/* Function: esl_hxp_FitGuess()
*
* Purpose: Given a sorted vector of <n> observed data samples <x[]>,
* from smallest <x[0]> to largest <x[n-1]>, calculate a
* very crude guesstimate of a fit -- suitable only as a starting
* point for further optimization -- and return those parameters
* in <h>.
*
* Assigns $q_k \propto \frac{1}{k}$ and $\mu = \min_i x_i$;
* splits $x$ into $K$ roughly equal-sized bins, and
* and assigns $\lambda_k$ as the ML estimate from bin $k$.
* (If $q_k$ coefficients have already been fixed to
* known values, this step is skipped.)
*/
int
esl_hxp_FitGuess(double *x, int n, ESL_HYPEREXP *h)
{
double tmu; /* current mu */
double mean; /* mean (x-tmu) in a bin */
int i,k;
int imin, imax;
h->mu = x[0]; /* minimum */
for (k = 0; k < h->K; k++)
{
if (! h->fixmix)
h->q[k] = 1 / (double)(k+1); /* priors ~ 1, 1/2, 1/3... */
imin = (int) ((double)(k*n)/(double)h->K);
imax = (int) ((double)((k+1)*n)/(double)h->K);
tmu = x[imin];
mean = 0.;
for (i = imin; i < imax; i++)
mean += x[i] - tmu;
mean /= (double)(imax-imin);
h->lambda[k] = 1 / mean;
}
esl_vec_DNorm(h->q, h->K);
return eslOK;
}
/* Function: esl_vec_DLogNorm()
* Synopsis: Normalize a log p-vector, make it a p-vector.
* Incept: SRE, Thu Apr 7 17:45:39 2005 [St. Louis]
*
* Purpose: Given an unnormalized log probability vector <vec>
* of length <n>, normalize it and make it a
* probability vector.
*
* <esl_vec_FLogNorm()> does the same, but for a vector
* of floats instead of doubles.
*
* Returns: (void); <vec> is changed in place.
*/
void
esl_vec_DLogNorm(double *vec, int n)
{
double denom;
denom = esl_vec_DLogSum(vec, n);
esl_vec_DIncrement(vec, n, -1.*denom);
esl_vec_DExp (vec, n);
esl_vec_DNorm(vec, n);
}
/* Function: esl_hyperexp_Read()
*
* Purpose: Reads hyperexponential parameters from an open <e>.
* which is an <ESL_FILEPARSER> tokenizer for an open stream.
*
* The first token is <K>, the number of mixture components.
* The second token is <mu>, the x offset shared by all components.
* Then for each mixture component <k=1..K>, it reads
* a mixture coefficient <q[k]> and a decay parameter
* <lambda[k]>.
*
* The <2K+2> data tokens must occur in this order, but
* they can be grouped into any number of lines, because the
* parser ignores line breaks.
*
* Anything after a <\#> character on a line is a comment, and
* is ignored.
*
* Returns: <eslOK> on success, and <ret_hxp> points to a new <ESL_HYPEREXP>
* object.
* <eslEFORMAT> on "normal" parse failure caused by a bad file
* format that's likely the user's fault.
*
* Throws: <eslEMEM> if allocation of the new <ESL_HYPEREXP> fails.
*
*
* FIXME: All our mixture models (esl_dirichlet, for example) should be
* reconciled w/ identical interfaces & behaviour.
