/* Function: Plan7ESTConfig()
*
* Purpose: Configure a Plan7 model for EST Smith/Waterman
* analysis.
*
* OUTDATED; DO NOT USE WITHOUT RECHECKING
*
* Args: hmm - hmm to configure.
* aacode - 0..63 vector mapping genetic code to amino acids
* estmodel - 20x64 translation matrix, w/ codon bias and substitution error
* dna2 - probability of a -1 frameshift in a triplet
* dna4 - probability of a +1 frameshift in a triplet
*/
void
Plan7ESTConfig(struct plan7_s *hmm, int *aacode, float **estmodel,
float dna2, float dna4)
{
int k;
int x;
float p;
float *tripnull; /* UNFINISHED!!! */
/* configure specials */
hmm->xt[XTN][MOVE] = 1./351.;
hmm->xt[XTN][LOOP] = 350./351.;
hmm->xt[XTE][MOVE] = 1.;
hmm->xt[XTE][LOOP] = 0.;
hmm->xt[XTC][MOVE] = 1./351.;
hmm->xt[XTC][LOOP] = 350./351.;
hmm->xt[XTJ][MOVE] = 1.;
hmm->xt[XTJ][LOOP] = 0.;
/* configure entry/exit */
hmm->begin[1] = 0.5;
FSet(hmm->begin+2, hmm->M-1, 0.5 / ((float)hmm->M - 1.));
hmm->end[hmm->M] = 1.;
FSet(hmm->end, hmm->M-1, 0.5 / ((float)hmm->M - 1.));
/* configure dna triplet/frameshift emissions */
for (k = 1; k <= hmm->M; k++)
{
/* translate aa to triplet probabilities */
for (x = 0; x < 64; x++) {
p = hmm->mat[k][aacode[x]] * estmodel[aacode[x]][x] * (1.-dna2-dna4);
hmm->dnam[x][k] = Prob2Score(p, tripnull[x]);
p = hmm->ins[k][aacode[x]] * estmodel[aacode[x]][x] * (1.-dna2-dna4);
hmm->dnai[x][k] = Prob2Score(p, tripnull[x]);
}
hmm->dnam[64][k] = 0; /* ambiguous codons score 0 (danger?) */
hmm->dna2 = Prob2Score(dna2, 1.);
hmm->dna4 = Prob2Score(dna4, 1.);
}
}
/* Function: CP9Logoddsify()
*
* Purpose: Take an HMM with valid probabilities, and
* fill in the integer log-odds score section of the model.
*
* Notes on log-odds scores (simplified from plan7.c):
* type of parameter probability score
* ----------------- ----------- ------
* any emission p_x log_2 p_x/null_x
* any transition t_x log_2 t_x
*
* Args: hmm - the hmm to calculate scores in.
*
* Return: (void)
* hmm scores are filled in.
*/
void
CP9Logoddsify(CP9_t *hmm)
{
/*printf("in CP9Logoddsify()\n");*/
int k; /* counter for model position */
int x; /* counter for symbols */
int *sc;
int status;
if (hmm->flags & CPLAN9_HASBITS) return;
ESL_ALLOC(sc, hmm->abc->Kp * sizeof(int));
/* Symbol emission scores
*/
sc[hmm->abc->K] = -INFTY; /* gap character */
sc[hmm->abc->Kp-1] = -INFTY; /* missing data character */
sc[hmm->abc->Kp-2] = -INFTY; /* non-residue data character */
/* Insert emission scores, relies on sc[K, Kp-1] initialization to -inf above */
for (k = 0; k <= hmm->M; k++) {
for (x = 0; x < hmm->abc->K; x++)
sc[x] = Prob2Score(hmm->ins[k][x], hmm->null[x]);
esl_abc_IExpectScVec(hmm->abc, sc, hmm->null);
for (x = 0; x < hmm->abc->Kp; x++) {
hmm->isc[x][k] = sc[x];
}
}
/* Match emission scores, 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] = Prob2Score(hmm->mat[k][x], hmm->null[x]);
esl_abc_IExpectScVec(hmm->abc, sc, hmm->null);
for (x = 0; x < hmm->abc->Kp; x++) {
hmm->msc[x][k] = sc[x];
}
}
for (k = 0; k <= hmm->M; k++)
{
hmm->tsc[CTMM][k] = Prob2Score(hmm->t[k][CTMM], 1.0);
hmm->tsc[CTMI][k] = Prob2Score(hmm->t[k][CTMI], 1.0);
hmm->tsc[CTMD][k] = Prob2Score(hmm->t[k][CTMD], 1.0);
hmm->tsc[CTMEL][k] = Prob2Score(hmm->t[k][CTMEL], 1.0);
hmm->tsc[CTIM][k] = Prob2Score(hmm->t[k][CTIM], 1.0);
hmm->tsc[CTII][k] = Prob2Score(hmm->t[k][CTII], 1.0);
hmm->tsc[CTID][k] = Prob2Score(hmm->t[k][CTID], 1.0);
if(k != 0)
{
hmm->tsc[CTDM][k] = Prob2Score(hmm->t[k][CTDM], 1.0);
