/* Function: esl_dst_CJukesCantor()
* Synopsis: Jukes-Cantor distance for two aligned strings.
* Incept: SRE, Tue Apr 18 14:00:37 2006 [St. Louis]
*
* Purpose: Calculate the generalized Jukes-Cantor distance between
* two aligned character strings <as1> and <as2>, in
* substitutions/site, for an alphabet of <K> residues
* (<K=4> for nucleic acid, <K=20> for proteins). The
* maximum likelihood estimate for the distance is
* optionally returned in <opt_distance>. The large-sample
* variance for the distance estimate is
* optionally returned in <opt_variance>.
*
* Alphabetic symbols <[a-zA-Z]> are compared
* case-insensitively to count the number of identities
* (<n1>) and mismatches (<n2>>). Any nonalphabetic
* character is assumed to be a gap symbol, and aligned
* columns containing gap symbols are ignored. The
* fractional difference <D> used to calculate the
* Jukes/Cantor distance is <n2/n1+n2>.
*
* Args: K - size of the alphabet (4 or 20)
* as1 - 1st aligned seq, 0..L-1, \0-terminated
* as2 - 2nd aligned seq, 0..L-1, \0-terminated
* opt_distance - optRETURN: ML estimate of distance d
* opt_variance - optRETURN: large-sample variance of d
*
* Returns: <eslOK> on success.
*
* Infinite distances are possible, in which case distance
* and variance are both <HUGE_VAL>. Caller has to deal
* with this case as it sees fit, perhaps by enforcing
* an arbitrary maximum distance.
*
* Throws: <eslEINVAL> if the two strings aren't the same length (and
* thus can't have been properly aligned).
* <eslEDIVZERO> if no aligned residues were counted.
* On either failure, distance and variance are both returned
* as <HUGE_VAL>.
*/
int
esl_dst_CJukesCantor(int K, const char *as1, const char *as2,
double *opt_distance, double *opt_variance)
{
int status;
int n1, n2; /* number of observed identities, substitutions */
int i; /* position in aligned seqs */
/* 1. Count identities, mismatches.
*/
n1 = n2 = 0;
for (i = 0; as1[i] != '\0' && as2[i] != '\0'; i++)
{
if (isalpha(as1[i]) && isalpha(as2[i]))
{
if (toupper(as1[i]) == toupper(as2[i])) n1++; else n2++;
}
}
if (as1[i] != '\0' || as2[i] != '\0')
ESL_XEXCEPTION(eslEINVAL, "strings not same length, not aligned");
return jukescantor(n1, n2, K, opt_distance, opt_variance); /* can throw eslEDIVZERO */
ERROR:
if (opt_distance != NULL) *opt_distance = HUGE_VAL;
if (opt_variance != NULL) *opt_variance = HUGE_VAL;
return status;
}
/* Function: p7_Handmodelmaker()
*
* Purpose: Manual model construction.
* Construct an HMM from a digital alignment, where the
* <#=RF> line of the alignment file is used to indicate the
* columns assigned to matches vs. inserts.
*
* The <msa> must be in digital mode, and it must have
* a reference annotation line.
*
* NOTE: <p7_Handmodelmaker()> will slightly revise the
* alignment if necessary, if the assignment of columns
* implies DI and ID transitions.
*
* Returns both the HMM in counts form (ready for applying
* Dirichlet priors as the next step), and fake tracebacks
* for each aligned sequence.
*
* Models must have at least one node, so if the <msa> defined
* no consensus columns, a <eslENORESULT> error is returned.
*
* Args: msa - multiple sequence alignment
* bld - holds information on regions requiring masking, optionally NULL -> no masking
* ret_hmm - RETURN: counts-form HMM
* opt_tr - optRETURN: array of tracebacks for aseq's
*
* Return: <eslOK> on success. <ret_hmm> and <opt_tr> are allocated
* here, and must be free'd by caller.
*
* Returns <eslENORESULT> if no consensus columns were annotated;
* in this case, <ret_hmm> and <opt_tr> are returned NULL.
*
* Returns <eslEFORMAT> if the <msa> doesn't have a reference
* annotation line.
*
* Throws: <eslEMEM> on allocation failure. Throws <eslEINVAL> if the <msa>
* isn't in digital mode.
*/
int
p7_Handmodelmaker(ESL_MSA *msa, P7_BUILDER *bld, P7_HMM **ret_hmm, P7_TRACE ***opt_tr)
{
int status;
int *matassign = NULL; /* MAT state assignments if 1; 1..alen */
int apos; /* counter for aligned columns */
if (! (msa->flags & eslMSA_DIGITAL)) ESL_XEXCEPTION(eslEINVAL, "need a digital msa");
if (msa->rf == NULL) return eslEFORMAT;
ESL_ALLOC(matassign, sizeof(int) * (msa->alen+1));
/* Watch for off-by-one. rf is [0..alen-1]; matassign is [1..alen] */
for (apos = 1; apos <= msa->alen; apos++)
matassign[apos] = (esl_abc_CIsGap(msa->abc, msa->rf[apos-1])? FALSE : TRUE);
/* matassign2hmm leaves ret_hmm, opt_tr in their proper state: */
if ((status = matassign2hmm(msa, matassign, ret_hmm, opt_tr)) != eslOK) goto ERROR;
free(matassign);
return eslOK;
ERROR:
if (matassign != NULL) free(matassign);
return status;
}
/* Function: esl_dst_XPairId()
* Synopsis: Pairwise identity of two aligned digital seqs.
* Incept: SRE, Tue Apr 18 09:24:05 2006 [St. Louis]
*
* Purpose: Digital version of <esl_dst_PairId()>: <adsq1> and
* <adsq2> are digitized aligned sequences, in alphabet
* <abc>. Otherwise, same as <esl_dst_PairId()>.
*
* Args: abc - digital alphabet in use
* ax1 - aligned digital seq 1
* ax2 - aligned digital seq 2
* opt_pid - optRETURN: pairwise identity, 0<=x<=1
* opt_nid - optRETURN: # of identities
* opt_n - optRETURN: denominator MIN(len1,len2)
*
* Returns: <eslOK> on success. <opt_distance>, <opt_nid>, <opt_n>
* contain the answers, for any of these that were passed
* non-<NULL> pointers.
*
* Throws: <eslEINVAL> if the strings are different lengths (not aligned).
*/
int
esl_dst_XPairId(const ESL_ALPHABET *abc, const ESL_DSQ *ax1, const ESL_DSQ *ax2,
double *opt_distance, int *opt_nid, int *opt_n)
{
int status;
int idents; /* total identical positions */
int len1, len2; /* lengths of seqs */
int i; /* position in aligned seqs */
idents = len1 = len2 = 0;
for (i = 1; ax1[i] != eslDSQ_SENTINEL && ax2[i] != eslDSQ_SENTINEL; i++)
{
if (esl_abc_XIsCanonical(abc, ax1[i])) len1++;
if (esl_abc_XIsCanonical(abc, ax2[i])) len2++;
if (esl_abc_XIsCanonical(abc, ax1[i]) && esl_abc_XIsCanonical(abc, ax2[i])
&& ax1[i] == ax2[i])
idents++;
}
if (len2 < len1) len1 = len2;
if (ax1[i] != eslDSQ_SENTINEL || ax2[i] != eslDSQ_SENTINEL)
ESL_XEXCEPTION(eslEINVAL, "strings not same length, not aligned");
if (opt_distance != NULL) *opt_distance = ( len1==0 ? 0. : (double) idents / (double) len1 );
if (opt_nid != NULL) *opt_nid = idents;
if (opt_n != NULL) *opt_n = len1;
return eslOK;
ERROR:
if (opt_distance != NULL) *opt_distance = 0.;
if (opt_nid != NULL) *opt_nid = 0;
if (opt_n != NULL) *opt_n = 0;
return status;
}
/* Function: esl_dst_XJukesCantor()
* Synopsis: Jukes-Cantor distance for two aligned digitized seqs.
* Incept: SRE, Tue Apr 18 15:26:51 2006 [St. Louis]
*
* Purpose: Calculate the generalized Jukes-Cantor distance between two
* aligned digital strings <ax> and <ay>, in substitutions/site,
* using alphabet <abc> to evaluate identities and differences.
* The maximum likelihood estimate for the distance is optionally returned in
* <opt_distance>. The large-sample variance for the distance
* estimate is optionally returned in <opt_variance>.
*
* Identical to <esl_dst_CJukesCantor()>, except that it takes
* digital sequences instead of character strings.
*
* Args: abc - bioalphabet to use for comparisons
* ax - 1st digital aligned seq
* ay - 2nd digital aligned seq
* opt_distance - optRETURN: ML estimate of distance d
* opt_variance - optRETURN: large-sample variance of d
*
* Returns: <eslOK> on success. As in <esl_dst_CJukesCantor()>, the
* distance and variance may be infinite, in which case they
* are returned as <HUGE_VAL>.
*
* Throws: <eslEINVAL> if the two strings aren't the same length (and
* thus can't have been properly aligned).
* <eslEDIVZERO> if no aligned residues were counted.
* On either failure, the distance and variance are set
* to <HUGE_VAL>.
*/
int
esl_dst_XJukesCantor(const ESL_ALPHABET *abc, const ESL_DSQ *ax, const ESL_DSQ *ay,
double *opt_distance, double *opt_variance)
{
int status;
int n1, n2; /* number of observed identities, substitutions */
int i; /* position in aligned seqs */
n1 = n2 = 0;
for (i = 1; ax[i] != eslDSQ_SENTINEL && ay[i] != eslDSQ_SENTINEL; i++)
{
if (esl_abc_XIsCanonical(abc, ax[i]) && esl_abc_XIsCanonical(abc, ay[i]))
{
if (ax[i] == ay[i]) n1++;
else n2++;
}
}
if (ax[i] != eslDSQ_SENTINEL || ay[i] != eslDSQ_SENTINEL)
ESL_XEXCEPTION(eslEINVAL, "strings not same length, not aligned");
return jukescantor(n1, n2, abc->K, opt_distance, opt_variance);
ERROR:
if (opt_distance != NULL) *opt_distance = HUGE_VAL;
if (opt_variance != NULL) *opt_variance = HUGE_VAL;
return status;
}
/* Function: esl_dst_CPairId()
* Synopsis: Pairwise identity of two aligned text strings.
* Incept: SRE, Mon Apr 17 20:06:07 2006 [St. Louis]
*
* Purpose: Calculates pairwise fractional identity between two
* aligned character strings <asq1> and <asq2>.
* Return this distance in <opt_pid>; return the
* number of identities counted in <opt_nid>; and
* return the denominator <MIN(len1,len2)> in
* <opt_n>.
*
* Alphabetic symbols <[a-zA-Z]> are compared
* case-insensitively for identity. Any nonalphabetic
* character is assumed to be a gap symbol.
*
* This simple comparison rule is unaware of synonyms and
* degeneracies in biological alphabets. For a more
* sophisticated and biosequence-aware comparison, use
* digitized sequences and the <esl_dst_XPairId()> function
* instead.
*
* Args: asq1 - aligned character string 1
* asq2 - aligned character string 2
* opt_pid - optRETURN: pairwise identity, 0<=x<=1
* opt_nid - optRETURN: # of identities
* opt_n - optRETURN: denominator MIN(len1,len2)
*
* Returns: <eslOK> on success. <opt_pid>, <opt_nid>, <opt_n>
* contain the answers (for whichever were passed non-NULL).
*
* Throws: <eslEINVAL> if the strings are different lengths
* (not aligned).
*/
int
esl_dst_CPairId(const char *asq1, const char *asq2,
double *opt_pid, int *opt_nid, int *opt_n)
{
int status;
int idents; /* total identical positions */
int len1, len2; /* lengths of seqs */
int i; /* position in aligned seqs */
idents = len1 = len2 = 0;
for (i = 0; asq1[i] != '\0' && asq2[i] != '\0'; i++)
{
if (isalpha(asq1[i])) len1++;
if (isalpha(asq2[i])) len2++;
if (isalpha(asq1[i]) && isalpha(asq2[i])
&& toupper(asq1[i]) == toupper(asq2[i]))
idents++;
}
if (asq1[i] != '\0' || asq2[i] != '\0')
ESL_XEXCEPTION(eslEINVAL, "strings not same length, not aligned");
if (opt_pid != NULL) *opt_pid = ( len1==0 ? 0. : (double) idents / (double) ESL_MIN(len1,len2));
if (opt_nid != NULL) *opt_nid = idents;
if (opt_n != NULL) *opt_n = len1;
return eslOK;
ERROR:
if (opt_pid != NULL) *opt_pid = 0.;
if (opt_nid != NULL) *opt_nid = 0;
if (opt_n != NULL) *opt_n = 0;
return status;
}
/* Function: esl_dst_XPairIdMx()
* Synopsis: NxN identity matrix for N aligned digital seqs.
* Incept: SRE, Thu Apr 27 09:08:11 2006 [New York]
*
* Purpose: Given a digitized multiple sequence alignment <ax>, consisting
* of <N> aligned digital sequences in alphabet <abc>; calculate
* a symmetric pairwise fractional identity matrix by $N(N-1)/2$
* calls to <esl_dst_XPairId()>, and return it in <ret_S>.
*
* Args: abc - digital alphabet in use
* ax - aligned dsq's, [0..N-1][1..alen]
* N - number of aligned sequences
* ret_S - RETURN: NxN matrix of fractional identities
*
* Returns: <eslOK> on success, and <ret_S> contains the distance
* matrix. Caller is obligated to free <S> with
* <esl_dmatrix_Destroy()>.
*
* Throws: <eslEINVAL> if a seq has a different
* length than others. On failure, <ret_S> is returned <NULL>
* and state of inputs is unchanged.
*/
int
esl_dst_XPairIdMx(const ESL_ALPHABET *abc, ESL_DSQ **ax, int N, ESL_DMATRIX **ret_S)
{
int status;
ESL_DMATRIX *S = NULL;
int i,j;
if (( S = esl_dmatrix_Create(N,N) ) == NULL) goto ERROR;
for (i = 0; i < N; i++)
{
S->mx[i][i] = 1.;
for (j = i+1; j < N; j++)
{
status = esl_dst_XPairId(abc, ax[i], ax[j], &(S->mx[i][j]), NULL, NULL);
if (status != eslOK)
ESL_XEXCEPTION(status, "Pairwise identity calculation failed at seqs %d,%d\n", i,j);
S->mx[j][i] = S->mx[i][j];
}
}
if (ret_S != NULL) *ret_S = S; else esl_dmatrix_Destroy(S);
return eslOK;
ERROR:
if (S != NULL) esl_dmatrix_Destroy(S);
if (ret_S != NULL) *ret_S = NULL;
return status;
}
/* Function: p7_oprofile_MPIRecv()
* Synopsis: Receives an OPROFILE as a work unit from an MPI sender.
* Incept: MSF, Wed Oct 21, 2009 [Janelia]
*
* Purpose: Receive a work unit that consists of a single OPROFILE
* sent by MPI <source> (<0..nproc-1>, or
* <MPI_ANY_SOURCE>) tagged as <tag> for MPI communicator <comm>.
*
* Work units are prefixed by a status code. If the unit's
* code is <eslOK> and no errors are encountered, this
* routine will return <eslOK> and a non-<NULL> <*ret_om>.
* If the unit's code is <eslEOD> (a shutdown signal),
* this routine returns <eslEOD> and <*ret_om> is <NULL>.
*
* Caller provides a working buffer <*buf> of size
* <*nalloc> characters. These are passed by reference, so
* that <*buf> can be reallocated and <*nalloc> increased
* if necessary. As a special case, if <*buf> is <NULL> and
* <*nalloc> is 0, the buffer will be allocated
* appropriately, but the caller is still responsible for
* free'ing it.
*
* Caller may or may not already know what alphabet the OPROFILE
* is expected to be in. A reference to the current
* alphabet is passed in <abc>. If the alphabet is unknown,
* pass <*abc = NULL>, and when the OPROFILE is received, an
* appropriate new alphabet object is allocated and passed
* back to the caller via <*abc>. If the alphabet is
* already known, <*ret_abc> is that alphabet, and the new
* OPROFILE's alphabet type is verified to agree with it. This
* mechanism allows an application to let the first OPROFILE
* determine the alphabet type for the application, while
* still keeping the alphabet under the application's scope
* of control.
*
* Returns: <eslOK> on success. <*ret_om> contains the received OPROFILE;
* it is allocated here, and the caller is responsible for
* free'ing it. <*buf> may have been reallocated to a
* larger size, and <*nalloc> may have been increased. If
* <*abc> was passed as <NULL>, it now points to an
* <ESL_ALPHABET> object that was allocated here; caller is
* responsible for free'ing this.
*
* Returns <eslEOD> if an end-of-data signal was received.