*/
int
esl_hyperexp_Read(ESL_FILEPARSER *e, ESL_HYPEREXP **ret_hxp)
{
ESL_HYPEREXP *hxp = NULL;
char *tok;
int status = eslOK;
int nc;
int k;
double sum;
esl_fileparser_SetCommentChar(e, '#');
if ((status = esl_fileparser_GetToken(e, &tok, NULL)) != eslOK) goto ERROR;
nc = atoi(tok);
if (nc < 1) {
sprintf(e->errbuf, "Expected # of components K >= 1 as first token");
goto ERROR;
}
if ((hxp = esl_hyperexp_Create(nc)) == NULL) return eslEMEM; /* percolation */
if ((status = esl_fileparser_GetToken(e, &tok, NULL)) != eslOK) goto ERROR;
hxp->mu = atof(tok);
for (k = 0; k < hxp->K; k++)
{
if ((status = esl_fileparser_GetToken(e, &tok, NULL)) != eslOK) goto ERROR;
hxp->q[k] = atof(tok);
if ((status = esl_fileparser_GetToken(e, &tok, NULL)) != eslOK) goto ERROR;
hxp->lambda[k] = atof(tok);
if (hxp->q[k] < 0. || hxp->q[k] > 1.) {
sprintf(e->errbuf, "Expected a mixture coefficient q[k], 0<=q[k]<=1");
goto ERROR;
}
if (hxp->lambda[k] <= 0.) {
sprintf(e->errbuf, "Expected a lambda parameter, lambda>0");
goto ERROR;
}
}
sum = esl_vec_DSum(hxp->q, hxp->K);
if (fabs(sum-1.0) > 0.05) {
sprintf(e->errbuf, "Expected mixture coefficients to sum to 1");
goto ERROR;
}
esl_vec_DNorm(hxp->q, hxp->K);
*ret_hxp = hxp;
return eslOK;
ERROR:
esl_hyperexp_Destroy(hxp);
return eslEFORMAT;
}
int main(void)
{
double *p;
char labels[] = "ACGT";
int n = 4;
p = malloc(sizeof(double) * n);
esl_vec_DSet(p, n, 1.0);
esl_vec_DNorm(p, n);
esl_vec_DDump(stdout, p, n, labels);
free(p);
return 0;
}
/* Function: esl_msaweight_BLOSUM()
* Synopsis: BLOSUM weights.
* Incept: SRE, Sun Nov 5 09:52:41 2006 [Janelia]
*
* Purpose: Given a multiple sequence alignment <msa> and an identity
* threshold <maxid>, calculate sequence weights using the
* BLOSUM algorithm (Henikoff and Henikoff, PNAS
* 89:10915-10919, 1992). These weights are stored
* internally in the <msa> object, replacing any weights
* that may have already been there. Weights are $\geq 0$
* and they sum to <msa->nseq>.
*
* The algorithm does a single linkage clustering by
* fractional id, defines clusters such that no two clusters
* have a pairwise link $\geq$ <maxid>), and assigns
* weights of $\frac{1}{M_i}$ to each of the $M_i$
* sequences in each cluster $i$. The <maxid> threshold
* is a fractional pairwise identity, in the range
* $0..1$.
*
* The <msa> may be in either digitized or text mode.
* Digital mode is preferred, so that the pairwise identity
* calculations deal with degenerate residue symbols
* properly.
*
* Returns: <eslOK> on success, and the weights inside <msa> have been
* modified.
*
* Throws: <eslEMEM> on allocation error. <eslEINVAL> if a pairwise
* identity calculation fails because of corrupted sequence
* data. In either case, the <msa> is unmodified.
*
* Xref: [Henikoff92]; squid::weight.c::BlosumWeights().
*/
int
esl_msaweight_BLOSUM(ESL_MSA *msa, double maxid)
{
int *c = NULL; /* cluster assignments for each sequence */
int *nmem = NULL; /* number of seqs in each cluster */
int nc; /* number of clusters */
int i; /* loop counter */
int status;
/* Contract checks
*/
ESL_DASSERT1( (maxid >= 0. && maxid <= 1.) );
ESL_DASSERT1( (msa->nseq >= 1) );
ESL_DASSERT1( (msa->alen >= 1) );
if (msa->nseq == 1) { msa->wgt[0] = 1.0; return eslOK; }
if ((status = esl_msacluster_SingleLinkage(msa, maxid, &c, NULL, &nc)) != eslOK) goto ERROR;
ESL_ALLOC(nmem, sizeof(int) * nc);
esl_vec_ISet(nmem, nc, 0);
for (i = 0; i < msa->nseq; i++) nmem[c[i]]++;
for (i = 0; i < msa->nseq; i++) msa->wgt[i] = 1. / (double) nmem[c[i]];
/* Make weights normalize up to nseq, and return.