hmm->tsc[CTDI][k] = Prob2Score(hmm->t[k][CTDI], 1.0);
hmm->tsc[CTDD][k] = Prob2Score(hmm->t[k][CTDD], 1.0);
}
else
{
hmm->tsc[CTDM][k] = -INFTY;
hmm->tsc[CTDD][k] = -INFTY; /*D_0 doesn't exist*/
hmm->tsc[CTDI][k] = -INFTY;
}
if(k != 0)
{
hmm->bsc[k] = Prob2Score(hmm->begin[k], 1.0);
hmm->esc[k] = Prob2Score(hmm->end[k], 1.0);
}
}
hmm->el_selfsc = Prob2Score(hmm->el_self, 1.0);
/* Finally, fill the efficiently reordered transition scores for this HMM. */
for (k = 0 ; k <= hmm->M; k++) {
int *otsc_k = hmm->otsc + k*cp9O_NTRANS;
otsc_k[cp9O_MM] = hmm->tsc[CTMM][k];
otsc_k[cp9O_MI] = hmm->tsc[CTMI][k];
otsc_k[cp9O_MD] = hmm->tsc[CTMD][k];
otsc_k[cp9O_IM] = hmm->tsc[CTIM][k];
otsc_k[cp9O_II] = hmm->tsc[CTII][k];
otsc_k[cp9O_DM] = hmm->tsc[CTDM][k];
otsc_k[cp9O_DD] = hmm->tsc[CTDD][k];
otsc_k[cp9O_ID] = hmm->tsc[CTID][k];
otsc_k[cp9O_DI] = hmm->tsc[CTDI][k];
otsc_k[cp9O_BM] = hmm->bsc[k];
otsc_k[cp9O_MEL]= hmm->tsc[CTMEL][k];
otsc_k[cp9O_ME] = hmm->esc[k];
}
hmm->flags |= CPLAN9_HASBITS; /* raise the log-odds ready flag */
free(sc);
return;
ERROR:
cm_Fail("Memory allocation error.\n");
return; /* never reached */
}
/* Function: cm_tr_penalties_Create()
* Date: EPN, Sat Jan 21 12:03:52 2012
*
* Purpose: Allocate and initialize a CM_TR_PENALTIES object.
* A CM and its emit map are required to determine
* truncation penalty scores. This is annoyingly
* complex, see verbose notes within code below.
*
* Some of the code in this function, specifically
* that which calculates the probability of a fragment
* aligning at a given node, is checkable, but only
* if we disallow truncated begins into insert states.
* However, we want to allow truncated begins in reality.
* I've left in a flag for ignoring inserts (<ignore_inserts>)
* I used in testing this function. Set it to TRUE to
* perform the test.
*
* Returns: Newly allocated CM_TR_PENALTIES object. NULL if out
* of memory.
*/
CM_TR_PENALTIES *
cm_tr_penalties_Create(CM_t *cm, int ignore_inserts, char *errbuf)
{
int status;
int v, nd, m, i1, i2;
int lpos, rpos;
int i;
/* variables used for determining ratio of inserts to match at each consensus position */
float *mexpocc = NULL; /* [0..c..clen] probability match state is used to emit at cons posn c */
float *iexpocc = NULL; /* [0..c..clen] probability insert state is used to emit after cons posn c */
double *psi = NULL; /* [0..v..M-1] expected occupancy of state v */
float m_psi, i1_psi, i2_psi; /* temp psi values */
float summed_psi;
CM_TR_PENALTIES *trp = NULL;
/* variables used for calculating global truncation penalties */
float g_5and3; /* fragment probability if 5' and 3' truncation are allowed */
float g_5or3; /* fragment probability if 5' or 3' truncation are allowed */
/* variables used for calculating local truncation penalties */
float *begin = NULL; /* local begin probabilities 0..v..M-1 */
int subtree_clen; /* consensus length of subtree under this node */
float prv53, prv5, prv3; /* previous node's fragment probability, 5'&3', 5' only, 3'only */
float cur53, cur5, cur3; /* current node's fragment probability, 5'&3', 5' only, 3'only */
int nfrag53, nfrag5, nfrag3; /* number of fragments, 5'&3', 5' only, 3'only */
if(cm == NULL || cm->emap == NULL) goto ERROR;
ESL_ALLOC(trp, sizeof(CM_TR_PENALTIES));
trp->M = cm->M;
trp->ignored_inserts = ignore_inserts;
/* Define truncation penalties for each state v. This will be
* the score for doing a truncated begin into state v.