* In this case, <*buf>, <*nalloc>, and <*abc> are left unchanged,
* and <*ret_om> is <NULL>.
*
* Returns <eslEINCOMPAT> if the OPROFILE is in a different alphabet
* than <*abc> said to expect. In this case, <*abc> is unchanged,
* <*buf> and <*nalloc> may have been changed, and <*ret_om> is
* <NULL>.
*
* Throws: <eslEMEM> on allocation error, in which case <*ret_om> is
* <NULL>.
*/
int
p7_oprofile_MPIRecv(int source, int tag, MPI_Comm comm, char **buf, int *nalloc, ESL_ALPHABET **abc, P7_OPROFILE **ret_om)
{
int status;
int code;
P7_OPROFILE *om = NULL;
int n;
int pos;
MPI_Status mpistatus;
/* Probe first, because we need to know if our buffer is big enough. */
MPI_Probe(source, tag, comm, &mpistatus);
MPI_Get_count(&mpistatus, MPI_PACKED, &n);
/* Make sure the buffer is allocated appropriately */
if (*buf == NULL || n > *nalloc) {
void *tmp;
ESL_RALLOC(*buf, tmp, sizeof(char) * n);
*nalloc = n;
}
/* Receive the packed work unit */
MPI_Recv(*buf, n, MPI_PACKED, source, tag, comm, &mpistatus);
/* Unpack it, looking at the status code prefix for EOD/EOK */
pos = 0;
if (MPI_Unpack(*buf, n, &pos, &code, 1, MPI_INT, comm) != 0) ESL_XEXCEPTION(eslESYS, "mpi unpack failed");
if (code == eslEOD) { *ret_om = NULL; return eslEOD; }
return p7_oprofile_MPIUnpack(*buf, *nalloc, &pos, comm, abc, ret_om);
ERROR:
if (om != NULL) p7_oprofile_Destroy(om);
return status;
}
/* Function: esl_workqueue_Create()
* Synopsis: Create a work queue object.
* Incept: MSF, Thu Jun 18 11:51:39 2009
*
* Purpose: Creates an <ESL_WORK_QUEUE> object of <size>. The
* queues are used to handle objects <void *> that
* are ready to be processed and that have been
* processed by worker threads.
*
* Returns: ptr to the new <ESL_WORK_QUEUE> object.
*
* Throws: <eslESYS> on allocation or initialization failure.
*/
ESL_WORK_QUEUE *
esl_workqueue_Create(int size)
{
int i;
int status;
ESL_WORK_QUEUE *queue = NULL;
ESL_ALLOC(queue, sizeof(ESL_WORK_QUEUE));
queue->readerQueue = NULL;
queue->readerQueueCnt = 0;
queue->readerQueueHead = 0;
queue->workerQueue = NULL;
queue->workerQueueCnt = 0;
queue->workerQueueHead = 0;
queue->queueSize = size;
queue->pendingWorkers = 0;
if (pthread_mutex_init(&queue->queueMutex, NULL) != 0) ESL_XEXCEPTION(eslESYS, "mutex init failed");
if (pthread_cond_init(&queue->readerQueueCond, NULL) != 0) ESL_XEXCEPTION(eslESYS, "cond reader init failed");
if (pthread_cond_init(&queue->workerQueueCond, NULL) != 0) ESL_XEXCEPTION(eslESYS, "cond worker init failed");
ESL_ALLOC(queue->readerQueue, sizeof(void *) * size);
ESL_ALLOC(queue->workerQueue, sizeof(void *) * size);
for (i = 0; i < queue->queueSize; ++i)
{
queue->readerQueue[i] = NULL;
queue->workerQueue[i] = NULL;
}
return queue;
ERROR:
esl_workqueue_Destroy(queue);
return NULL;
}
/* Function: esl_workqueue_WorkerUpdate()
* Synopsis: Consumer routine.
* Incept: MSF, Thu Jun 18 11:51:39 2009
*
* Purpose: The consumer (i.e. Worker) places an object that has
* been processed on the producers (i.e. Readers) queue.
*
* If the <in> object is not null, it is placed on the
* readers queue. If the reader is waiting for an object,
* it is signaled it to wake up.
*
* If the worker routine has supplied an <out> pointer,
* an object that is ready for processing by a worker,
* is placed in <out> so the worker thread can continue.
*
* Returns: <eslOK> on success.
*
* Throws: <eslESYS> if thread synchronization fails somewhere.
* <eslEINVAL> if something's wrong with <queue>.
*/
int esl_workqueue_WorkerUpdate(ESL_WORK_QUEUE *queue, void *in, void **out)
{
int cnt;
int inx;
int queueSize;
int status;
if (queue == NULL) ESL_XEXCEPTION(eslEINVAL, "Invalid queue object");
if (pthread_mutex_lock (&queue->queueMutex) != 0) ESL_XEXCEPTION(eslESYS, "mutex lock failed");
queueSize = queue->queueSize;
/* check if the caller is queuing up an item */
if (in != NULL)
{
/* check to make sure we don't overflow */
if (queue->readerQueueCnt >= queueSize) ESL_XEXCEPTION(eslEINVAL, "Reader queue overflow");
inx = (queue->readerQueueHead + queue->readerQueueCnt) % queueSize;
queue->readerQueue[inx] = in;
cnt = queue->readerQueueCnt++;
if (cnt == 0)
{
if (pthread_cond_signal (&queue->readerQueueCond) != 0) ESL_XEXCEPTION(eslESYS, "cond signal failed");
}
}
/* check if the caller is waiting for a queued item */
if (out != NULL)
{
if (queue->workerQueueCnt == 0)
{
/* wait for a processed buffers to be returned */
++queue->pendingWorkers;
while (queue->workerQueueCnt == 0)
{
if (pthread_cond_wait (&queue->workerQueueCond, &queue->queueMutex) != 0) ESL_XEXCEPTION(eslESYS, "cond wait failed");
}
--queue->pendingWorkers;
}
inx = queue->workerQueueHead;
*out = queue->workerQueue[inx];
queue->workerQueue[inx] = NULL;
queue->workerQueueHead = (queue->workerQueueHead + 1) % queueSize;
--queue->workerQueueCnt;
}
if (pthread_mutex_unlock (&queue->queueMutex) != 0) ESL_XEXCEPTION(eslESYS, "mutex unlock failed");
return eslOK;
ERROR:
if (out) *out = NULL;
return status;
}
/* Function: p7_SingleBuilder()
* Synopsis: Build a new HMM from a single sequence.
*
* Purpose: Take the sequence <sq> and a build configuration <bld>, and
* build a new HMM.
*
* The single sequence scoring system in the <bld>
* configuration must have been previously initialized by
* <p7_builder_SetScoreSystem()>.
*
* Args: bld - build configuration
* sq - query sequence
* bg - null model (needed to paramaterize insert emission probs)
* opt_hmm - optRETURN: new HMM
* opt_gm - optRETURN: profile corresponding to <hmm>
* opt_om - optRETURN: optimized profile corresponding to <gm>
*
* Returns: <eslOK> on success.
*
* Throws: <eslEMEM> on allocation error.
* <eslEINVAL> if <bld> isn't properly configured somehow.
*/
int
p7_SingleBuilder(P7_BUILDER *bld, ESL_SQ *sq, P7_BG *bg, P7_HMM **opt_hmm,
P7_TRACE **opt_tr, P7_PROFILE **opt_gm, P7_OPROFILE **opt_om)
{
P7_HMM *hmm = NULL;
P7_TRACE *tr = NULL;
int k;
int status;
bld->errbuf[0] = '\0';
if (! bld->Q) ESL_XEXCEPTION(eslEINVAL, "score system not initialized");
if ((status = p7_Seqmodel(bld->abc, sq->dsq, sq->n, sq->name, bld->Q, bg->f, bld->popen, bld->pextend, &hmm)) != eslOK) goto ERROR;
if ((status = p7_hmm_SetComposition(hmm)) != eslOK) goto ERROR;
if ((status = p7_hmm_SetConsensus(hmm, sq)) != eslOK) goto ERROR;
if ((status = calibrate(bld, hmm, bg, opt_gm, opt_om)) != eslOK) goto ERROR;
if ( bld->abc->type == eslDNA || bld->abc->type == eslRNA ) {
if (bld->w_len > 0) hmm->max_length = bld->w_len;
else if (bld->w_beta == 0.0) hmm->max_length = hmm->M *4;
else if ( (status = p7_Builder_MaxLength(hmm, bld->w_beta)) != eslOK) goto ERROR;
}
/* build a faux trace: relative to core model (B->M_1..M_L->E) */
if (opt_tr != NULL)
{
if ((tr = p7_trace_Create()) == NULL) goto ERROR;
if ((status = p7_trace_Append(tr, p7T_B, 0, 0)) != eslOK) goto ERROR;
for (k = 1; k <= sq->n; k++)
if ((status = p7_trace_Append(tr, p7T_M, k, k)) != eslOK) goto ERROR;
if ((status = p7_trace_Append(tr, p7T_E, 0, 0)) != eslOK) goto ERROR;
tr->M = sq->n;
tr->L = sq->n;
}
/* note that <opt_gm> and <opt_om> were already set by calibrate() call above. */
if (opt_hmm != NULL) *opt_hmm = hmm; else p7_hmm_Destroy(hmm);
if (opt_tr != NULL) *opt_tr = tr;
return eslOK;
ERROR:
p7_hmm_Destroy(hmm);
if (tr != NULL) p7_trace_Destroy(tr);
if (opt_gm != NULL) p7_profile_Destroy(*opt_gm);
if (opt_om != NULL) p7_oprofile_Destroy(*opt_om);
return status;
}
/* Function: p7_Fastmodelmaker()
*
* Purpose: Heuristic model construction.
* Construct an HMM from an alignment by a simple rule,
* based on the fractional occupancy of each columns w/
* residues vs gaps. Any column w/ a fractional
* occupancy of $\geq$ <symfrac> is assigned as a MATCH column;
* for instance, if thresh = 0.5, columns w/ $\geq$ 50\%
* residues are assigned to match... roughly speaking.
*
* "Roughly speaking" because sequences may be weighted
* in the input <msa>, and because missing data symbols are
* ignored, in order to deal with sequence fragments.
*
* The <msa> must be in digital mode.
*
* If the caller wants to designate any sequences as
* fragments, it does so by converting all N-terminal and
* C-terminal flanking gap symbols to missing data symbols.
*
* NOTE: p7_Fastmodelmaker() will slightly revise the
* alignment if the assignment of columns implies
* DI and ID transitions.
*
* Returns the HMM in counts form (ready for applying Dirichlet
* priors as the next step). Also returns fake traceback
* for each training sequence.
*
* Models must have at least one node, so if the <msa> defined
* no consensus columns, a <eslENORESULT> error is returned.
*
* Args: msa - multiple sequence alignment
* symfrac - threshold for residue occupancy; >= assigns MATCH
* bld - holds information on regions requiring masking, optionally NULL -> no masking
* ret_hmm - RETURN: counts-form HMM
* opt_tr - optRETURN: array of tracebacks for aseq's
*
* Return: <eslOK> on success. ret_hmm and opt_tr allocated here,
* and must be free'd by the caller (FreeTrace(tr[i]), free(tr),
* FreeHMM(hmm)).
*
* Returns <eslENORESULT> if no consensus columns were annotated;
* in this case, <ret_hmm> and <opt_tr> are returned NULL.
*
* Throws: <eslEMEM> on allocation failure; <eslEINVAL> if the
* <msa> isn't in digital mode.
*/
int
p7_Fastmodelmaker(ESL_MSA *msa, float symfrac, P7_BUILDER *bld, P7_HMM **ret_hmm, P7_TRACE ***opt_tr)
{
int status; /* return status flag */
int *matassign = NULL; /* MAT state assignments if 1; 1..alen */
int idx; /* counter over sequences */
int apos; /* counter for aligned columns */
float r; /* weighted residue count */
float totwgt; /* weighted residue+gap count */
if (! (msa->flags & eslMSA_DIGITAL)) ESL_XEXCEPTION(eslEINVAL, "need digital MSA");
/* Allocations: matassign is 1..alen array of bit flags.
*/
ESL_ALLOC(matassign, sizeof(int) * (msa->alen+1));
/* Determine weighted sym freq in each column, set matassign[] accordingly.
*/
for (apos = 1; apos <= msa->alen; apos++)
{
r = totwgt = 0.;
for (idx = 0; idx < msa->nseq; idx++)
{
if (esl_abc_XIsResidue(msa->abc, msa->ax[idx][apos])) { r += msa->wgt[idx]; totwgt += msa->wgt[idx]; }
else if (esl_abc_XIsGap(msa->abc, msa->ax[idx][apos])) { totwgt += msa->wgt[idx]; }
else if (esl_abc_XIsMissing(msa->abc, msa->ax[idx][apos])) continue;
}
if (r > 0. && r / totwgt >= symfrac) matassign[apos] = TRUE;
else matassign[apos] = FALSE;
}
/* Once we have matassign calculated, modelmakers behave
* the same; matassign2hmm() does this stuff (traceback construction,
* trace counting) and sets up ret_hmm and opt_tr.
*/
if ((status = matassign2hmm(msa, matassign, ret_hmm, opt_tr)) != eslOK) {
fprintf (stderr, "hmm construction error during trace counting\n");
goto ERROR;
}
free(matassign);
return eslOK;
ERROR:
if (matassign != NULL) free(matassign);
return status;
}
/* Function: esl_randomness_Create()
* Synopsis: Create an RNG with a given seed.
* Incept: SRE, Wed Jul 14 13:02:18 2004 [St. Louis]
*
* Purpose: Create a random number generator using
* a given random seed. Seed must be $>0$.
*
* Args: seed $>= 0$.
*
* Returns: an initialized <ESL_RANDOMNESS *> on success.
* Caller free's with <esl_randomness_Destroy()>.
*
* Throws: <NULL> on failure.
*
* Xref: STL8/p57.
*/
ESL_RANDOMNESS *
esl_randomness_Create(long seed)
{
ESL_RANDOMNESS *r = NULL;
int burnin = 7;
int status;
if (seed <= 0) ESL_XEXCEPTION(eslEINVAL, "bad seed");
ESL_ALLOC(r, sizeof(ESL_RANDOMNESS));
r->seed = seed;
r->reseeding = TRUE;
/* we observe that the first random number isn't very random, if
* closely spaced seeds are used, like what we get with using
* time(). So, "burn in" the random chain just a little.
*/
while (burnin--) esl_random(r);
return r;
ERROR:
return NULL;
}
/* Function: esl_dst_XJukesCantorMx()
* Synopsis: NxN Jukes/Cantor distance matrix for N aligned digital seqs.
* Incept: SRE, Thu Apr 27 08:38:08 2006 [New York City]
*
* Purpose: Given a digitized multiple sequence alignment <ax>,
* consisting of <nseq> aligned digital sequences in
* bioalphabet <abc>, calculate a symmetric Jukes/Cantor
* pairwise distance matrix for all sequence pairs;
* optionally return the distance matrix in <ret_D> and
* a matrix of the large-sample variances for those ML distance
* estimates in <ret_V>.
*
* Infinite distances (and variances) are possible. They
* are represented as <HUGE_VAL> in <D> and <V>. Caller must
* be prepared to deal with them as appropriate.
*
* Args: abc - bioalphabet for <aseq>
* ax - aligned digital sequences [0.nseq-1][1..L]
* nseq - number of aseqs
* opt_D - optRETURN: [0..nseq-1]x[0..nseq-1] symmetric distance mx
* opt_V - optRETURN: matrix of variances.
*
* Returns: <eslOK> on success. <D> (and optionally <V>) contain the
* distance matrix (and variances). Caller frees these with
* <esl_dmatrix_Destroy()>.
*
* Throws: <eslEINVAL> if any pair of sequences have differing lengths
* (and thus cannot have been properly aligned).
* <eslEDIVZERO> if some pair of sequences had no aligned
* residues. On failure, <D> and <V> are both returned <NULL>
* and state of inputs is unchanged.