*/
esl_vec_DNorm(msa->wgt, msa->nseq);
esl_vec_DScale(msa->wgt, msa->nseq, (double) msa->nseq);
msa->flags |= eslMSA_HASWGTS;
free(nmem);
free(c);
return eslOK;
ERROR:
if (c != NULL) free(c);
if (nmem != NULL) free(nmem);
return status;
}
static void
utest_pvectors(void)
{
char *msg = "pvector unit test failed";
double p1[4] = { 0.25, 0.25, 0.25, 0.25 };
double p2[4];
double p3[4];
float p1f[4];
float p2f[4] = { 0.0, 0.5, 0.5, 0.0 };
float p3f[4];
int n = 4;
double result;
esl_vec_D2F(p1, n, p1f);
esl_vec_F2D(p2f, n, p2);
if (esl_vec_DValidate(p1, n, 1e-12, NULL) != eslOK) esl_fatal(msg);
if (esl_vec_FValidate(p1f, n, 1e-7, NULL) != eslOK) esl_fatal(msg);
result = esl_vec_DEntropy(p1, n); if (esl_DCompare(2.0, result, 1e-9) != eslOK) esl_fatal(msg);
result = esl_vec_FEntropy(p1f, n); if (esl_DCompare(2.0, result, 1e-9) != eslOK) esl_fatal(msg);
result = esl_vec_DEntropy(p2, n); if (esl_DCompare(1.0, result, 1e-9) != eslOK) esl_fatal(msg);
result = esl_vec_FEntropy(p2f, n); if (esl_DCompare(1.0, result, 1e-9) != eslOK) esl_fatal(msg);
result = esl_vec_DRelEntropy(p2, p1, n); if (esl_DCompare(1.0, result, 1e-9) != eslOK) esl_fatal(msg);
result = esl_vec_FRelEntropy(p2f, p1f, n); if (esl_DCompare(1.0, result, 1e-9) != eslOK) esl_fatal(msg);
result = esl_vec_DRelEntropy(p1, p2, n); if (result != eslINFINITY) esl_fatal(msg);
result = esl_vec_FRelEntropy(p1f, p2f, n); if (result != eslINFINITY) esl_fatal(msg);
esl_vec_DLog(p2, n);
if (esl_vec_DLogValidate(p2, n, 1e-12, NULL) != eslOK) esl_fatal(msg);
esl_vec_DExp(p2, n);
if (p2[0] != 0.) esl_fatal(msg);
esl_vec_FLog(p2f, n);
if (esl_vec_FLogValidate(p2f, n, 1e-7, NULL) != eslOK) esl_fatal(msg);
esl_vec_FExp(p2f, n);
if (p2f[0] != 0.) esl_fatal(msg);
esl_vec_DCopy(p2, n, p3);
esl_vec_DScale(p3, n, 10.);
esl_vec_DNorm(p3, n);
if (esl_vec_DCompare(p2, p3, n, 1e-12) != eslOK) esl_fatal(msg);
esl_vec_DLog(p3, n);
result = esl_vec_DLogSum(p3, n); if (esl_DCompare(0.0, result, 1e-12) != eslOK) esl_fatal(msg);
esl_vec_DIncrement(p3, n, 2.0);
esl_vec_DLogNorm(p3, n);
if (esl_vec_DCompare(p2, p3, n, 1e-12) != eslOK) esl_fatal(msg);
esl_vec_FCopy(p2f, n, p3f);
esl_vec_FScale(p3f, n, 10.);
esl_vec_FNorm(p3f, n);
if (esl_vec_FCompare(p2f, p3f, n, 1e-7) != eslOK) esl_fatal(msg);
esl_vec_FLog(p3f, n);
result = esl_vec_FLogSum(p3f, n); if (esl_DCompare(0.0, result, 1e-7) != eslOK) esl_fatal(msg);
esl_vec_FIncrement(p3f, n, 2.0);
esl_vec_FLogNorm(p3f, n);
if (esl_vec_FCompare(p2f, p3f, n, 1e-7) != eslOK) esl_fatal(msg);
return;
}
/* Function: esl_msaweight_GSC()
* Synopsis: GSC weights.