*
* Important note: For this discussion we assume that sequences can
* only be truncated at consensus positions, which means we don't
* have to worry about truncated begins into inserts. This is an
* approximation (also made by Diana and Sean in the 2009 trCYK
* paper) that greatly simplifies the explanation of the calculation
* of the truncation penalties. The examples in my ELN3 notebook
* also use this simplification. However, I need to be able to do
* truncated begins into insert states in some cases (some pass/mode
* combinations see ELN bottom of p.47). I explain first the
* rationale for calculating truncation penalties ignoring inserts
* and then I describe how I adapt those penalties to allow
* for inserts.
*
* This is a lengthy comment. I've divided it into 3 sections:
* Section 1. Global mode truncation penalties, ignoring inserts.
* Section 2. Local mode truncation penalties, ignoring inserts.
* Section 3. Adapting truncation penalties to allow for inserts.
*
**************************************************************
* Section 1. Global mode truncation penalties, ignoring inserts.
*
* We want the truncation penalty to be the log of the probability
* that the particular fragment we're aligning was generated from
* the following generative process. The generative process differs
* between global and local mode.
*
* In global mode:
* o Sample global parsetree which spans consensus positions 1..clen.
* o Randomly choose g and h in range 1..clen, where h >= g and
* truncate sequence from g..h. The first residue will either be
* an insert before position g, or a match at position g of the
* model. The final residue will either be an insert after position
* h or a match at position h of the model.
*
* All g,h fragments are equiprobable, so the probability of any
* particular fragment is 2 / (clen * (clen+1)). So log_2 of this
* value is the truncation penalty for all truncated alignments in
* global mode where both 5' and 3' truncation are allowed.
*
* We store this penalty, per-state in the
* g_ptyAA[TRPENALTY_5P_AND_3P][0..v..M-1]. The penalty is
* identical for all emitting states. The penalty value for
* non-emitters is IMPOSSIBLE because truncated begins are
* not allowed into non-emitters.
*
* If only 5' OR 3' truncation is allowed, we only truncate at g or
* h, which menas there's 1/clen possible fragments and log_2
* (1/clen) is our global truncation penalty.
*
* However, if 5' truncation is allowed we can only do a truncated
* begin into states that with a consensus subtree that spans
* position clen (since we don't allow a truncation at the 3' end).
* Thus any state whose subtree that doesn't span clen gets
* an IMPOSSIBLE value for its truncation score in:
* g_ptyAA[TRPENALTY_5P_ONLY][0..v..M-1].
*
* Likewise, if 3' truncation is allowed we can only do a truncated
* begin into states that with a consensus subtree that spans
* position 1 (since we don't allow a truncation at the 5' end).
*
* There's an example of computing all three types of penalties for
* a simple CM in ELN 3 p43.
*
************************************************************
* Section 2. Local mode truncation penalties, ignoring inserts.
*
* Generative process that generates fragments in local mode:
* o Sample local begin state b with consensus subtree from i..j from
* local begin state distribution.
* o Randomly choose g and h in range i..j, where h >= g and
* truncate sequence from g..h. The first residue will either be
* an insert before position g, or a match at position g of the
* model. The final residue will either be an insert after position
* h or a match at position h of the model.
*
* Unlike in global mode, in local mode all fragments are not
* equiprobable since the local begin state distribution can be
* anything, and each b allows different sets of fragments to be
* generated (because they can only span from i to j).
*
* The truncation penalty should be the log of the probability of
* aligning the current fragment to the model. So we need to know
* the probability of generating each possible fragment.
* We could calculate probability of any fragment g,h with the
* following inefficient algorithm:
*
* For each start fragment point g,
* For each start fragment point h,
* For each state v,
* If lpos[v] <= g && rpos[v] >= h, then
* prob[g][h] += begin[v] * 2. / (st_clen[v] * (st_clen[v]+1));
*
* Where lpos[v]/rpos[v] are the left/right consensus positions in
* consensus subtree rooted at state v. And st_clen[v] is rpos[v] -
* lpos[v] + 1, the consensus length of that subtree.
*
* This gives us prob[g][h], the probability of generating fragment
* g,h. But we want to apply the penalty to a state, not to a
* fragment, to avoid needing to know the fragment boundaries g,h
* during the DP recursion when applying the penalty.