*/
int
esl_dst_XJukesCantorMx(const ESL_ALPHABET *abc, ESL_DSQ **ax, int nseq,
ESL_DMATRIX **opt_D, ESL_DMATRIX **opt_V)
{
ESL_DMATRIX *D = NULL;
ESL_DMATRIX *V = NULL;
int status;
int i,j;
if (( D = esl_dmatrix_Create(nseq, nseq) ) == NULL) goto ERROR;
if (( V = esl_dmatrix_Create(nseq, nseq) ) == NULL) goto ERROR;
for (i = 0; i < nseq; i++)
{
D->mx[i][i] = 0.;
V->mx[i][i] = 0.;
for (j = i+1; j < nseq; j++)
{
status = esl_dst_XJukesCantor(abc, ax[i], ax[j],
&(D->mx[i][j]), &(V->mx[i][j]));
if (status != eslOK)
ESL_XEXCEPTION(status, "J/C calculation failed at digital aseqs %d,%d", i,j);
D->mx[j][i] = D->mx[i][j];
V->mx[j][i] = V->mx[i][j];
}
}
if (opt_D != NULL) *opt_D = D; else esl_dmatrix_Destroy(D);
if (opt_V != NULL) *opt_V = V; else esl_dmatrix_Destroy(V);
return eslOK;
ERROR:
if (D != NULL) esl_dmatrix_Destroy(D);
if (V != NULL) esl_dmatrix_Destroy(V);
if (opt_D != NULL) *opt_D = NULL;
if (opt_V != NULL) *opt_V = NULL;
return status;
}
/* Function: p7_oprofile_MPISend()
* Synopsis: Send an OPROFILE as an MPI work unit.
* Incept: MSF, Wed Oct 21, 2009 [Janelia]
*
* Purpose: Sends an OPROFILE <om> as a work unit to MPI process
* <dest> (where <dest> ranges from 0..<nproc-1>), tagged
* with MPI tag <tag>, for MPI communicator <comm>, as
* the sole workunit or result.
*
* Work units are prefixed by a status code. If <hmm> is
* <non-NULL>, the work unit is an <eslOK> code followed by
* the packed HMM. If <hmm> is NULL, the work unit is an
* <eslEOD> code, which <p7_hmm_MPIRecv()> knows how to
* interpret; this is typically used for an end-of-data
* signal to cleanly shut down worker processes.
*
* In order to minimize alloc/free cycles in this routine,
* caller passes a pointer to a working buffer <*buf> of
* size <*nalloc> characters. If necessary (i.e. if <hmm> is
* too big to fit), <*buf> will be reallocated and <*nalloc>
* increased to the new size. As a special case, if <*buf>
* is <NULL> and <*nalloc> is 0, the buffer will be
* allocated appropriately, but the caller is still
* responsible for free'ing it.
*
* Returns: <eslOK> on success; <*buf> may have been reallocated and
* <*nalloc> may have been increased.
*
* Throws: <eslESYS> if an MPI call fails; <eslEMEM> if a malloc/realloc
* fails. In either case, <*buf> and <*nalloc> remain valid and useful
* memory (though the contents of <*buf> are undefined).
*
* Note: Compare to p7_hmmfile_WriteBinary(). The two operations (sending
* an HMM via MPI, or saving it as a binary file to disk) are
* similar.
*/
int
p7_oprofile_MPISend(P7_OPROFILE *om, int dest, int tag, MPI_Comm comm, char **buf, int *nalloc)
{
int status;
int code;
int sz, n, pos;
/* Figure out size */
if (MPI_Pack_size(1, MPI_INT, comm, &n) != 0) ESL_XEXCEPTION(eslESYS, "mpi pack size failed");
if (om != NULL) {
if ((status = p7_oprofile_MPIPackSize(om, comm, &sz)) != eslOK) return status;
n += sz;
}
/* Make sure the buffer is allocated appropriately */
if (*buf == NULL || n > *nalloc) {
void *tmp;
ESL_RALLOC(*buf, tmp, sizeof(char) * n);
*nalloc = n;
}
/* Pack the status code and OPROFILE into the buffer */
pos = 0;
code = (om == NULL) ? eslEOD : eslOK;
if (MPI_Pack(&code, 1, MPI_INT, *buf, n, &pos, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi pack failed");
if (om != NULL) {
if ((status = p7_oprofile_MPIPack(om, *buf, n, &pos, comm)) != eslOK) return status;
}
/* Send the packed OPROFILE to the destination. */
if (MPI_Send(*buf, n, MPI_PACKED, dest, tag, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi send failed");
return eslOK;
ERROR:
return status;
}
/* Function: esl_msashuffle_CQRNA()
* Synopsis: Gap-preserving column shuffle of a pairwise alignment.
* Incept: SRE, Tue Jan 22 08:45:34 2008 [Market Street Cafe, Leesburg]
*
* Purpose: Shuffle a pairwise alignment <x>,<y> while preserving the
* position of gaps, using the random number generator <r>.
* Return the shuffled alignment in <xs>,
* <ys>. Caller provides allocated space for <xs> and <ys>.
*
* An alphabet <abc> must also be provided, solely for the
* definition of gap characters. Because Easel's default
* alphabets (DNA, RNA, and protein) all use the same
* definition of gap characters <-_.>, you can actually
* provide any alphabet here, and get the same results.
* (This may save having to determine the alphabet of input
* sequences.)
*
* Works by doing three separate
* shuffles, of (1) columns with residues in both
* <x> and <y>, (2) columns with residue in <x> and gap in <y>,
* and (3) columns with gap in <x> and residue in <y>.
*
* <xs>,<x> and <ys>,<y> may be identical: that is, to shuffle
* an alignment "in place", destroying the original
* alignment, just call <esl_msashuffle_CQRNA(r, abc, x,y,x,y)>.
*
* Returns: <eslOK> on success, and the shuffled alignment is
* returned in <xs>, <ys>.
*
* Throws: <eslEMEM> on allocation failure.
*/
int
esl_msashuffle_CQRNA(ESL_RANDOMNESS *r, ESL_ALPHABET *abc, char *x, char *y, char *xs, char *ys)
{
int L;
int *xycol = NULL;
int *xcol = NULL;
int *ycol = NULL;
int nxy, nx, ny;
int i;
int pos, c;
char xsym, ysym;
int status;
if (xs != x) strcpy(xs, x);
if (ys != y) strcpy(ys, y);
/* First, construct three arrays containing lists of the column positions
* of the three types of columns. (If a column contains gaps in both x and y,
* we've already simply copied it to the shuffled sequence.)
*/
L = strlen(x);
if (strlen(y) != L) ESL_XEXCEPTION(eslEINVAL, "sequences of different lengths in qrna shuffle");
ESL_ALLOC(xycol, sizeof(int) * L);
ESL_ALLOC(xcol, sizeof(int) * L);
ESL_ALLOC(ycol, sizeof(int) * L);
nxy = nx = ny = 0;
for (i = 0; i < L; i++)
{
if ( esl_abc_CIsGap(abc, x[i]) && esl_abc_CIsGap(abc, y[i])) { continue; }
else if (! esl_abc_CIsGap(abc, x[i]) && ! esl_abc_CIsGap(abc, y[i])) { xycol[nxy] = i; nxy++; }
else if ( esl_abc_CIsGap(abc, x[i])) { ycol[ny] = i; ny++; }
else if ( esl_abc_CIsGap(abc, y[i])) { xcol[nx] = i; nx++; }
}
/* Second, shuffle the sequences indirectly, via shuffling these arrays.
* Yow, careful with those indices, and with order of the statements...
*/
for (; nxy > 1; nxy--) {
pos = esl_rnd_Roll(r, nxy);
xsym = xs[xycol[pos]]; ysym = ys[xycol[pos]]; c = xycol[pos];
xs[xycol[pos]] = xs[xycol[nxy-1]]; ys[xycol[pos]] = ys[xycol[nxy-1]]; xycol[pos] = xycol[nxy-1];
xs[xycol[nxy-1]] = xsym; ys[xycol[nxy-1]] = ysym; xycol[pos] = xycol[nxy-1];
}
for (; nx > 1; nx--) {
pos = esl_rnd_Roll(r, nx);
xsym = xs[xcol[pos]]; ysym = ys[xcol[pos]]; c = xcol[pos];
xs[xcol[pos]] = xs[xcol[nx-1]]; ys[xcol[pos]] = ys[xcol[nx-1]]; xcol[pos] = xcol[nx-1];
xs[xcol[nx-1]] = xsym; ys[xcol[nx-1]] = ysym; xcol[nx-1] = c;
}
for (; ny > 1; ny--) {
pos = esl_rnd_Roll(r, ny);
xsym = xs[ycol[pos]]; ysym = ys[ycol[pos]]; c = ycol[pos];
xs[ycol[pos]] = xs[ycol[ny-1]]; ys[ycol[pos]] = ys[ycol[ny-1]]; ycol[pos] = ycol[ny-1];
xs[ycol[ny-1]] = xsym; ys[ycol[ny-1]] = ysym; ycol[ny-1] = c;
}
free(xycol); free(xcol); free(ycol);
return eslOK;
ERROR:
if (xycol != NULL) free(xycol);
if (xcol != NULL) free(xcol);
if (ycol != NULL) free(ycol);
return status;
}
/* ideal_local_endpoints()
*
* Purpose: Implementation of the "two-step" fragment sampling
* algorithm, sampling a uniform local fragment w.r.t.
* sequence coords, by first sampling a complete
* sequence of length L from <hmm>; then choosing
* a random fragment <i1..i2> uniformly from all
* possible $\frac{L(L+1)/2}$ fragments; then finding
* local alignment coordinates wrt model and sequence,
* using convention that local alignment starts/stops
* with match states. (Thus, if the initially selected
* i1 or i2 were generated by insert states, bounds
* are moved to reach first/last match state.)
*
* The caller also provides an allocated sequence <sq> and
* traceback <tr>, as storage to be provided to
* <p7_CoreEmit()>. They contain the generated global
* sequence and trace upon return (not a local trace, note).
*
* i endpoints are normalized/discretized to 1..<Lbins>, so
* we can collate i statistics from sampled sequences of
* varying L. Note this causes discretization artifacts,
* leading to underrepresentation of j=M and
* overrepresentation of i=1.
*
* This routine is only intended for collecting endpoint
* statistics (i1,i2,k1,k2); it does not generate a local
* alignment trace. (xref milestone 2, STL11/115).
*
* Returns: <eslOK> on success; returns normalized/binned sequence
* coords in <*ret_i1> and <*ret_i2> in range <1..Lbins> and
* the model entry/exit coords in <*ret_k1> and <*ret_k2> in
* range <1..M>. By internal def'n of local alignment endpoints,
* M_k1 emits residue x_i1, M_k2 emits residue x_i2.
*
* Xref: STL11/142-143
*/
static int
ideal_local_endpoints(ESL_RANDOMNESS *r, P7_HMM *hmm, ESL_SQ *sq, P7_TRACE *tr, int Lbins,
int *ret_i1, int *ret_i2, int *ret_k1, int *ret_k2)
{
int status;
int tpos;
int i1, i2, k1,k2, t1,t2;
int all_insert;
int failsafe = 0; /* a failsafe timer for rejection sampling */
do {
if (failsafe++ == 1000) ESL_XEXCEPTION(eslENOHALT, "failed to obtain local alignment that wasn't all inserts");
if ((status = p7_CoreEmit(r, hmm, sq, tr)) != eslOK) goto ERROR;
/* a simple way to sample uniformly from upper triangle is by rejection
* this do/while cannot infinite loop, doesn't need failsafe
*/
do {
i1 = 1 + esl_rnd_Roll(r, sq->n);
i2 = 1 + esl_rnd_Roll(r, sq->n);
} while (i1 > i2);
/* Get initial k1,k2 coords: this step must work in a core model,
* i1/i2 were generated by an M or I. Also record t1,t2 endpoints
* on core's trace.
*/
for (tpos = 0; tpos < tr->N; tpos++)
if (tr->i[tpos] == i1) { t1 = tpos; k1 = tr->k[tpos]; break; }
for (tpos = tr->N-1; tpos >= 0; tpos--)
if (tr->i[tpos] == i2) { t2 = tpos; k2 = tr->k[tpos]; break; }
/* Enforce the definition of local alignment endpoints being
* match-delimited - roll up any leading/trailing I states.
* Watch out for pathological case of a local fragment that
* includes no M state at all.
*/
all_insert = FALSE;
for (; t1 <= t2; t1++) if (tr->st[t1] == p7T_M) break;
for (; t2 >= t1; t2--) if (tr->st[t2] == p7T_M) break;
if (t2 < t1) all_insert = TRUE; /* sufficient to check both. */
i1 = tr->i[t1]; i2 = tr->i[t2];
k1 = tr->k[t1]; k2 = tr->k[t2];
} while (all_insert);
/* Normalize sequence coords.
* They're 1..L now; make them 1..Lbins
*/
*ret_i1 = ((i1-1) * Lbins / sq->n) + 1;
*ret_i2 = ((i2-1) * Lbins / sq->n) + 1;
*ret_k1 = k1;
*ret_k2 = k2;
return eslOK;
ERROR:
*ret_i1 = 0.;
*ret_i2 = 0.;
*ret_k1 = 0;
*ret_k2 = 0;
return status;
}
/* profile_local_endpoints()
*
* Purpose: Wrapper around <p7_ProfileEmit()>, sampling a local
* alignment fragment from the profile's probabilistic model
* (which may be the implicit model of HMMER3, or the
* Plan7 model of HMMER2), and reporting coordinates
* of the fragment w.r.t. both model and sequence.
*
* To simplify the implementation, the profile must be in
* <p7_UNILOCAL> mode, not <p7_LOCAL> mode, so we know we
* only have to deal with a single hit per sampled
* sequence.
*
* We want <i1..i2> to be relative to the sequence coords
* of a complete (global) sampled sequence that we could
* have sampled this local alignment from; but the <i1..i2>
* we initially get are relative to our profile-sampled
* trace, so they are offset both by N-generated residues
* that occur in the profile and by residues that the
* profile's local entry skipped. To translate from
* profile/sequence coords to core model/sequence coords,
* we use rejection sampling: sample traces from the core
* model until we find one that uses the same statetypes
* at *initial* entry/exit points <k1>,<k2>, then use
* that sample's sequence to determine offsets and correct
* <i1..i2> reference frame.
*
* Local alignment endpoints are defined to be
* match-delimited. However, an H3 model allows exit on
* either a D or M state. Thus, the initially sampled end
* point k2 may need to be rolled back to last M state, to
* satisfy local alignment endpoint definition. Entries are
* not a problem; both H2 and H3 profiles can only enter on
* a M state. (This rollback has to occur after we've
* matched a core trace to the profile trace to determine
* i offsets.)
*
* Then, sampling from both the core model and the profile
* in the same routine introduces a complication:
* conceivably, profile configuration alters the transition
* probabilities in the core model (by adding <M->E>
* transitions and renormalizing the M transition
* distributions, for example; H2 configuration does this,
* though H3 does not). So you can't <CoreSample()> the
* <gm->hmm> safely. To avoid such things, the caller
* provides a clean copy of the core model in <core>.
*
* i endpoints are normalized/discretized to 1..<Lbins>, so
* we can collate i statistics from sampled sequences of
* varying L. Note this causes discretization artifacts,
* leading to underrepresentation of j=M and
* overrepresentation of i=1.
*
* Returns: <eslOK> on success; returns normalized sequence coords in
* <*ret_i1> and <*ret_i2>, and the model entry/exit coords
* in <*ret_k1> and <*ret_k2>.
*
* Xref: STL11/142-143
*/
static int
profile_local_endpoints(ESL_RANDOMNESS *r, P7_HMM *core, P7_PROFILE *gm, ESL_SQ *sq, P7_TRACE *tr, int Lbins,
int *ret_i1, int *ret_i2, int *ret_k1, int *ret_k2)
{
int status;
int i1,i2;
int k1,k2;
int t1,t2; /* entry/exit positions in local trace, tr */
int tg1, tg2; /* entry/exit positions in global trace, tr2 */
int tpos;
int nterm, cterm; /* offsets at N, C terminus. */
int L; /* inferred length from 3-part patching */
ESL_SQ *sq2 = NULL;
P7_TRACE *tr2 = NULL;
int failsafe = 0;
if (gm->mode != p7_UNILOCAL) ESL_XEXCEPTION(eslEINVAL, "profile must be unilocal");
if ((sq2 = esl_sq_CreateDigital(gm->abc)) == NULL) { status = eslEMEM; goto ERROR; }
if ((tr = p7_trace_Create()) == NULL) { status = eslEMEM; goto ERROR; }
/* sample local alignment from the implicit model */
if (gm->h2_mode) {
if ((status = p7_H2_ProfileEmit(r, gm, sq, tr)) != eslOK) goto ERROR;
} else {
if ((status = p7_ProfileEmit(r, gm, sq, tr)) != eslOK) goto ERROR;
}
/* Get initial trace coords */
for (tpos = 0; tpos < tr->N; tpos++) if (tr->st[tpos] == p7T_B) { t1 = tpos+1; break; }
for (tpos = tr->N-1; tpos >= 0; tpos--) if (tr->st[tpos] == p7T_E) { t2 = tpos-1; break; }
/* Match a core trace to this local trace by rejection sampling;
* this is to let us calculate sequence offsets; see comments above in preamble
*/
do {
if (failsafe++ == 100000) ESL_XEXCEPTION(eslENOHALT, "failed to match core,local traces in %d tries\n", failsafe);
if ((status = p7_CoreEmit(r, core, sq2, tr2)) != eslOK) goto ERROR;
for (tpos = 0; tpos < tr2->N; tpos++)
if (tr2->k[tpos] == tr->k[t1]) { tg1 = tpos; break; }
for (tpos = tr2->N-1; tpos >= 0; tpos--)
if (tr2->k[tpos] == tr->k[t2]) { tg2 = tpos; break; }
} while (tr2->st[tg1] != tr->st[t1] && tr2->st[tg2] != tr->st[t2]);
/* tg1..tg2 in core trace is now matched to t1..t2 in the profile trace.