* Incept: SRE, Fri Nov 3 13:31:14 2006 [Janelia]
*
* Purpose: Given a multiple sequence alignment <msa>, calculate
* sequence weights according to the
* Gerstein/Sonnhammer/Chothia algorithm. These weights
* are stored internally in the <msa> object, replacing
* any weights that may have already been there. Weights
* are $\geq 0$ and they sum to <msa->nseq>.
*
* The <msa> may be in either digitized or text mode.
* Digital mode is preferred, so that distance calculations
* used by the GSC algorithm are robust against degenerate
* residue symbols.
*
* This is an implementation of Gerstein et al., "A method to
* weight protein sequences to correct for unequal
* representation", JMB 236:1067-1078, 1994.
*
* The algorithm is $O(N^2)$ memory (it requires a pairwise
* distance matrix) and $O(N^3 + LN^2)$ time ($N^3$ for a UPGMA
* tree building step, $LN^2$ for distance matrix construction)
* for an alignment of N sequences and L columns.
*
* In the current implementation, the actual memory
* requirement is dominated by two full NxN distance
* matrices (one tmp copy in UPGMA, and one here): for
* 8-byte doubles, that's $16N^2$ bytes. To keep the
* calculation under memory limits, don't process large
* alignments: max 1400 sequences for 32 MB, max 4000
* sequences for 256 MB, max 8000 seqs for 1 GB. Watch
* out, because Pfam alignments can easily blow this up.
*
* Note: Memory usage could be improved. UPGMA consumes a distance
* matrix, but that can be D itself, not a copy, if the
* caller doesn't mind the destruction of D. Also, D is
* symmetrical, so we could use upper or lower triangular
* matrices if we rewrote dmatrix to allow them.
*
* I also think UPGMA can be reduced to O(N^2) time, by
* being more tricky about rapidly identifying the minimum
* element: could keep min of each row, and update that,
* I think.
*
* Returns: <eslOK> on success, and the weights inside <msa> have been
* modified.
*
* Throws: <eslEINVAL> if the alignment data are somehow invalid and
* distance matrices can't be calculated. <eslEMEM> on an
* allocation error. In either case, the original <msa> is
* left unmodified.
*
* Xref: [Gerstein94]; squid::weight.c::GSCWeights(); STL11/81.
*/
int
esl_msaweight_GSC(ESL_MSA *msa)
{
ESL_DMATRIX *D = NULL; /* distance matrix */
ESL_TREE *T = NULL; /* UPGMA tree */
double *x = NULL; /* storage per node, 0..N-2 */
double lw, rw; /* total branchlen on left, right subtrees */
double lx, rx; /* distribution of weight to left, right side */
int i; /* counter over nodes */
int status;
/* Contract checks
*/
ESL_DASSERT1( (msa != NULL) );
ESL_DASSERT1( (msa->nseq >= 1) );
ESL_DASSERT1( (msa->alen >= 1) );
ESL_DASSERT1( (msa->wgt != NULL) );
if (msa->nseq == 1) { msa->wgt[0] = 1.0; return eslOK; }
/* GSC weights use a rooted tree with "branch lengths" calculated by
* UPGMA on a fractional difference matrix - pretty crude.
*/
if (! (msa->flags & eslMSA_DIGITAL)) {
if ((status = esl_dst_CDiffMx(msa->aseq, msa->nseq, &D)) != eslOK) goto ERROR;
}
#ifdef eslAUGMENT_ALPHABET
else {
if ((status = esl_dst_XDiffMx(msa->abc, msa->ax, msa->nseq, &D)) != eslOK) goto ERROR;
}
#endif
/* oi, look out here. UPGMA is correct, but old squid library uses
* single linkage, so for regression tests ONLY, we use single link.