* Calculate # of residues preceding tg1 and following tg2 in the core trace.
* A core trace can only generate residues from M or I states.
*/
for (nterm = 0, tpos = 0; tpos < tg1; tpos++)
if (tr2->st[tpos] == p7T_M || tr2->st[tpos] == p7T_I) nterm++;
for (cterm = 0, tpos = tr2->N-1; tpos > tg2; tpos--)
if (tr2->st[tpos] == p7T_M || tr2->st[tpos] == p7T_I) cterm++;
/* rectify the t2 endpoint, rolling back any trailing D path
*/
for (; t2 >= 0; t2--) if (tr->st[t2] == p7T_M) break;
if (t2 < t1) ESL_XEXCEPTION(eslEINCONCEIVABLE, "this only happens on an all-D path through profile");
/* determine initial endpoint coords from t1 and t2 */
i1 = tr->i[t1]; i2 = tr->i[t2];
k1 = tr->k[t1]; k2 = tr->k[t2];
/* offset the i coords. */
L = (i2-i1+1) + nterm + cterm;
i2 = (i2-i1+1) + nterm;
i1 = nterm+1;
/* normalize the i coords into range 1..Lbins, instead of 1..L */
i1 = ((i1-1) * Lbins / L) + 1;
i2 = ((i2-1) * Lbins / L) + 1;
*ret_i1 = i1;
*ret_i2 = i2;
*ret_k1 = k1;
*ret_k2 = k2;
p7_trace_Destroy(tr2);
esl_sq_Destroy(sq2);
return eslOK;
ERROR:
if (sq2 != NULL) esl_sq_Destroy(sq2);
if (tr2 != NULL) p7_trace_Destroy(tr2);
*ret_i1 = 0.;
*ret_i2 = 0.;
*ret_k1 = 0;
*ret_k2 = 0;
return status;
}
/* All input sources funnel through here.
* Here, <afp> is already allocated and initialized, and the input
* <bf> is opened successfully.
*/
static int
profillic_msafile_OpenBuffer(ESL_ALPHABET **byp_abc, ESL_BUFFER *bf, int format, ESLX_MSAFILE_FMTDATA *fmtd, ESLX_MSAFILE *afp)
{
ESL_ALPHABET *abc = NULL;
int alphatype = eslUNKNOWN;
int status;
/* if caller provided <fmtd>, copy it into afp->fmtd */
if (fmtd) eslx_msafile_fmtdata_Copy(fmtd, &(afp->fmtd));
/* Determine the format */
if (format == eslMSAFILE_UNKNOWN)
{
status = eslx_msafile_GuessFileFormat(afp->bf, &format, &(afp->fmtd));
if (status == eslENOFORMAT) ESL_XFAIL(eslENOFORMAT, afp->errmsg, "couldn't determine alignment input format"); /* ENOFORMAT is normal failure */
else if (status != eslOK) goto ERROR;
}
afp->format = format;
/* Determine the alphabet; set <abc>. (<abc> == NULL means text mode.) */
/* Note that GuessAlphabet() functions aren't allowed to use the inmap, because it isn't set yet */
#ifdef eslAUGMENT_ALPHABET
if (byp_abc && *byp_abc) /* Digital mode, and caller provided the alphabet */
{
abc = *byp_abc;
alphatype = abc->type;
}
else if (byp_abc) /* Digital mode, and caller wants us to guess and create an alphabet */
{
status = eslx_msafile_GuessAlphabet(afp, &alphatype);
if (status == eslENOALPHABET) ESL_XFAIL(eslENOALPHABET, afp->errmsg, "couldn't guess alphabet (maybe try --dna/--rna/--amino if available)");
else if (status != eslOK) goto ERROR;
if ( (abc = esl_alphabet_Create(alphatype)) == NULL) { status = eslEMEM; goto ERROR; }
}
#endif
if (abc && ! byp_abc) ESL_EXCEPTION(eslEINCONCEIVABLE, "Your version of Easel does not include digital alphabet code.");
/* ^^^^^^^^^^^^^^^^^ this test interacts tricksily with the #ifdef above */
afp->abc = abc; /* with afp->abc set, the inmap config functions know whether to do digital/text */
/**
* <pre>
* Configure the format-specific, digital or text mode character
* input map in afp->inmap.
* All of these must:
*
* set inmap[0] to an appropriate 'unknown' character, to replace
* invalid input with.
* set ' ' to eslDSQ_IGNORE (if we're supposed to accept and skip
* it), or map it to a gap, or set it as eslDSQ_ILLEGAL.
* in digital mode, copy the abc->inmap
* in text mode, decide if we should accept most any
* non-whitespace character (isgraph()), or if the format is
* inherently restrictive and we should go with isalpha() +
* some other valid characters "_-.~*" instead.
* </pre>
*/
switch (afp->format) {
case eslMSAFILE_A2M: status = esl_msafile_a2m_SetInmap( afp); break;
case eslMSAFILE_AFA: status = esl_msafile_afa_SetInmap( afp); break;
case eslMSAFILE_CLUSTAL: status = esl_msafile_clustal_SetInmap( afp); break;
case eslMSAFILE_CLUSTALLIKE: status = esl_msafile_clustal_SetInmap( afp); break;
case eslMSAFILE_PFAM: status = esl_msafile_stockholm_SetInmap(afp); break;
case eslMSAFILE_PHYLIP: status = esl_msafile_phylip_SetInmap( afp); break;
case eslMSAFILE_PHYLIPS: status = esl_msafile_phylip_SetInmap( afp); break;
case eslMSAFILE_PSIBLAST: status = esl_msafile_psiblast_SetInmap( afp); break;
case eslMSAFILE_SELEX: status = esl_msafile_selex_SetInmap( afp); break;
case eslMSAFILE_STOCKHOLM: status = esl_msafile_stockholm_SetInmap(afp); break;
case eslMSAFILE_PROFILLIC: status = eslOK; break; /// \todo status = profillic_esl_msafile_profile_SetInmap(afp); */ break;
default: ESL_XEXCEPTION(eslENOFORMAT, "no such alignment file format"); break;
}
if (esl_byp_IsReturned(byp_abc)) *byp_abc = abc;
return eslOK;
ERROR: /* on normal errors, afp is returned in an error state */
if (abc && ! esl_byp_IsProvided(byp_abc)) { esl_alphabet_Destroy(abc); }
if (esl_byp_IsReturned(byp_abc)) *byp_abc = NULL;
afp->abc = NULL;
return status;
}
/* Function: p7_ProfileConfig()
* Synopsis: Configure a search profile.
*
* Purpose: Given a model <hmm> with core probabilities, the null1
* model <bg>, a desired search <mode> (one of <p7_LOCAL>,
* <p7_GLOCAL>, <p7_UNILOCAL>, or <p7_UNIGLOCAL>), and an
* expected target sequence length <L>; configure the
* search model in <gm> with lod scores relative to the
* background frequencies in <bg>.
*
* Returns: <eslOK> on success; the profile <gm> now contains
* scores and is ready for searching target sequences.
*
* Throws: <eslEMEM> on allocation error.
*/
int
p7_ProfileConfig(const P7_HMM *hmm, const P7_BG *bg, P7_PROFILE *gm, int L, int mode)
{
int k, x, z; /* counters over states, residues, annotation */
int status;
float *occ = NULL;
float *tp, *rp;
float sc[p7_MAXCODE];
float Z;
/* Contract checks */
if (gm->abc->type != hmm->abc->type) ESL_XEXCEPTION(eslEINVAL, "HMM and profile alphabet don't match");
if (hmm->M > gm->allocM) ESL_XEXCEPTION(eslEINVAL, "profile too small to hold HMM");
if (! (hmm->flags & p7H_CONS)) ESL_XEXCEPTION(eslEINVAL, "HMM must have a consensus to transfer to the profile");
/* Copy some pointer references and other info across from HMM */
gm->M = hmm->M;
gm->max_length = hmm->max_length;
gm->mode = mode;
gm->roff = -1;
gm->eoff = -1;
gm->offs[p7_MOFFSET] = -1;
gm->offs[p7_FOFFSET] = -1;
gm->offs[p7_POFFSET] = -1;
if (gm->name != NULL) free(gm->name);
if (gm->acc != NULL) free(gm->acc);
if (gm->desc != NULL) free(gm->desc);
if ((status = esl_strdup(hmm->name, -1, &(gm->name))) != eslOK) goto ERROR;
if ((status = esl_strdup(hmm->acc, -1, &(gm->acc))) != eslOK) goto ERROR;
if ((status = esl_strdup(hmm->desc, -1, &(gm->desc))) != eslOK) goto ERROR;
if (hmm->flags & p7H_RF) strcpy(gm->rf, hmm->rf);
if (hmm->flags & p7H_MMASK) strcpy(gm->mm, hmm->mm);
if (hmm->flags & p7H_CONS) strcpy(gm->consensus, hmm->consensus); /* must be present, actually, so the flag test is just for symmetry w/ other optional HMM fields */
if (hmm->flags & p7H_CS) strcpy(gm->cs, hmm->cs);
for (z = 0; z < p7_NEVPARAM; z++) gm->evparam[z] = hmm->evparam[z];
for (z = 0; z < p7_NCUTOFFS; z++) gm->cutoff[z] = hmm->cutoff[z];
for (z = 0; z < p7_MAXABET; z++) gm->compo[z] = hmm->compo[z];
/* Entry scores. */
if (p7_profile_IsLocal(gm))
{
/* Local mode entry: occ[k] /( \sum_i occ[i] * (M-i+1))
* (Reduces to uniform 2/(M(M+1)) for occupancies of 1.0) */
Z = 0.;
ESL_ALLOC(occ, sizeof(float) * (hmm->M+1));
if ((status = p7_hmm_CalculateOccupancy(hmm, occ, NULL)) != eslOK) goto ERROR;
for (k = 1; k <= hmm->M; k++)
Z += occ[k] * (float) (hmm->M-k+1);
for (k = 1; k <= hmm->M; k++)
p7P_TSC(gm, k-1, p7P_BM) = log(occ[k] / Z); /* note off-by-one: entry at Mk stored as [k-1][BM] */
free(occ);
}
else /* glocal modes: left wing retraction; must be in log space for precision */
{
Z = log(hmm->t[0][p7H_MD]);
p7P_TSC(gm, 0, p7P_BM) = log(1.0 - hmm->t[0][p7H_MD]);
for (k = 1; k < hmm->M; k++)
{
p7P_TSC(gm, k, p7P_BM) = Z + log(hmm->t[k][p7H_DM]);
Z += log(hmm->t[k][p7H_DD]);
}
}
/* E state loop/move probabilities: nonzero for MOVE allows loops/multihits
* N,C,J transitions are set later by length config
*/
if (p7_profile_IsMultihit(gm)) {
gm->xsc[p7P_E][p7P_MOVE] = -eslCONST_LOG2;
gm->xsc[p7P_E][p7P_LOOP] = -eslCONST_LOG2;
gm->nj = 1.0f;
} else {
gm->xsc[p7P_E][p7P_MOVE] = 0.0f;
gm->xsc[p7P_E][p7P_LOOP] = -eslINFINITY;
gm->nj = 0.0f;
}
/* Transition scores. */
for (k = 1; k < gm->M; k++) {
tp = gm->tsc + k * p7P_NTRANS;
tp[p7P_MM] = log(hmm->t[k][p7H_MM]);
tp[p7P_MI] = log(hmm->t[k][p7H_MI]);
tp[p7P_MD] = log(hmm->t[k][p7H_MD]);
tp[p7P_IM] = log(hmm->t[k][p7H_IM]);
tp[p7P_II] = log(hmm->t[k][p7H_II]);
tp[p7P_DM] = log(hmm->t[k][p7H_DM]);
tp[p7P_DD] = log(hmm->t[k][p7H_DD]);
}
/* Match emission scores. */
sc[hmm->abc->K] = -eslINFINITY; /* gap character */
sc[hmm->abc->Kp-2] = -eslINFINITY; /* nonresidue character */
sc[hmm->abc->Kp-1] = -eslINFINITY; /* missing data character */
for (k = 1; k <= hmm->M; k++) {
for (x = 0; x < hmm->abc->K; x++)
sc[x] = log((double)hmm->mat[k][x] / bg->f[x]);
esl_abc_FExpectScVec(hmm->abc, sc, bg->f);
for (x = 0; x < hmm->abc->Kp; x++) {
rp = gm->rsc[x] + k * p7P_NR;
rp[p7P_MSC] = sc[x];
}
}
/* Insert emission scores */
/* SRE, Fri Dec 5 08:41:08 2008: We currently hardwire insert scores
* to 0, i.e. corresponding to the insertion emission probabilities
* being equal to the background probabilities. Benchmarking shows
* that setting inserts to informative emission distributions causes
* more problems than it's worth: polar biased composition hits
* driven by stretches of "insertion" occur, and are difficult to
* correct for.
*/
for (x = 0; x < gm->abc->Kp; x++)
{
for (k = 1; k < hmm->M; k++) p7P_ISC(gm, k, x) = 0.0f;
p7P_ISC(gm, hmm->M, x) = -eslINFINITY; /* init I_M to impossible. */
}
for (k = 1; k <= hmm->M; k++) p7P_ISC(gm, k, gm->abc->K) = -eslINFINITY; /* gap symbol */
for (k = 1; k <= hmm->M; k++) p7P_ISC(gm, k, gm->abc->Kp-2) = -eslINFINITY; /* nonresidue symbol */
for (k = 1; k <= hmm->M; k++) p7P_ISC(gm, k, gm->abc->Kp-1) = -eslINFINITY; /* missing data symbol */
#if 0
/* original (informative) insert setting: relies on sc[K, Kp-1] initialization to -inf above */
for (k = 1; k < hmm->M; k++) {
for (x = 0; x < hmm->abc->K; x++)
sc[x] = log(hmm->ins[k][x] / bg->f[x]);
esl_abc_FExpectScVec(hmm->abc, sc, bg->f);
for (x = 0; x < hmm->abc->Kp; x++) {
rp = gm->rsc[x] + k*p7P_NR;
rp[p7P_ISC] = sc[x];
}
}
for (x = 0; x < hmm->abc->Kp; x++)
p7P_ISC(gm, hmm->M, x) = -eslINFINITY; /* init I_M to impossible. */
#endif
/* Remaining specials, [NCJ][MOVE | LOOP] are set by ReconfigLength()
*/
gm->L = 0; /* force ReconfigLength to reconfig */
if ((status = p7_ReconfigLength(gm, L)) != eslOK) goto ERROR;
return eslOK;
ERROR:
if (occ != NULL) free(occ);
return status;
}
/* Function: p7_ProfileConfig()
* Synopsis: Configure a search profile.
* Incept: SRE, Sun Sep 25 12:21:25 2005 [St. Louis]
*
* Purpose: Given a model <hmm> with core probabilities, the null1
* model <bg>, a desired search <mode> (one of <p7_LOCAL>,
* <p7_GLOCAL>, <p7_UNILOCAL>, or <p7_UNIGLOCAL>), and an
* expected target sequence length <L>; configure the
* search model in <gm> with lod scores relative to the
* background frequencies in <bg>.
*
* Returns: <eslOK> on success; the profile <gm> now contains
* scores and is ready for searching target sequences.
*
* Throws: <eslEMEM> on allocation error.
*/
int
p7_ProfileConfig(const P7_HMM *hmm, const P7_BG *bg, P7_PROFILE *gm, int L, int mode)
{
int k, x, z; /* counters over states, residues, annotation */
int status;
float *occ = NULL;
float *tp, *rp;
float sc[p7_MAXCODE];
float mthresh;
float Z;
/* Contract checks */
if (gm->abc->type != hmm->abc->type) ESL_XEXCEPTION(eslEINVAL, "HMM and profile alphabet don't match");
if (hmm->M > gm->allocM) ESL_XEXCEPTION(eslEINVAL, "profile too small to hold HMM");
/* Copy some pointer references and other info across from HMM */
gm->M = hmm->M;
gm->mode = mode;
gm->roff = -1;
gm->eoff = -1;
gm->offs[p7_MOFFSET] = -1;
gm->offs[p7_FOFFSET] = -1;
gm->offs[p7_POFFSET] = -1;
if (gm->name != NULL) free(gm->name);
if (gm->acc != NULL) free(gm->acc);
if (gm->desc != NULL) free(gm->desc);
if ((status = esl_strdup(hmm->name, -1, &(gm->name))) != eslOK) goto ERROR;
if ((status = esl_strdup(hmm->acc, -1, &(gm->acc))) != eslOK) goto ERROR;
if ((status = esl_strdup(hmm->desc, -1, &(gm->desc))) != eslOK) goto ERROR;
if (hmm->flags & p7H_RF) strcpy(gm->rf, hmm->rf);
if (hmm->flags & p7H_CS) strcpy(gm->cs, hmm->cs);
for (z = 0; z < p7_NEVPARAM; z++) gm->evparam[z] = hmm->evparam[z];
for (z = 0; z < p7_NCUTOFFS; z++) gm->cutoff[z] = hmm->cutoff[z];
for (z = 0; z < p7_MAXABET; z++) gm->compo[z] = hmm->compo[z];
/* Determine the "consensus" residue for each match position.
* This is only used for alignment displays, not in any calculations.
*/
if (hmm->abc->type == eslAMINO) mthresh = 0.5;
else if (hmm->abc->type == eslDNA) mthresh = 0.9;
else if (hmm->abc->type == eslRNA) mthresh = 0.9;
else mthresh = 0.5;
gm->consensus[0] = ' ';
for (k = 1; k <= hmm->M; k++) {
x = esl_vec_FArgMax(hmm->mat[k], hmm->abc->K);
gm->consensus[k] = ((hmm->mat[k][x] > mthresh) ? toupper(hmm->abc->sym[x]) : tolower(hmm->abc->sym[x]));
}
gm->consensus[hmm->M+1] = '\0';
/* Entry scores. */
if (p7_profile_IsLocal(gm))
{
/* Local mode entry: occ[k] /( \sum_i occ[i] * (M-i+1))
* (Reduces to uniform 2/(M(M+1)) for occupancies of 1.0) */
Z = 0.;
ESL_ALLOC_WITH_TYPE(occ, float*, sizeof(float) * (hmm->M+1));
if ((status = p7_hmm_CalculateOccupancy(hmm, occ, NULL)) != eslOK) goto ERROR;
for (k = 1; k <= hmm->M; k++)
Z += occ[k] * (float) (hmm->M-k+1);
for (k = 1; k <= hmm->M; k++)
p7P_TSC(gm, k-1, p7P_BM) = log((double)(occ[k] / Z)); /* note off-by-one: entry at Mk stored as [k-1][BM] */
free(occ);
}
else /* glocal modes: left wing retraction; must be in log space for precision */
{
/* Function: p7_hmm_mpi_Unpack()
* Synopsis: Unpacks one HMM from an MPI buffer.
*
* Purpose: Unpack one HMM from MPI packed buffer
* <buf>, starting from position <*pos>, where the total length
* of the buffer in bytes is <n>. The new HMM is allocated here.
*
* Caller may or may not already know what alphabet the HMM
* is expected to be in. A reference to the current
* alphabet is passed in <byp_abc>. If the alphabet is unknown,
* pass <*byp_abc = NULL>, and when the HMM is received, an
* appropriate new alphabet object is allocated and passed
* back to the caller via <*byp_abc>. If the alphabet is
* already known, <*byp_abc> is that alphabet, and the new
* HMM's alphabet type is verified to agree with it. This
* mechanism allows an application to let the first HMM
* determine the alphabet type for the application, while
* still keeping the alphabet under the application's scope
* of control.
*
* Args: buf - MPI packed buffer to unpack
* n - total length of <buf> in bytes
* pos - current parsing/unpacking position in <buf>
* comm - MPI communicator
* byp_abc - BYPASS: <*byp_abc> == ESL_ALPHABET *> if known;
* <*byp_abc> == NULL> if alphabet unknown;
* ret_hmm - RETURN: ptr to newly allocated, unpacked profile
*
* Returns: <eslOK> on success. <*pos> is updated to the position of
* the next element in <buf> to unpack (if any). <*ret_hmm>
* contains a newly allocated HMM, which the caller is
* responsible for free'ing. If <*byp_abc> was passed as
* <NULL>, it now points to an <ESL_ALPHABET> object that
* was allocated here; caller is responsible for free'ing
* this.
*
* Returns <eslEINCOMPAT> if the HMM is in a different
* alphabet than <*byp_abc> said to expect. In this case,
* <*byp_abc> is unchanged, <*buf> and <*nalloc> may have been
* changed, and <*ret_hmm> is <NULL>.
*
* Throws: <eslESYS> on an MPI call failure. <eslEMEM> on allocation failure.
* In either case, <*ret_hmm> is <NULL>, and the state of <buf>
* and <*pos> is undefined and should be considered to be corrupted.
*/
int
p7_hmm_mpi_Unpack(char *buf, int n, int *pos, MPI_Comm comm, ESL_ALPHABET **byp_abc, P7_HMM **ret_hmm)
{
P7_HMM *hmm = NULL;
ESL_ALPHABET *abc = NULL;
int64_t offset;
int M, K, atype;
int status;
/* Use the CreateShell/CreateBody interface, because that interface allocates our optional fields, using <flags> */
if (( hmm = p7_hmm_CreateShell() ) == NULL) { status = eslEMEM; goto ERROR; }
/* First, unpack info that we need for HMM body allocation */
if (MPI_Unpack(buf, n, pos, &M, 1, MPI_INT, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack(buf, n, pos, &(hmm->flags), 1, MPI_INT, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack(buf, n, pos, &atype, 1, MPI_INT, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed");
/* Set or verify the alphabet */
if (*byp_abc == NULL) { /* alphabet unknown. create new one */
if ( (abc = esl_alphabet_Create(atype)) == NULL) { status = eslEMEM; goto ERROR; }
} else { /* already known: check it */
abc = *byp_abc;
if (abc->type != atype){ status = eslEINCOMPAT; goto ERROR; }
}
K = abc->K; /* For convenience below. */
/* Allocate the HMM body */
if ((status = p7_hmm_CreateBody(hmm, M, abc)) != eslOK) goto ERROR;
/* Unpack the rest of the HMM */
if (MPI_Unpack( buf, n, pos, hmm->t[0], p7H_NTRANSITIONS*(M+1), MPI_FLOAT, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack( buf, n, pos, hmm->mat[0], K*(M+1), MPI_FLOAT, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack( buf, n, pos, hmm->ins[0], K*(M+1), MPI_FLOAT, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed");
if ((status = esl_mpi_UnpackOpt( buf, n, pos, (void**) &(hmm->name), NULL, MPI_CHAR, comm)) != eslOK) goto ERROR;
if ((status = esl_mpi_UnpackOpt( buf, n, pos, (void**) &(hmm->acc), NULL, MPI_CHAR, comm)) != eslOK) goto ERROR;
if ((status = esl_mpi_UnpackOpt( buf, n, pos, (void**) &(hmm->desc), NULL, MPI_CHAR, comm)) != eslOK) goto ERROR;
if (hmm->flags & p7H_RF) { if (MPI_Unpack(buf, n, pos, hmm->rf, M+2, MPI_CHAR, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed"); }
if (hmm->flags & p7H_MMASK) { if (MPI_Unpack(buf, n, pos, hmm->mm, M+2, MPI_CHAR, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed"); }
if (hmm->flags & p7H_CONS) { if (MPI_Unpack(buf, n, pos, hmm->consensus, M+2, MPI_CHAR, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed"); }
if (hmm->flags & p7H_CS) { if (MPI_Unpack(buf, n, pos, hmm->cs, M+2, MPI_CHAR, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed"); }
if (hmm->flags & p7H_CA) { if (MPI_Unpack(buf, n, pos, hmm->ca, M+2, MPI_CHAR, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed"); }
if ((status = esl_mpi_UnpackOpt( buf, n, pos, (void**)&(hmm->comlog), NULL, MPI_CHAR, comm)) != eslOK) goto ERROR;
if (MPI_Unpack( buf, n, pos, &(hmm->nseq), 1, MPI_INT, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack( buf, n, pos, &(hmm->eff_nseq), 1, MPI_FLOAT, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack( buf, n, pos, &(hmm->max_length), 1, MPI_FLOAT, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed");
if ((status = esl_mpi_UnpackOpt( buf, n, pos, (void**) &(hmm->ctime), NULL, MPI_CHAR, comm)) != eslOK) goto ERROR;
if (hmm->flags & p7H_MAP) { if (MPI_Unpack(buf, n, pos, hmm->map, M+1, MPI_INT, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed"); }
if (MPI_Unpack( buf, n, pos, &(hmm->checksum), 1, MPI_UINT32_T, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack( buf, n, pos, hmm->evparam, p7_NEVPARAM, MPI_FLOAT, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack( buf, n, pos, hmm->cutoff, p7_NCUTOFFS, MPI_FLOAT, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack( buf, n, pos, hmm->compo, p7_MAXABET, MPI_FLOAT, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack( buf, n, pos, &offset, 1, MPI_INT64_T, comm) != MPI_SUCCESS) ESL_XEXCEPTION(eslESYS, "mpi unpack failed");
hmm->offset = offset; /* receive as int64_t, then cast to off_t, which is probably int64_t (but not guaranteed) */
*byp_abc = abc; /* works even if caller provided *byp_abc, because then abc==*byp_abc already */
*ret_hmm = hmm;
return eslOK;
ERROR:
if (hmm) p7_hmm_Destroy(hmm);
if (abc && *byp_abc == NULL) esl_alphabet_Destroy(abc);
*ret_hmm = NULL;
return status;
}
/* Function: esl_stats_LinearRegression()
* Synopsis: Fit data to a straight line.
* Incept: SRE, Sat May 26 11:33:46 2007 [Janelia]
*
* Purpose: Fit <n> points <x[i]>, <y[i]> to a straight line
* $y = a + bx$ by linear regression.
*
* The $x_i$ are taken to be known, and the $y_i$ are taken
* to be observed quantities associated with a sampling
* error $\sigma_i$. If known, the standard deviations
* $\sigma_i$ for $y_i$ are provided in the <sigma> array.
* If they are unknown, pass <sigma = NULL>, and the
* routine will proceed with the assumption that $\sigma_i
* = 1$ for all $i$.
*
* The maximum likelihood estimates for $a$ and $b$ are
* optionally returned in <opt_a> and <opt_b>.
*
* The estimated standard deviations of $a$ and $b$ and
* their estimated covariance are optionally returned in
* <opt_sigma_a>, <opt_sigma_b>, and <opt_cov_ab>.
*
* The Pearson correlation coefficient is optionally
* returned in <opt_cc>.
*
* The $\chi^2$ P-value for the regression fit is
* optionally returned in <opt_Q>. This P-value may only be
* obtained when the $\sigma_i$ are known. If <sigma> is
* passed as <NULL> and <opt_Q> is requested, <*opt_Q> is
* set to 1.0.
*
* This routine follows the description and algorithm in
* \citep[pp.661-666]{Press93}.
*
* <n> must be greater than 2; at least two x[i] must
* differ; and if <sigma> is provided, all <sigma[i]> must
* be $>0$. If any of these conditions isn't met, the
* routine throws <eslEINVAL>.
*
* Args: x - x[0..n-1]
* y - y[0..n-1]
* sigma - sample error in observed y_i
* n - number of data points
* opt_a - optRETURN: intercept estimate
* opt_b - optRETURN: slope estimate
* opt_sigma_a - optRETURN: error in estimate of a
* opt_sigma_b - optRETURN: error in estimate of b
* opt_cov_ab - optRETURN: covariance of a,b estimates
* opt_cc - optRETURN: Pearson correlation coefficient for x,y
* opt_Q - optRETURN: X^2 P-value for linear fit
*
* Returns: <eslOK> on success.
*
* Throws: <eslEMEM> on allocation error;
* <eslEINVAL> if a contract condition isn't met;
* <eslENORESULT> if the chi-squared test fails.
* In these cases, all optional return values are set to 0.
*/
int
esl_stats_LinearRegression(const double *x, const double *y, const double *sigma, int n,
double *opt_a, double *opt_b,
double *opt_sigma_a, double *opt_sigma_b, double *opt_cov_ab,
double *opt_cc, double *opt_Q)
{
int status;
double *t = NULL;
double S, Sx, Sy, Stt;
double Sxy, Sxx, Syy;
double a, b, sigma_a, sigma_b, cov_ab, cc, X2, Q;
double xdev, ydev;
double tmp;
int i;
/* Contract checks. */
if (n <= 2) ESL_XEXCEPTION(eslEINVAL, "n must be > 2 for linear regression fitting");
if (sigma != NULL)
for (i = 0; i < n; i++) if (sigma[i] <= 0.) ESL_XEXCEPTION(eslEINVAL, "sigma[%d] <= 0", i);
status = eslEINVAL;
for (i = 0; i < n; i++) if (x[i] != 0.) { status = eslOK; break; }
if (status != eslOK) ESL_XEXCEPTION(eslEINVAL, "all x[i] are 0.");
/* Allocations */
ESL_ALLOC(t, sizeof(double) * n);
/* S = \sum_{i=1}{n} \frac{1}{\sigma_i^2}. (S > 0.) */
if (sigma != NULL) { for (S = 0., i = 0; i < n; i++) S += 1./ (sigma[i] * sigma[i]); }
else S = (double) n;
/* S_x = \sum_{i=1}{n} \frac{x[i]}{ \sigma_i^2} (Sx real.) */
for (Sx = 0., i = 0; i < n; i++) {
if (sigma == NULL) Sx += x[i];
else Sx += x[i] / (sigma[i] * sigma[i]);
}
/* S_y = \sum_{i=1}{n} \frac{y[i]}{\sigma_i^2} (Sy real.) */
for (Sy = 0., i = 0; i < n; i++) {
if (sigma == NULL) Sy += y[i];
else Sy += y[i] / (sigma[i] * sigma[i]);
}
/* t_i = \frac{1}{\sigma_i} \left( x_i - \frac{S_x}{S} \right) (t_i real) */
for (i = 0; i < n; i++) {
t[i] = x[i] - Sx/S;
if (sigma != NULL) t[i] /= sigma[i];
}
/* S_{tt} = \sum_{i=1}^n t_i^2 (if at least one x is != 0, Stt > 0) */
for (Stt = 0., i = 0; i < n; i++) { Stt += t[i] * t[i]; }
/* b = \frac{1}{S_{tt}} \sum_{i=1}^{N} \frac{t_i y_i}{\sigma_i} */
for (b = 0., i = 0; i < n; i++) {
if (sigma != NULL) { b += t[i]*y[i] / sigma[i]; }
else { b += t[i]*y[i]; }
}
b /= Stt;
/* a = \frac{ S_y - S_x b } {S} */
a = (Sy - Sx * b) / S;
/* \sigma_a^2 = \frac{1}{S} \left( 1 + \frac{ S_x^2 }{S S_{tt}} \right) */
sigma_a = sqrt ((1. + (Sx*Sx) / (S*Stt)) / S);
/* \sigma_b = \frac{1}{S_{tt}} */
sigma_b = sqrt (1. / Stt);
/* Cov(a,b) = - \frac{S_x}{S S_{tt}} */
cov_ab = -Sx / (S * Stt);
/* Pearson correlation coefficient */
Sxy = Sxx = Syy = 0.;
for (i = 0; i < n; i++) {
if (sigma != NULL) {
xdev = (x[i] / (sigma[i] * sigma[i])) - (Sx / n);
ydev = (y[i] / (sigma[i] * sigma[i])) - (Sy / n);
} else {
xdev = x[i] - (Sx / n);
ydev = y[i] - (Sy / n);
}
Sxy += xdev * ydev;
Sxx += xdev * xdev;
Syy += ydev * ydev;
}
cc = Sxy / (sqrt(Sxx) * sqrt(Syy));
/* \chi^2 */
for (X2 = 0., i = 0; i < n; i++) {
tmp = y[i] - a - b*x[i];
if (sigma != NULL) tmp /= sigma[i];
X2 += tmp*tmp;
}
/* We can calculate a goodness of fit if we know the \sigma_i */
if (sigma != NULL) {
if (esl_stats_ChiSquaredTest(n-2, X2, &Q) != eslOK) { status = eslENORESULT; goto ERROR; }
} else Q = 1.0;
/* If we didn't use \sigma_i, adjust the sigmas for a,b */
if (sigma == NULL) {
tmp = sqrt(X2 / (double)(n-2));
sigma_a *= tmp;
sigma_b *= tmp;
}
/* Done. Set up for normal return.