*/
#ifdef eslMSAWEIGHT_REGRESSION
if ((status = esl_tree_SingleLinkage(D, &T)) != eslOK) goto ERROR;
#else
if ((status = esl_tree_UPGMA(D, &T)) != eslOK) goto ERROR;
#endif
esl_tree_SetCladesizes(T);
ESL_ALLOC(x, sizeof(double) * (T->N-1));
/* Postorder traverse (leaves to root) to calculate the total branch
* length under each internal node; store this in x[]. Remember the
* total branch length (x[0]) for a future sanity check.
*/
for (i = T->N-2; i >= 0; i--)
{
x[i] = T->ld[i] + T->rd[i];
if (T->left[i] > 0) x[i] += x[T->left[i]];
if (T->right[i] > 0) x[i] += x[T->right[i]];
}
/* Preorder traverse (root to leaves) to calculate the weights. Now
* we use x[] to mean, the total weight *above* this node that we will
* apportion to the node's left and right children. The two
* meanings of x[] never cross: every x[] beneath x[i] is still a
* total branch length.
*
* Because the API guarantees that msa is returned unmodified in case
* of an exception, and we're touching msa->wgt here, no exceptions
* may be thrown from now on in this function.
*/
x[0] = 0; /* initialize: no branch to the root. */
for (i = 0; i <= T->N-2; i++)
{
lw = T->ld[i]; if (T->left[i] > 0) lw += x[T->left[i]];
rw = T->rd[i]; if (T->right[i] > 0) rw += x[T->right[i]];
if (lw+rw == 0.)
{
/* A special case arises in GSC weights when all branch lengths in a subtree are 0.
* In this case, all seqs in this clade should get equal weights, sharing x[i] equally.
* So, split x[i] in proportion to cladesize, not to branch weight.
*/
if (T->left[i] > 0) lx = x[i] * ((double) T->cladesize[T->left[i]] / (double) T->cladesize[i]);
else lx = x[i] / (double) T->cladesize[i];
if (T->right[i] > 0) rx = x[i] * ((double) T->cladesize[T->right[i]] / (double) T->cladesize[i]);
else rx = x[i] / (double) T->cladesize[i];
}
else /* normal case: x[i] split in proportion to branch weight. */
{
lx = x[i] * lw/(lw+rw);
rx = x[i] * rw/(lw+rw);
}
if (T->left[i] <= 0) msa->wgt[-(T->left[i])] = lx + T->ld[i];
else x[T->left[i]] = lx + T->ld[i];
if (T->right[i] <= 0) msa->wgt[-(T->right[i])] = rx + T->rd[i];
else x[T->right[i]] = rx + T->rd[i];
}
/* Renormalize weights to sum to N.
*/
esl_vec_DNorm(msa->wgt, msa->nseq);
esl_vec_DScale(msa->wgt, msa->nseq, (double) msa->nseq);
msa->flags |= eslMSA_HASWGTS;
free(x);
esl_tree_Destroy(T);
esl_dmatrix_Destroy(D);
return eslOK;
ERROR:
if (x != NULL) free(x);
if (T != NULL) esl_tree_Destroy(T);
if (D != NULL) esl_dmatrix_Destroy(D);
return status;
}
/* Function: esl_msaweight_PB()
* Synopsis: PB (position-based) weights.
* Incept: SRE, Sun Nov 5 08:59:28 2006 [Janelia]
*
* Purpose: Given a multiple alignment <msa>, calculate sequence
* weights according to the position-based weighting
* algorithm (Henikoff and Henikoff, JMB 243:574-578,
* 1994). These weights are stored internally in the <msa>
* object, replacing any weights that may have already been
* there. Weights are $\geq 0$ and they sum to <msa->nseq>.
*
* The <msa> may be in either digitized or text mode.
* Digital mode is preferred, so that the algorithm
* deals with degenerate residue symbols properly.