*/
free(t);
if (opt_a != NULL) *opt_a = a;
if (opt_b != NULL) *opt_b = b;
if (opt_sigma_a != NULL) *opt_sigma_a = sigma_a;
if (opt_sigma_b != NULL) *opt_sigma_b = sigma_b;
if (opt_cov_ab != NULL) *opt_cov_ab = cov_ab;
if (opt_cc != NULL) *opt_cc = cc;
if (opt_Q != NULL) *opt_Q = Q;
return eslOK;
ERROR:
if (t != NULL) free(t);
if (opt_a != NULL) *opt_a = 0.;
if (opt_b != NULL) *opt_b = 0.;
if (opt_sigma_a != NULL) *opt_sigma_a = 0.;
if (opt_sigma_b != NULL) *opt_sigma_b = 0.;
if (opt_cov_ab != NULL) *opt_cov_ab = 0.;
if (opt_cc != NULL) *opt_cc = 0.;
if (opt_Q != NULL) *opt_Q = 0.;
return status;
}
/* Function: matassign2hmm()
*
* Purpose: Given an assignment of alignment columns to match vs.
* insert, finish the final part of the model construction
* calculation that is constant between model construction
* algorithms.
*
* Args: msa - multiple sequence alignment
* matassign - 1..alen bit flags for column assignments
* ret_hmm - RETURN: counts-form HMM
* opt_tr - optRETURN: array of tracebacks for aseq's
*
* Return: <eslOK> on success.
* <eslENORESULT> if no consensus columns are identified.
*
* ret_hmm and opt_tr alloc'ed here.
*/
static int
matassign2hmm(ESL_MSA *msa, int *matassign, P7_HMM **ret_hmm, P7_TRACE ***opt_tr)
{
int status; /* return status */
P7_HMM *hmm = NULL; /* RETURN: new hmm */
P7_TRACE **tr = NULL; /* RETURN: 0..nseq-1 fake traces */
int M; /* length of new model in match states */
int idx; /* counter over sequences */
int apos; /* counter for aligned columns */
#ifdef p7_DEBUGGING
char errbuf[eslERRBUFSIZE];
#endif
/* apply the model mask in the 'GC MM' row */
do_modelmask(msa);
/* How many match states in the HMM? */
for (M = 0, apos = 1; apos <= msa->alen; apos++)
if (matassign[apos]) M++;
if (M == 0) { status = eslENORESULT; goto ERROR; }
/* Make fake tracebacks for each seq */
ESL_ALLOC(tr, sizeof(P7_TRACE *) * msa->nseq);
if ((status = p7_trace_FauxFromMSA(msa, matassign, p7_MSA_COORDS, tr)) != eslOK) goto ERROR;
for (idx = 0; idx < msa->nseq; idx++)
{
if ((status = p7_trace_Doctor(tr[idx], NULL, NULL)) != eslOK) goto ERROR;
#ifdef p7_DEBUGGING
if ((status = p7_trace_Validate(tr[idx], msa->abc, msa->ax[idx], errbuf)) != eslOK)
ESL_XEXCEPTION(eslFAIL, "validation failed: %s", errbuf);
#endif
}
/* Build count model from tracebacks */
if ((hmm = p7_hmm_Create(M, msa->abc)) == NULL) { status = eslEMEM; goto ERROR; }
if ((status = p7_hmm_Zero(hmm)) != eslOK) goto ERROR;
for (idx = 0; idx < msa->nseq; idx++) {
if (tr[idx] == NULL) continue; /* skip rare examples of empty sequences */
if ((status = p7_trace_Count(hmm, msa->ax[idx], msa->wgt[idx], tr[idx])) != eslOK) goto ERROR;
}
hmm->nseq = msa->nseq;
hmm->eff_nseq = msa->nseq;
/* Transfer annotation from the MSA to the new model
*/
if ((status = annotate_model(hmm, matassign, msa)) != eslOK) goto ERROR;
/* Reset #=RF line of alignment to reflect our assignment
* of match, delete. matassign is valid from 1..alen and is off
* by one from msa->rf.
*/
if (msa->rf == NULL) ESL_ALLOC(msa->rf, sizeof(char) * (msa->alen + 1));
for (apos = 1; apos <= msa->alen; apos++)
msa->rf[apos-1] = matassign[apos] ? 'x' : '.';
msa->rf[msa->alen] = '\0';
if (opt_tr != NULL) *opt_tr = tr;
else p7_trace_DestroyArray(tr, msa->nseq);
*ret_hmm = hmm;
return eslOK;
ERROR:
if (tr != NULL) p7_trace_DestroyArray(tr, msa->nseq);
if (hmm != NULL) p7_hmm_Destroy(hmm);
if (opt_tr != NULL) *opt_tr = NULL;
*ret_hmm = NULL;
return status;
}
/* Function: p7_CoreEmit()
* Incept: SRE, Tue Jan 9 10:20:51 2007 [Janelia]
*
* Purpose: Generate (sample) a sequence from a core HMM <hmm>.
*
* Optionally return the sequence and/or its trace in <sq>
* and <tr>, respectively, which the caller has
* allocated. Having the caller provide these reusable
* objects allows re-use of both <sq> and <tr> in repeated
* calls, saving malloc/free wastage. Either can be passed
* as <NULL> if it isn't needed.
*
* This does not set any fields in the <sq> except for the
* sequence itself. Caller must set the name, and any other
* annotation it wants to add.
*
* Trace is relative to the core model: it may include
* I_0 and I_M states, B->DD->M entry is explicit, and a
* 0 length generated sequence is possible.
*
* Args: r - source of randomness
* hmm - core HMM to generate from
* sq - opt: digital sequence sampled (or NULL)
* tr - opt: trace sampled (or NULL)
*
* Returns: <eslOK> on success;
* optionally return the digital sequence through <ret_sq>,
* and optionally return its trace in <ret_tr>.
*
* Throws: <eslECORRUPT> if emission gets us into an illegal state,
* probably indicating that a probability that should have
* been zero wasn't.
*
* Throws <eslEMEM> on a reallocation error.
*
* In these cases, the contents of <sq> and <tr> may be
* corrupted. Caller should not trust their data, but may
* safely reuse them.
*
* Xref: STL11/124.
*/
int
p7_CoreEmit(ESL_RANDOMNESS *r, const P7_HMM *hmm, ESL_SQ *sq, P7_TRACE *tr)
{
int k = 0; /* position in model nodes 1..M */
int i = 0; /* position in sequence 1..L */
char st = p7T_B; /* state type */
int x; /* sampled residue */
int status;
if (sq != NULL) esl_sq_Reuse(sq);
if (tr != NULL) {
if ((status = p7_trace_Reuse(tr)) != eslOK) goto ERROR;
if ((status = p7_trace_Append(tr, st, k, i)) != eslOK) goto ERROR;
}
while (st != p7T_E)
{
/* Sample next state type, given current state type (and current k) */
switch (st) {
case p7T_B:
case p7T_M:
switch (esl_rnd_FChoose(r, hmm->t[k], 3)) {
case 0: st = p7T_M; break;
case 1: st = p7T_I; break;
case 2: st = p7T_D; break;
default: ESL_XEXCEPTION(eslEINCONCEIVABLE, "impossible.");
}
break;
case p7T_I:
switch (esl_rnd_FChoose(r, hmm->t[k]+3, 2)) {
case 0: st = p7T_M; break;
case 1: st = p7T_I; break;
default: ESL_XEXCEPTION(eslEINCONCEIVABLE, "impossible.");
}
break;
case p7T_D:
switch (esl_rnd_FChoose(r, hmm->t[k]+5, 2)) {
case 0: st = p7T_M; break;
case 1: st = p7T_D; break;
default: ESL_XEXCEPTION(eslEINCONCEIVABLE, "impossible.");
}
break;
default: ESL_XEXCEPTION(eslECORRUPT, "impossible state reached during emission");
}
/* Bump k,i if needed, depending on new state type */
if (st == p7T_M || st == p7T_D) k++;
if (st == p7T_M || st == p7T_I) i++;
/* a transit to M_M+1 is a transit to the E state */
if (k == hmm->M+1) {
if (st == p7T_M) { st = p7T_E; k = 0; }
else ESL_XEXCEPTION(eslECORRUPT, "failed to reach E state properly");
}
/* Sample new residue x if in match or insert */
if (st == p7T_M) x = esl_rnd_FChoose(r, hmm->mat[k], hmm->abc->K);
else if (st == p7T_I) x = esl_rnd_FChoose(r, hmm->ins[k], hmm->abc->K);
else x = eslDSQ_SENTINEL;
/* Add state to trace */
if (tr != NULL) {
if ((status = p7_trace_Append(tr, st, k, i)) != eslOK) goto ERROR;
}
/* Add x to sequence */
if (sq != NULL && x != eslDSQ_SENTINEL)
if ((status = esl_sq_XAddResidue(sq, x)) != eslOK) goto ERROR;
}
/* Terminate the trace and sequence (both are optional, remember) */
if (tr != NULL) { tr->M = hmm->M; tr->L = i; }
if (sq != NULL && (status = esl_sq_XAddResidue(sq, eslDSQ_SENTINEL)) != eslOK) goto ERROR;
return eslOK;
ERROR:
return status;
}
/* Function: p7_ProfileEmit()
* Synopsis: Sample a sequence from the search form of the model.
* Incept: SRE, Mon Jan 22 10:23:28 2007 [Janelia]
*
* Purpose: Sample a sequence from the implicit
* probabilistic model of a Plan7 profile <gm>. This
* requires also having the core probabilities of
* the accompanying <hmm>, and the background
* frequencies of null1 model <bg>.
*
* Optionally return the sequence and/or its trace in <sq>
* and <tr>, respectively. Caller has allocated space for
* both of these, though they may get reallocated/grown
* here. Either can be passed as <NULL> if unneeded.
*
* Only the sequence field is set in the <sq>. Caller must
* set the name, plus any other fields it wants to set. If
* the <sq> was created in digital mode, this is the <sq->dsq>;
* if the <sq> was created in text mode, this is <sq->seq>.
*
* <p7_ProfileEmit()> deliberately uses an <ESL_SQ> object
* instead of a plain <ESL_DSQ *> or <char *> string, to
* take advantage of the object's support for dynamic
* reallocation of seq length, and to allow both digital and
* text mode generation.
*
* Args: r - source of randomness
* hmm - core probabilities of the profile
* gm - configured search profile
* sq - optRETURN: sampled sequence
* tr - optRETURN: sampled trace
*
* Throws: (no abnormal error conditions)
*/
int
p7_ProfileEmit(ESL_RANDOMNESS *r, const P7_HMM *hmm, const P7_PROFILE *gm, const P7_BG *bg, ESL_SQ *sq, P7_TRACE *tr)
{
char prv, st; /* prev, current state type */
int k = 0; /* position in model nodes 1..M */
int i = 0; /* position in sequence 1..L */
int x; /* sampled residue */
int kend = hmm->M; /* predestined end node */
int status;
float xt[p7P_NXSTATES][p7P_NXTRANS];
/* Backcalculate the probabilities in the special states (loop and length model) */
for (i = 0; i < p7P_NXSTATES; i++)
for (x = 0; x < p7P_NXTRANS; x++)
xt[i][x] = exp(gm->xsc[i][x]);
if (sq != NULL) esl_sq_Reuse(sq);
if (tr != NULL) {
if ((status = p7_trace_Reuse(tr)) != eslOK) goto ERROR;
if ((status = p7_trace_Append(tr, p7T_S, k, i)) != eslOK) goto ERROR;
if ((status = p7_trace_Append(tr, p7T_N, k, i)) != eslOK) goto ERROR;
}
st = p7T_N;
i = 0;
while (st != p7T_T)
{
/* Sample a state transition. After this section, prv and st (prev->current state) are set;
* k also gets set if we make a B->Mk entry transition.
*/
prv = st;
switch (st) {
case p7T_B:
if (p7_profile_IsLocal(gm))
{ /* local mode: enter the implicit profile: choose our entry and our predestined exit */
if ((status = sample_endpoints(r, gm, &k, &kend)) != eslOK) goto ERROR;
st = p7T_M; /* must be, because left wing is retracted */
}
else
{ /* glocal mode: treat B as M_0, use its transitions to MID. */
/* FIXME: this is wrong. It should sample from B->Mk distribution! */
switch (esl_rnd_FChoose(r, P7H_TMAT(hmm, 0), p7H_NTMAT)) {
case 0: st = p7T_M; k = 1; break;
case 1: st = p7T_I; k = 0; break;
case 2: st = p7T_D; k = 1; break;
default: ESL_XEXCEPTION(eslEINCONCEIVABLE, "impossible.");
}
}
break;
case p7T_M:
if (k == kend) st = p7T_E; /* check our preordained fate */
else {
switch (esl_rnd_FChoose(r, P7H_TMAT(hmm, k), p7H_NTMAT)) {
case 0: st = p7T_M; break;
case 1: st = p7T_I; break;
case 2: st = p7T_D; break;
default: ESL_XEXCEPTION(eslEINCONCEIVABLE, "impossible.");
}
}
break;
case p7T_D:
if (k == kend) st = p7T_E;
else st = (esl_rnd_FChoose(r, P7H_TDEL(hmm, k), p7H_NTDEL) == 0) ? p7T_M : p7T_D;
break;
case p7T_I: st = (esl_rnd_FChoose(r, P7H_TINS(hmm, k), p7H_NTINS) == 0) ? p7T_M : p7T_I; break;
case p7T_N: st = (esl_rnd_FChoose(r, xt[p7P_N], p7P_NXTRANS) == p7P_MOVE) ? p7T_B : p7T_N; break;
case p7T_E: st = (esl_rnd_FChoose(r, xt[p7P_E], p7P_NXTRANS) == p7P_MOVE) ? p7T_C : p7T_J; break;
case p7T_C: st = (esl_rnd_FChoose(r, xt[p7P_C], p7P_NXTRANS) == p7P_MOVE) ? p7T_T : p7T_C; break;
case p7T_J: st = (esl_rnd_FChoose(r, xt[p7P_J], p7P_NXTRANS) == p7P_MOVE) ? p7T_B : p7T_J; break;
default: ESL_XEXCEPTION(eslECORRUPT, "impossible state reached during emission");
}
/* Based on the transition we just sampled, update k. */
if (st == p7T_E) k = 0;
else if (st == p7T_M && prv != p7T_B) k++; /* be careful about B->Mk, where we already set k */
else if (st == p7T_D) k++;
/* Based on the transition we just sampled, generate a residue. */
if (st == p7T_M) x = esl_rnd_FChoose(r, hmm->mat[k], hmm->abc->K);
else if (st == p7T_I) x = esl_rnd_FChoose(r, hmm->ins[k], hmm->abc->K);
else if ((st == p7T_N || st == p7T_C || st == p7T_J) && prv==st) x = esl_rnd_FChoose(r, bg->f, hmm->abc->K);
else x = eslDSQ_SENTINEL;
if (x != eslDSQ_SENTINEL) i++;
/* Add residue (if any) to sequence */
if (sq != NULL && x != eslDSQ_SENTINEL && (status = esl_sq_XAddResidue(sq, x)) != eslOK) goto ERROR;
/* Add state to trace. */
if (tr != NULL) {
if ((status = p7_trace_Append(tr, st, k, i)) != eslOK) goto ERROR;
}
}
/* Terminate the trace and sequence (both are optional, remember) */
if (tr != NULL) { tr->M = hmm->M; tr->L = i; }
if (sq != NULL && (status = esl_sq_XAddResidue(sq, eslDSQ_SENTINEL)) != eslOK) goto ERROR;
return eslOK;
ERROR:
return status;
}
/* Function: p7_alidisplay_Create()
* Synopsis: Create an alignment display, from trace and oprofile.
* Incept: SRE, Sun Dec 30 09:13:31 2007 [Janelia]
*
* Purpose: Creates and returns an alignment display for domain number
* <which> in traceback <tr>, where the traceback
* corresponds to an alignment of optimized profile <om> to digital sequence
* <dsq>, and the unique name of that target
* sequence <dsq> is <sqname>. The <which> index starts at 0.
*
* It will be a little faster if the trace is indexed with
* <p7_trace_Index()> first. The number of domains is then
* in <tr->ndom>. If the caller wants to create alidisplays
* for all of these, it would loop <which> from
* <0..tr->ndom-1>.
*
* However, even without an index, the routine will work fine.
*
* Args: tr - traceback
* which - domain number, 0..tr->ndom-1
* om - optimized profile (query)
* sq - digital sequence (target)
*
* Returns: <eslOK> on success.
*
* Throws: <NULL> on allocation failure, or if something's internally corrupt
* in the data.