*
* The Henikoffs' algorithm does not give rules for dealing
* with gaps or degenerate residue symbols. The rule here
* is to ignore them. This means that longer sequences
* initially get more weight; hence a "double
* normalization" in which the weights are first divided by
* sequence length in canonical residues (to compensate for
* that effect), then normalized to sum to nseq.
*
* An advantage of the PB method is efficiency.
* It is $O(1)$ in memory and $O(NL)$ time, for an alignment of
* N sequences and L columns. This makes it a good method
* for ad hoc weighting of very deep alignments.
*
* When the alignment is in simple text mode, IUPAC
* degenerate symbols are not dealt with correctly; instead,
* the algorithm simply uses the 26 letters as "residues"
* (case-insensitively), and treats all other residues as
* gaps.
*
* Returns: <eslOK> on success, and the weights inside <msa> have been
* modified.
*
* Throws: <eslEMEM> on allocation error, in which case <msa> is
* returned unmodified.
*
* Xref: [Henikoff94b]; squid::weight.c::PositionBasedWeights().
*/
int
esl_msaweight_PB(ESL_MSA *msa)
{
int *nres = NULL; /* counts of each residue observed in a column */
int ntotal; /* number of different symbols observed in a column */
int rlen; /* number of residues in a sequence */
int idx, pos, i;
int K; /* alphabet size */
int status;
/* Contract checks
*/
ESL_DASSERT1( (msa->nseq >= 1) );
ESL_DASSERT1( (msa->alen >= 1) );
if (msa->nseq == 1) { msa->wgt[0] = 1.0; return eslOK; }
/* Initialize
*/
if (! (msa->flags & eslMSA_DIGITAL))
{ ESL_ALLOC(nres, sizeof(int) * 26); K = 26; }
#ifdef eslAUGMENT_ALPHABET
else
{ ESL_ALLOC(nres, sizeof(int) * msa->abc->K); K = msa->abc->K; }
#endif
esl_vec_DSet(msa->wgt, msa->nseq, 0.);
/* This section handles text alignments */
if (! (msa->flags & eslMSA_DIGITAL))
{
for (pos = 0; pos < msa->alen; pos++)
{
/* Collect # of letters A..Z in this column, and total */
esl_vec_ISet(nres, K, 0.);
for (idx = 0; idx < msa->nseq; idx++)
if (isalpha((int) msa->aseq[idx][pos]))
nres[toupper((int) msa->aseq[idx][pos]) - 'A'] ++;
for (ntotal = 0, i = 0; i < K; i++) if (nres[i] > 0) ntotal++;
/* Bump weight on each seq by PB rule */
if (ntotal > 0) {
for (idx = 0; idx < msa->nseq; idx++) {
if (isalpha((int) msa->aseq[idx][pos]))
msa->wgt[idx] += 1. /
(double) (ntotal * nres[toupper((int) msa->aseq[idx][pos]) - 'A'] );
}
}
}
/* first normalization by # of residues counted in each seq */
for (idx = 0; idx < msa->nseq; idx++) {
for (rlen = 0, pos = 0; pos < msa->alen; pos++)
if (isalpha((int) msa->aseq[idx][pos])) rlen++;
if (ntotal > 0) msa->wgt[idx] /= (double) rlen;
/* if rlen == 0 for this seq, its weight is still 0.0, as initialized. */
}
}
/* This section handles digital alignments. */
#ifdef eslAUGMENT_ALPHABET
else
{
for (pos = 1; pos <= msa->alen; pos++)
{
/* Collect # of residues 0..K-1 in this column, and total # */
esl_vec_ISet(nres, K, 0.);
for (idx = 0; idx < msa->nseq; idx++)
if (esl_abc_XIsCanonical(msa->abc, msa->ax[idx][pos]))
nres[(int) msa->ax[idx][pos]] ++;
for (ntotal = 0, i = 0; i < K; i++) if (nres[i] > 0) ntotal++;
/* Bump weight on each sequence by PB rule */
if (ntotal > 0) {
for (idx = 0; idx < msa->nseq; idx++) {
if (esl_abc_XIsCanonical(msa->abc, msa->ax[idx][pos]))
msa->wgt[idx] += 1. / (double) (ntotal * nres[msa->ax[idx][pos]]);
}
}
}
/* first normalization by # of residues counted in each seq */
for (idx = 0; idx < msa->nseq; idx++)
{
for (rlen = 0, pos = 1; pos <= msa->alen; pos++)
if (esl_abc_XIsCanonical(msa->abc, msa->ax[idx][pos])) rlen++;
if (rlen > 0) msa->wgt[idx] /= (double) rlen;
/* if rlen == 0 for this seq, its weight is still 0.0, as initialized. */
}
}
#endif
/* Make weights normalize up to nseq, and return. In pathological
* case where all wgts were 0 (no seqs contain any unambiguous
* residues), weights become 1.0.