*/
P7_ALIDISPLAY *
p7_alidisplay_Create(const P7_TRACE *tr, int which, const P7_OPROFILE *om, const ESL_SQ *sq)
{
P7_ALIDISPLAY *ad = NULL;
char *Alphabet = om->abc->sym;
int n, pos, z;
int z1,z2;
int k,x,i,s;
int hmm_namelen, hmm_acclen, hmm_desclen;
int sq_namelen, sq_acclen, sq_desclen;
int status;
/* First figure out which piece of the trace (from first match to last match)
* we're going to represent, and how big it is.
*/
if (tr->ndom > 0) { /* if we have an index, this is a little faster: */
for (z1 = tr->tfrom[which]; z1 < tr->N; z1++) if (tr->st[z1] == p7T_M) break; /* find next M state */
if (z1 == tr->N) return NULL; /* no M? corrupt trace */
for (z2 = tr->tto[which]; z2 >= 0 ; z2--) if (tr->st[z2] == p7T_M) break; /* find prev M state */
if (z2 == -1) return NULL; /* no M? corrupt trace */
} else { /* without an index, we can still do it fine: */
for (z1 = 0; which >= 0 && z1 < tr->N; z1++) if (tr->st[z1] == p7T_B) which--; /* find the right B state */
if (z1 == tr->N) return NULL; /* no such domain <which> */
for (; z1 < tr->N; z1++) if (tr->st[z1] == p7T_M) break; /* find next M state */
if (z1 == tr->N) return NULL; /* no M? corrupt trace */
for (z2 = z1; z2 < tr->N; z2++) if (tr->st[z2] == p7T_E) break; /* find the next E state */
for (; z2 >= 0; z2--) if (tr->st[z2] == p7T_M) break; /* find prev M state */
if (z2 == -1) return NULL; /* no M? corrupt trace */
}
/* Now we know that z1..z2 in the trace will be represented in the
* alidisplay; that's z2-z1+1 positions. We need a \0 trailer on all
* our display lines, so allocate z2-z1+2. We know each position is
* M, D, or I, so there's a 1:1 correspondence of trace positions
* with alignment display positions. We also know the display
* starts and ends with M states.
*
* So now let's allocate. The alidisplay is packed into a single
* memory space, so this appears to be intricate, but it's just
* bookkeeping.
*/
n = (z2-z1+2) * 3; /* model, mline, aseq mandatory */
if (om->rf[0] != 0) n += z2-z1+2; /* optional reference line */
if (om->cs[0] != 0) n += z2-z1+2; /* optional structure line */
if (tr->pp != NULL) n += z2-z1+2; /* optional posterior prob line */
hmm_namelen = strlen(om->name); n += hmm_namelen + 1;
hmm_acclen = (om->acc != NULL ? strlen(om->acc) : 0); n += hmm_acclen + 1;
hmm_desclen = (om->desc != NULL ? strlen(om->desc) : 0); n += hmm_desclen + 1;
sq_namelen = strlen(sq->name); n += sq_namelen + 1;
sq_acclen = strlen(sq->acc); n += sq_acclen + 1; /* sq->acc is "\0" when unset */
sq_desclen = strlen(sq->desc); n += sq_desclen + 1; /* same for desc */
ESL_ALLOC(ad, sizeof(P7_ALIDISPLAY));
ad->mem = NULL;
pos = 0;
ad->memsize = sizeof(char) * n;
ESL_ALLOC(ad->mem, ad->memsize);
if (om->rf[0] != 0) { ad->rfline = ad->mem + pos; pos += z2-z1+2; } else { ad->rfline = NULL; }
if (om->cs[0] != 0) { ad->csline = ad->mem + pos; pos += z2-z1+2; } else { ad->csline = NULL; }
ad->model = ad->mem + pos; pos += z2-z1+2;
ad->mline = ad->mem + pos; pos += z2-z1+2;
ad->aseq = ad->mem + pos; pos += z2-z1+2;
if (tr->pp != NULL) { ad->ppline = ad->mem + pos; pos += z2-z1+2;} else { ad->ppline = NULL; }
ad->hmmname = ad->mem + pos; pos += hmm_namelen +1;
ad->hmmacc = ad->mem + pos; pos += hmm_acclen +1;
ad->hmmdesc = ad->mem + pos; pos += hmm_desclen +1;
ad->sqname = ad->mem + pos; pos += sq_namelen +1;
ad->sqacc = ad->mem + pos; pos += sq_acclen +1;
ad->sqdesc = ad->mem + pos; pos += sq_desclen +1;
strcpy(ad->hmmname, om->name);
if (om->acc != NULL) strcpy(ad->hmmacc, om->acc); else ad->hmmacc[0] = 0;
if (om->desc != NULL) strcpy(ad->hmmdesc, om->desc); else ad->hmmdesc[0] = 0;
strcpy(ad->sqname, sq->name);
strcpy(ad->sqacc, sq->acc);
strcpy(ad->sqdesc, sq->desc);
/* Determine hit coords */
ad->hmmfrom = tr->k[z1];
ad->hmmto = tr->k[z2];
ad->M = om->M;
ad->sqfrom = tr->i[z1];
ad->sqto = tr->i[z2];
ad->L = sq->n;
/* optional rf line */
if (ad->rfline != NULL) {
for (z = z1; z <= z2; z++) ad->rfline[z-z1] = ((tr->st[z] == p7T_I) ? '.' : om->rf[tr->k[z]]);
ad->rfline[z-z1] = '\0';
}
/* optional cs line */
if (ad->csline != NULL) {
for (z = z1; z <= z2; z++) ad->csline[z-z1] = ((tr->st[z] == p7T_I) ? '.' : om->cs[tr->k[z]]);
ad->csline[z-z1] = '\0';
}
/* optional pp line */
if (ad->ppline != NULL) {
for (z = z1; z <= z2; z++) ad->ppline[z-z1] = ( (tr->st[z] == p7T_D) ? '.' : p7_alidisplay_EncodePostProb(tr->pp[z]));
ad->ppline[z-z1] = '\0';
}
/* mandatory three alignment display lines: model, mline, aseq */
for (z = z1; z <= z2; z++)
{
k = tr->k[z];
i = tr->i[z];
x = sq->dsq[i];
s = tr->st[z];
switch (s) {
case p7T_M:
ad->model[z-z1] = om->consensus[k];
if (x == esl_abc_DigitizeSymbol(om->abc, om->consensus[k])) ad->mline[z-z1] = ad->model[z-z1];
else if (p7_oprofile_FGetEmission(om, k, x) > 1.0) ad->mline[z-z1] = '+'; /* >1 not >0; om has odds ratios, not scores */
else ad->mline[z-z1] = ' ';
ad->aseq [z-z1] = toupper(Alphabet[x]);
break;
case p7T_I:
ad->model [z-z1] = '.';
ad->mline [z-z1] = ' ';
ad->aseq [z-z1] = tolower(Alphabet[x]);
break;
case p7T_D:
ad->model [z-z1] = om->consensus[k];
ad->mline [z-z1] = ' ';
ad->aseq [z-z1] = '-';
break;
default: ESL_XEXCEPTION(eslEINVAL, "invalid state in trace: not M,D,I");
}
}
ad->model [z2-z1+1] = '\0';
ad->mline [z2-z1+1] = '\0';
ad->aseq [z2-z1+1] = '\0';
ad->N = z2-z1+1;
return ad;
ERROR:
p7_alidisplay_Destroy(ad);
return NULL;
}
/* Function: p7_alidisplay_Backconvert()
* Synopsis: Convert an alidisplay to a faux trace and subsequence.
* Incept: SRE, Wed Dec 10 09:49:28 2008 [Janelia]
*
* Purpose: Convert alignment display object <ad> to a faux subsequence
* and faux subsequence trace, returning them in <ret_sq> and
* <ret_tr>.
*
* The subsequence <*ret_sq> is digital; ascii residues in
* <ad> are digitized using digital alphabet <abc>.
*
* The subsequence and trace are suitable for passing as
* array elements to <p7_MultipleAlignment>. This is the
* main purpose of backconversion. Results of a profile
* search are stored in a hit list as a processed
* <P7_ALIDISPLAY>, not as a <P7_TRACE> and <ESL_SQ>, to
* reduce space and to reduce communication overhead in
* parallelized search implementations. After reduction
* to a final hit list, a master may want to construct a
* multiple alignment of all the significant hits.
*
* Returns: <eslOK> on success.
*
* Throws: <eslEMEM> on allocation failures. <eslECORRUPT> on unexpected internal
* data corruption. On any exception, <*ret_sq> and <*ret_tr> are
* <NULL>.
*
* Xref: J4/29.
*/
int
p7_alidisplay_Backconvert(const P7_ALIDISPLAY *ad, const ESL_ALPHABET *abc, ESL_SQ **ret_sq, P7_TRACE **ret_tr)
{
ESL_SQ *sq = NULL; /* RETURN: faux subsequence */
P7_TRACE *tr = NULL; /* RETURN: faux trace */
int subL = 0; /* subsequence length in the <ad> */
int a, i, k; /* coords for <ad>, <sq->dsq>, model */
char st; /* state type: MDI */
int status;
/* Make a first pass over <ad> just to calculate subseq length */
for (a = 0; a < ad->N; a++)
if (! esl_abc_CIsGap(abc, ad->aseq[a])) subL++;
/* Allocations */
if ((sq = esl_sq_CreateDigital(abc)) == NULL) { status = eslEMEM; goto ERROR; }
if ((status = esl_sq_GrowTo(sq, subL)) != eslOK) goto ERROR;
if ((tr = (ad->ppline == NULL) ? p7_trace_Create() : p7_trace_CreateWithPP()) == NULL) { status = eslEMEM; goto ERROR; }
if ((status = p7_trace_GrowTo(tr, subL+6)) != eslOK) goto ERROR; /* +6 is for SNB/ECT */
/* Construction of dsq, trace */
sq->dsq[0] = eslDSQ_SENTINEL;
if ((status = ((ad->ppline == NULL) ? p7_trace_Append(tr, p7T_S, 0, 0) : p7_trace_AppendWithPP(tr, p7T_S, 0, 0, 0.0))) != eslOK) goto ERROR;
if ((status = ((ad->ppline == NULL) ? p7_trace_Append(tr, p7T_N, 0, 0) : p7_trace_AppendWithPP(tr, p7T_N, 0, 0, 0.0))) != eslOK) goto ERROR;
if ((status = ((ad->ppline == NULL) ? p7_trace_Append(tr, p7T_B, 0, 0) : p7_trace_AppendWithPP(tr, p7T_B, 0, 0, 0.0))) != eslOK) goto ERROR;
k = ad->hmmfrom;
i = 1;
for (a = 0; a < ad->N; a++)
{
if (esl_abc_CIsResidue(abc, ad->model[a])) { st = (esl_abc_CIsResidue(abc, ad->aseq[a]) ? p7T_M : p7T_D); } else st = p7T_I;
if ((status = ((ad->ppline == NULL) ? p7_trace_Append(tr, st, k, i) : p7_trace_AppendWithPP(tr, st, k, i, p7_alidisplay_DecodePostProb(ad->ppline[a])))) != eslOK) goto ERROR;
switch (st) {
case p7T_M: sq->dsq[i] = esl_abc_DigitizeSymbol(abc, ad->aseq[a]); k++; i++; break;
case p7T_I: sq->dsq[i] = esl_abc_DigitizeSymbol(abc, ad->aseq[a]); i++; break;
case p7T_D: k++; break;
}
}
if ((status = ((ad->ppline == NULL) ? p7_trace_Append(tr, p7T_E, 0, 0) : p7_trace_AppendWithPP(tr, p7T_E, 0, 0, 0.0))) != eslOK) goto ERROR;
if ((status = ((ad->ppline == NULL) ? p7_trace_Append(tr, p7T_C, 0, 0) : p7_trace_AppendWithPP(tr, p7T_C, 0, 0, 0.0))) != eslOK) goto ERROR;
if ((status = ((ad->ppline == NULL) ? p7_trace_Append(tr, p7T_T, 0, 0) : p7_trace_AppendWithPP(tr, p7T_T, 0, 0, 0.0))) != eslOK) goto ERROR;
sq->dsq[i] = eslDSQ_SENTINEL;
/* some sanity checks */
if (tr->N != ad->N + 6) ESL_XEXCEPTION(eslECORRUPT, "backconverted trace ended up with unexpected size (%s/%s)", ad->sqname, ad->hmmname);
if (k != ad->hmmto + 1) ESL_XEXCEPTION(eslECORRUPT, "backconverted trace didn't end at expected place on model (%s/%s)", ad->sqname, ad->hmmname);
if (i != subL + 1) ESL_XEXCEPTION(eslECORRUPT, "backconverted subseq didn't end at expected length (%s/%s)", ad->sqname, ad->hmmname);
/* Set up <sq> annotation as a subseq of a source sequence */
if ((status = esl_sq_FormatName(sq, "%s/%ld-%ld", ad->sqname, ad->sqfrom, ad->sqto)) != eslOK) goto ERROR;
if ((status = esl_sq_FormatDesc(sq, "[subseq from] %s", ad->sqdesc[0] != '\0' ? ad->sqdesc : ad->sqname)) != eslOK) goto ERROR;
if ((status = esl_sq_SetSource (sq, ad->sqname)) != eslOK) goto ERROR;
if (ad->sqacc[0] != '\0') { if ((status = esl_sq_SetAccession (sq, ad->sqacc)) != eslOK) goto ERROR; }
sq->n = subL;
sq->start = ad->sqfrom;
sq->end = ad->sqto;
sq->C = 0;
sq->W = subL;
sq->L = ad->L;
tr->M = ad->M;
tr->L = ad->L;
*ret_sq = sq;
*ret_tr = tr;
return eslOK;
ERROR:
if (sq != NULL) esl_sq_Destroy(sq);
if (tr != NULL) p7_trace_Destroy(tr);
*ret_sq = NULL;
*ret_tr = NULL;
return status;
}
/* Function: p7_GStochasticTrace()
* Synopsis: Stochastic traceback of a Forward matrix.
* Incept: SRE, Thu Jan 3 15:39:20 2008 [Janelia]
*
* Purpose: Stochastic traceback of Forward matrix <gx> to
* sample an alignment of digital sequence <dsq>
* (of length <L>) to the profile <gm>.
*
* The sampled traceback is returned in <tr>, which the
* caller must have at least made an initial allocation of
* (the <tr> will be grown as needed here).
*
* Args: r - source of random numbers
* dsq - digital sequence aligned to, 1..L
* L - length of dsq
* gm - profile
* mx - Forward matrix to trace, L x M
* tr - storage for the recovered traceback.
*
* Returns: <eslOK> on success.
*/
int
p7_GStochasticTrace(ESL_RANDOMNESS *r, const ESL_DSQ *dsq, int L, const P7_PROFILE *gm, const P7_GMX *gx, P7_TRACE *tr)
{
int status;
int i; /* position in seq (1..L) */
int k; /* position in model (1..M) */
int M = gm->M;
float **dp = gx->dp;
float *xmx = gx->xmx;
float const *tsc = gm->tsc;
float *sc; /* scores of possible choices: up to 2M-1, in the case of exits to E */
int scur, sprv;
/* we'll index M states as 1..M, and D states as 2..M = M+2..2M: M0, D1 are impossibles. */
ESL_ALLOC(sc, sizeof(float) * (2*M+1));
k = 0;
i = L;
if ((status = p7_trace_Append(tr, p7T_T, k, i)) != eslOK) goto ERROR;
if ((status = p7_trace_Append(tr, p7T_C, k, i)) != eslOK) goto ERROR;
sprv = p7T_C;
while (sprv != p7T_S)
{
switch (tr->st[tr->N-1]) {
/* C(i) comes from C(i-1) or E(i) */
case p7T_C:
if (XMX(i,p7G_C) == -eslINFINITY) ESL_XEXCEPTION(eslFAIL, "impossible C reached at i=%d", i);
sc[0] = XMX(i-1, p7G_C) + gm->xsc[p7P_C][p7P_LOOP];
sc[1] = XMX(i, p7G_E) + gm->xsc[p7P_E][p7P_MOVE];
esl_vec_FLogNorm(sc, 2);
scur = (esl_rnd_FChoose(r, sc, 2) == 0) ? p7T_C : p7T_E;
break;
/* E connects from any M or D state. k set here */
case p7T_E:
if (XMX(i, p7G_E) == -eslINFINITY) ESL_XEXCEPTION(eslFAIL, "impossible E reached at i=%d", i);
if (p7_profile_IsLocal(gm)) { /* local models come from any M, D */
sc[0] = sc[M+1] = -eslINFINITY;
for (k = 1; k <= M; k++) sc[k] = MMX(i,k);
for (k = 2; k <= M; k++) sc[k+M] = DMX(i,k);
esl_vec_FLogNorm(sc, 2*M+1); /* now sc is a prob vector */
k = esl_rnd_FChoose(r, sc, 2*M+1);
if (k <= M) scur = p7T_M;
else { k -= M; scur = p7T_D; }
} else { /* glocal models come from M_M or D_M */
k = M;
sc[0] = MMX(i,M);
sc[1] = DMX(i,M);
esl_vec_FLogNorm(sc, 2); /* now sc is a prob vector */
scur = (esl_rnd_FChoose(r, sc, 2) == 0) ? p7T_M : p7T_D;
}
break;
/* M connects from {MDI} i-1,k-1, or B */
case p7T_M:
if (MMX(i,k) == -eslINFINITY) ESL_XEXCEPTION(eslFAIL, "impossible M reached at k=%d,i=%d", k,i);
sc[0] = XMX(i-1,p7G_B) + TSC(p7P_BM, k-1);
sc[1] = MMX(i-1,k-1) + TSC(p7P_MM, k-1);
sc[2] = IMX(i-1,k-1) + TSC(p7P_IM, k-1);
sc[3] = DMX(i-1,k-1) + TSC(p7P_DM, k-1);
esl_vec_FLogNorm(sc, 4);
switch (esl_rnd_FChoose(r, sc, 4)) {
case 0: scur = p7T_B; break;
case 1: scur = p7T_M; break;
case 2: scur = p7T_I; break;
case 3: scur = p7T_D; break;
}
k--;
i--;
break;
/* D connects from M,D at i,k-1 */
case p7T_D:
if (DMX(i, k) == -eslINFINITY) ESL_XEXCEPTION(eslFAIL, "impossible D reached at k=%d,i=%d", k,i);
sc[0] = MMX(i, k-1) + TSC(p7P_MD, k-1);
sc[1] = DMX(i, k-1) + TSC(p7P_DD, k-1);
esl_vec_FLogNorm(sc, 2);
scur = (esl_rnd_FChoose(r, sc, 2) == 0) ? p7T_M : p7T_D;
k--;
break;
/* I connects from M,I at i-1,k */
case p7T_I:
if (IMX(i,k) == -eslINFINITY) ESL_XEXCEPTION(eslFAIL, "impossible I reached at k=%d,i=%d", k,i);
sc[0] = MMX(i-1,k) + TSC(p7P_MI, k);
sc[1] = IMX(i-1,k) + TSC(p7P_II, k);
esl_vec_FLogNorm(sc, 2);
scur = (esl_rnd_FChoose(r, sc, 2) == 0) ? p7T_M : p7T_I;
i--;
break;
/* N connects from S, N */
case p7T_N:
if (XMX(i, p7G_N) == -eslINFINITY) ESL_XEXCEPTION(eslFAIL, "impossible N reached at i=%d", i);
scur = (i == 0) ? p7T_S : p7T_N;
break;
/* B connects from N, J */
case p7T_B:
if (XMX(i,p7G_B) == -eslINFINITY) ESL_XEXCEPTION(eslFAIL, "impossible B reached at i=%d", i);
sc[0] = XMX(i, p7G_N) + gm->xsc[p7P_N][p7P_MOVE];
sc[1] = XMX(i, p7G_J) + gm->xsc[p7P_J][p7P_MOVE];
esl_vec_FLogNorm(sc, 2);
scur = (esl_rnd_FChoose(r, sc, 2) == 0) ? p7T_N : p7T_J;
break;
/* J connects from E(i) or J(i-1) */
case p7T_J:
if (XMX(i,p7G_J) == -eslINFINITY) ESL_XEXCEPTION(eslFAIL, "impossible J reached at i=%d", i);
sc[0] = XMX(i-1,p7G_J) + gm->xsc[p7P_J][p7P_LOOP];
sc[1] = XMX(i, p7G_E) + gm->xsc[p7P_E][p7P_LOOP];
esl_vec_FLogNorm(sc, 2);
scur = (esl_rnd_FChoose(r, sc, 2) == 0) ? p7T_J : p7T_E;
break;
default: ESL_XEXCEPTION(eslFAIL, "bogus state in traceback");
} /* end switch over statetype[tpos-1] */
/* Append this state and the current i,k to be explained to the growing trace */
if ((status = p7_trace_Append(tr, scur, k, i)) != eslOK) goto ERROR;
/* For NCJ, we had to defer i decrement. */
if ( (scur == p7T_N || scur == p7T_J || scur == p7T_C) && scur == sprv) i--;
sprv = scur;
} /* end traceback, at S state */
if ((status = p7_trace_Reverse(tr)) != eslOK) goto ERROR;
tr->M = gm->M;
tr->L = L;
free(sc);
return eslOK;
ERROR:
if (sc != NULL) free(sc);
return status;
}
/* Function: p7_oprofile_MPIPackSize()
* Synopsis: Calculates size needed to pack an OPROFILE.
* Incept: MSF, Wed Oct 21, 2009 [Janelia]
*
* Purpose: Calculate an upper bound on the number of bytes
* that <p7_oprofile_MPIPack()> will need to pack an
* OPROFILE <om> in a packed MPI message for MPI
* communicator <comm>; return that number of bytes
* in <*ret_n>.
*
* Returns: <eslOK> on success, and <*ret_n> contains the answer.
*
* Throws: <eslESYS> if an MPI call fails, and <*ret_n> is 0.
*/
int
p7_oprofile_MPIPackSize(P7_OPROFILE *om, MPI_Comm comm, int *ret_n)
{
int status;
int n = 0;
int K = om->abc->Kp;
int len = 0;
int cnt;
int sz;
int Q4 = p7O_NQF(om->M);
int Q8 = p7O_NQW(om->M);
int Q16 = p7O_NQB(om->M);
int vsz = sizeof(vector float);
/* MSV Filter information */
if (MPI_Pack_size(5, MPI_CHAR, comm, &sz) != 0) ESL_XEXCEPTION(eslESYS, "pack size failed"); n += sz;
if (MPI_Pack_size(1, MPI_FLOAT, comm, &sz) != 0) ESL_XEXCEPTION(eslESYS, "pack size failed"); n += sz;
if (MPI_Pack_size(vsz*Q16, MPI_CHAR, comm, &sz) != 0) ESL_XEXCEPTION(eslESYS, "pack size failed"); n += (K*sz);
/* Viterbi Filter information */
if (MPI_Pack_size(1, MPI_SHORT, comm, &sz) != 0) ESL_XEXCEPTION(eslESYS, "pack size failed"); n += ((p7O_NXSTATES*p7O_NXTRANS+2)*sz);
if (MPI_Pack_size(2, MPI_FLOAT, comm, &sz) != 0) ESL_XEXCEPTION(eslESYS, "pack size failed"); n += sz;
if (MPI_Pack_size(K*vsz*Q8, MPI_CHAR, comm, &sz) != 0) ESL_XEXCEPTION(eslESYS, "pack size failed"); n += sz;
if (MPI_Pack_size(8*vsz*Q8, MPI_CHAR, comm, &sz) != 0) ESL_XEXCEPTION(eslESYS, "pack size failed"); n += sz;
/* Forward/Backward information */
if (MPI_Pack_size(1, MPI_FLOAT, comm, &sz) != 0) ESL_XEXCEPTION(eslESYS, "pack size failed"); n += (p7O_NXSTATES*p7O_NXTRANS*sz);
if (MPI_Pack_size(K*vsz*Q4, MPI_CHAR, comm, &sz) != 0) ESL_XEXCEPTION(eslESYS, "pack size failed"); n += sz;
if (MPI_Pack_size(8*vsz*Q4, MPI_CHAR, comm, &sz) != 0) ESL_XEXCEPTION(eslESYS, "pack size failed"); n += sz;
/* disk offsets */
if (MPI_Pack_size(1, MPI_LONG_LONG_INT, comm, &sz) != 0) ESL_XEXCEPTION(eslESYS, "pack size failed"); n += ((p7_NOFFSETS+2)*sz);
/* annotation info */
if (om->name != NULL) len += strlen(om->name) + 1;
if (om->acc != NULL) len += strlen(om->acc) + 1;
if (om->desc != NULL) len += strlen(om->desc) + 1;
if (om->rf != NULL) len += strlen(om->rf) + 1;
if (om->cs != NULL) len += strlen(om->cs) + 1;
if (om->consensus != NULL) len += strlen(om->consensus) + 1;
if (MPI_Pack_size(6, MPI_INT, comm, &sz) != 0) ESL_XEXCEPTION(eslESYS, "pack size failed"); n += sz;
if (MPI_Pack_size(len, MPI_CHAR, comm, &sz) != 0) ESL_XEXCEPTION(eslESYS, "pack size failed"); n += sz;
cnt = p7_NEVPARAM + p7_NCUTOFFS + p7_MAXABET;
if (MPI_Pack_size(cnt, MPI_FLOAT, comm, &sz) != 0) ESL_XEXCEPTION(eslESYS, "pack size failed"); n += sz;
/* current model size */
if (MPI_Pack_size(4, MPI_INT, comm, &sz) != 0) ESL_XEXCEPTION(eslESYS, "pack size failed"); n += sz;
if (MPI_Pack_size(1, MPI_FLOAT, comm, &sz) != 0) ESL_XEXCEPTION(eslESYS, "pack size failed"); n += sz;
*ret_n = n;
return eslOK;
ERROR:
*ret_n = 0;
return status;
}
/* Function: p7_oprofile_MPIUnpack()
* Synopsis: Unpacks an OPROFILE from an MPI buffer.
* Incept: MSF, Wed Oct 21, 2009 [Janelia]
*
* Purpose: Unpack a newly allocated OPROFILE from MPI packed buffer
* <buf>, starting from position <*pos>, where the total length
* of the buffer in bytes is <n>.
*
* Caller may or may not already know what alphabet the OPROFILE
* is expected to be in. A reference to the current
* alphabet is passed in <abc>. If the alphabet is unknown,
* pass <*abc = NULL>, and when the OPROFILE is received, an
* appropriate new alphabet object is allocated and passed
* back to the caller via <*abc>. If the alphabet is
* already known, <*abc> is that alphabet, and the new
* OPROFILE's alphabet type is verified to agree with it. This
* mechanism allows an application to let the first OPROFILE
* determine the alphabet type for the application, while
* still keeping the alphabet under the application's scope
* of control.
*
* Returns: <eslOK> on success. <*pos> is updated to the position of
* the next element in <buf> to unpack (if any). <*ret_om>
* contains a newly allocated OPROFILE, which the caller is
* responsible for free'ing. If <*abc> was passed as
* <NULL>, it now points to an <ESL_ALPHABET> object that
* was allocated here; caller is responsible for free'ing
* this.
*
* Returns <eslEINCOMPAT> if the OPROFILE is in a different
* alphabet than <*abc> said to expect. In this case,
* <*abc> is unchanged, <*buf> and <*nalloc> may have been
* changed, and <*ret_om> is <NULL>.
*
* Throws: <eslESYS> on an MPI call failure. <eslEMEM> on allocation failure.
* In either case, <*ret_om> is <NULL>, and the state of <buf>
* and <*pos> is undefined and should be considered to be corrupted.
*/
int
p7_oprofile_MPIUnpack(char *buf, int n, int *pos, MPI_Comm comm, ESL_ALPHABET **abc, P7_OPROFILE **ret_om)
{
int status;
int M, K, atype;
int len;
int x;
int Q4, Q8, Q16;
int vsz = sizeof(vector float);
P7_OPROFILE *om = NULL;
if (MPI_Unpack(buf, n, pos, &M, 1, MPI_INT, comm) != 0) ESL_XEXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack(buf, n, pos, &atype, 1, MPI_INT, comm) != 0) ESL_XEXCEPTION(eslESYS, "mpi unpack failed");
/* Set or verify the alphabet */
if (*abc == NULL) { /* still unknown: set it, pass control of it back to caller */
if ((*abc = esl_alphabet_Create(atype)) == NULL) { status = eslEMEM; goto ERROR; }
} else { /* already known: check it */
if ((*abc)->type != atype) { status = eslEINCOMPAT; goto ERROR; }
}
Q4 = p7O_NQF(M);
Q8 = p7O_NQW(M);
Q16 = p7O_NQB(M);
if ((om = p7_oprofile_Create(M, *abc)) == NULL) { status = eslEMEM; goto ERROR; }
om->M = M;
K = (*abc)->Kp;
/* model configuration */
if (MPI_Unpack(buf, n, pos, &om->L, 1, MPI_INT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack(buf, n, pos, &om->mode, 1, MPI_INT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack(buf, n, pos, &om->nj, 1, MPI_FLOAT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
/* MSV Filter information */
if (MPI_Unpack(buf, n, pos, &om->tbm_b, 1, MPI_CHAR, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack(buf, n, pos, &om->tec_b, 1, MPI_CHAR, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack(buf, n, pos, &om->tjb_b, 1, MPI_CHAR, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack(buf, n, pos, &om->scale_b, 1, MPI_FLOAT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack(buf, n, pos, &om->base_b, 1, MPI_CHAR, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack(buf, n, pos, &om->bias_b, 1, MPI_CHAR, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
for (x = 0; x < K; x++)
if (MPI_Unpack(buf, n, pos, om->rbv[x], vsz*Q16, MPI_CHAR, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
/* Viterbi Filter information */
if (MPI_Unpack(buf, n, pos, &om->scale_w, 1, MPI_FLOAT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack(buf, n, pos, &om->base_w, 1, MPI_SHORT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack(buf, n, pos, &om->ddbound_w, 1, MPI_SHORT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack(buf, n, pos, &om->ncj_roundoff, 1, MPI_FLOAT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack(buf, n, pos, om->twv, 8*vsz*Q8, MPI_CHAR, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
for (x = 0; x < p7O_NXSTATES; x++)
if (MPI_Unpack(buf, n, pos, om->xw[x], p7O_NXTRANS, MPI_SHORT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
for (x = 0; x < K; x++)
if (MPI_Unpack(buf, n, pos, om->rwv[x], vsz*Q8, MPI_CHAR, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
/* Forward/Backward information */
if (MPI_Unpack(buf, n, pos, om->tfv, 8*vsz*Q4, MPI_CHAR, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
for (x = 0; x < p7O_NXSTATES; x++)
if (MPI_Unpack(buf, n, pos, om->xf[x], p7O_NXTRANS, MPI_FLOAT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
for (x = 0; x < K; x++)
if (MPI_Unpack(buf, n, pos, om->rfv[x], vsz*Q4, MPI_CHAR, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
/* Forward/Backward information */
if (MPI_Unpack(buf, n, pos, om->offs, p7_NOFFSETS, MPI_LONG_LONG_INT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack(buf, n, pos, &om->roff, 1, MPI_LONG_LONG_INT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack(buf, n, pos, &om->eoff, 1, MPI_LONG_LONG_INT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
/* Annotation information */
if (MPI_Unpack(buf, n, pos, &len, 1, MPI_INT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (len > 0) {
ESL_ALLOC(om->name, len);
if (MPI_Unpack(buf, n, pos, om->name, len, MPI_CHAR, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
om->name[len-1] = '\0';
}
if (MPI_Unpack(buf, n, pos, &len, 1, MPI_INT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (len > 0) {
ESL_ALLOC(om->acc, len);
if (MPI_Unpack(buf, n, pos, om->acc, len, MPI_CHAR, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
om->acc[len-1] = '\0';
}
if (MPI_Unpack(buf, n, pos, &len, 1, MPI_INT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (len > 0) {
ESL_ALLOC(om->desc, len);
if (MPI_Unpack(buf, n, pos, om->desc, len, MPI_CHAR, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
om->desc[len-1] = '\0';
}
if (MPI_Unpack(buf, n, pos, &len, 1, MPI_INT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (len > 0) {
ESL_ALLOC(om->rf, len);
if (MPI_Unpack(buf, n, pos, om->rf, len, MPI_CHAR, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
om->rf[len-1] = '\0';
}
if (MPI_Unpack(buf, n, pos, &len, 1, MPI_INT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (len > 0) {
ESL_ALLOC(om->cs, len);
if (MPI_Unpack(buf, n, pos, om->cs, len, MPI_CHAR, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
om->cs[len-1] = '\0';
}
if (MPI_Unpack(buf, n, pos, &len, 1, MPI_INT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (len > 0) {
ESL_ALLOC(om->consensus, len);
if (MPI_Unpack(buf, n, pos, om->consensus, len, MPI_CHAR, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
om->consensus[len-1] = '\0';
}
if (MPI_Unpack(buf, n, pos, om->evparam, p7_NEVPARAM, MPI_FLOAT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack(buf, n, pos, om->cutoff, p7_NCUTOFFS, MPI_FLOAT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (MPI_Unpack(buf, n, pos, om->compo, p7_MAXABET, MPI_FLOAT, comm) != 0) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
*ret_om = om;
return eslOK;
ERROR:
if (om != NULL) p7_oprofile_Destroy(om);
return status;
}