*/
esl_vec_DNorm(msa->wgt, msa->nseq);
esl_vec_DScale(msa->wgt, msa->nseq, (double) msa->nseq);
msa->flags |= eslMSA_HASWGTS;
free(nres);
return eslOK;
ERROR:
if (nres != NULL) free(nres);
return status;
}
/* dump_infocontent_info
*
* Given an MSA with RF annotation, dump information content per column data to
* an open output file.
*/
static int dump_infocontent_info(FILE *fp, ESL_ALPHABET *abc, double **abc_ct, int use_weights, int nali, int64_t alen, int nseq, int *i_am_rf, char *msa_name, char *alifile, char *errbuf)
{
int status;
int apos, rfpos;
double bg_ent;
double *bg = NULL;
double *abc_freq = NULL;
double nnongap;
ESL_ALLOC(bg, sizeof(double) * abc->K);
esl_vec_DSet(bg, abc->K, 1./(abc->K));
bg_ent = esl_vec_DEntropy(bg, abc->K);
free(bg);
ESL_ALLOC(abc_freq, sizeof(double) * abc->K);
fprintf(fp, "# Information content per column (bits):\n");
fprintf(fp, "# Alignment file: %s\n", alifile);
fprintf(fp, "# Alignment idx: %d\n", nali);
if(msa_name != NULL) { fprintf(fp, "# Alignment name: %s\n", msa_name); }
fprintf(fp, "# Number of sequences: %d\n", nseq);
if(use_weights) { fprintf(fp, "# IMPORTANT: Counts are weighted based on sequence weights in alignment file.\n"); }
else { fprintf(fp, "# Sequence weights from alignment were ignored (if they existed).\n"); }
fprintf(fp, "#\n");
if(i_am_rf != NULL) {
fprintf(fp, "# %7s %7s %10s %10s\n", "rfpos", "alnpos", "freqnongap", "info(bits)");
fprintf(fp, "# %7s %7s %10s %10s\n", "-------", "-------", "----------", "----------");
}
else {
fprintf(fp, "# %7s %10s %10s\n", "alnpos", "freqnongap", "info(bits)");
fprintf(fp, "# %7s %10s %10s\n", "-------", "----------", "----------");
}
rfpos = 0;
for(apos = 0; apos < alen; apos++) {
if(i_am_rf != NULL) {
if(i_am_rf[apos]) {
fprintf(fp, " %7d", rfpos+1);
rfpos++;
}
else {
fprintf(fp, " %7s", "-");
}
}
nnongap = esl_vec_DSum(abc_ct[apos], abc->K);
esl_vec_DCopy(abc_ct[apos], abc->K, abc_freq);
esl_vec_DNorm(abc_freq, abc->K);
fprintf(fp, " %7d %10.8f %10.8f\n", apos+1,
nnongap / (nnongap + abc_ct[apos][abc->K]),
(bg_ent - esl_vec_DEntropy(abc_freq, abc->K)));
}
fprintf(fp, "//\n");
if(abc_freq != NULL) free(abc_freq);
return eslOK;
ERROR:
ESL_FAIL(eslEINVAL, errbuf, "out of memory");
return status; /* NEVERREACHED */
}