*
* To facilitate this, we need to find state t, the state with
* smallest subtree that contains g,h. State t is relevant because
* it is the state which will root the alignment of the fragment g,h
* by using a truncated begin transition into t. This gives a new
* algorithm:
*
* For each start fragment point g,
* For each start fragment point h,
* Identify state t, the max valued state for which
* lpos[v] <= g && rpos[v] >= h, then {
* prob[t] += prob[g][h]
* fcount[t]++;
* }
*
* prob[t] will be the probability of observing an alignment that
* uses a truncated begin into t to align any fragment. Then we take
* average over all fragments: prob[t] / fcount[t] (since we'll only
* be aligning one of those fragments) and use the log of that
* probability as the penalty for observing a truncated alignment
* rooted at state t. Conveniently, it turns out that all fragments
* that share t are equiprobable (have equal prob[g][h] values), so
* the average probability is the actual probability for each
* fragment, and thus the correct penalty to apply.
*
* Fortunately, we can compute the correct penalty much more
* efficiently than the two algorithms shown above. The
* efficient way is implemented below. A test that the penalties
* are correctly computed is in cm_tr_penalties_Validate().
*
* This discussion assumes we're truncating 5' and 3', but if we're
* only truncating 5' or 3' The situation is a little different.
*
* There's an example of computing all three types of penalties for
* a simple CM in ELN3 p44-45.
*
************************************************************
* Section 3. Adapting truncation penalties to allow for inserts.
*
* We need to be able to do truncated begins into insert states
* because we enforce that the first/final residue of a sequence be
* included in 5'/3' truncated alignments and we want to be able
* to properly align those residues if they're probably emitted
* by insert states.
*
* The methods/logic explained in sections 1 and 2 above I believe
* is correct IF we ignore inserts (assume truncated begins into
* them are impossible). But we need to allow inserts, so I modify
* the truncation penalties as described above to allow for inserts
* as follows. We can calculate the appropriate truncated begin
* penalty for all MATP_MP, MATL_ML, MATR_MR, BIF_B states as with
* the methods described above by ignoring inserts. This gives us a
* probability p of using that state as the root of the truncated
* alignment, i.e. the truncated begin state. (The log_2 of this
* probability is the penalty.) We then partition p amongst the
* MATP_MP, MATL_ML, MATR_MR, BIF_B states and any parent insert
* states, i.e. any insert state that can transition into the
* match/bif state. For each match/bif state there's 0, 1 or 2
* parent inserts. We then partition p based on the relative
* expected occupancy of these inserts versus the match/bif state.
*
* This is certainly 'incorrect' in that it doesn't reflect the
* true probability of a fragment being aligned to each of the
* states, but it should be a close approximation. I think doing
* it correctly is basically impossible in the context of a single
* state-specific penalty (i.e. the penalty would have to be per-fragment
* which would be hard to deal with in the DP functions).
*/
/* allocate and initialize the penalty arrays */
ESL_ALLOC(trp->g_ptyAA, sizeof(float *) * NTRPENALTY);
ESL_ALLOC(trp->l_ptyAA, sizeof(float *) * NTRPENALTY);
ESL_ALLOC(trp->ig_ptyAA, sizeof(int *) * NTRPENALTY);
ESL_ALLOC(trp->il_ptyAA, sizeof(int *) * NTRPENALTY);
for(i = 0; i < NTRPENALTY; i++) {
trp->g_ptyAA[i] = NULL;
trp->l_ptyAA[i] = NULL;
trp->il_ptyAA[i] = NULL;
trp->ig_ptyAA[i] = NULL;
ESL_ALLOC(trp->g_ptyAA[i], sizeof(float) * cm->M);
ESL_ALLOC(trp->l_ptyAA[i], sizeof(float) * cm->M);
ESL_ALLOC(trp->ig_ptyAA[i], sizeof(int) * cm->M);
ESL_ALLOC(trp->il_ptyAA[i], sizeof(int) * cm->M);
esl_vec_FSet(trp->g_ptyAA[i], cm->M, IMPOSSIBLE);
esl_vec_FSet(trp->l_ptyAA[i], cm->M, IMPOSSIBLE);
esl_vec_ISet(trp->ig_ptyAA[i], cm->M, -INFTY);
esl_vec_ISet(trp->il_ptyAA[i], cm->M, -INFTY);
}
/* DumpEmitMap(stdout, cm->emap, cm); */
/* Calculate local begin probabilities and expected occupancy */
ESL_ALLOC(begin, sizeof(float) * cm->M);
cm_CalculateLocalBeginProbs(cm, cm->pbegin, cm->t, begin);
if((status = cm_ExpectedPositionOccupancy(cm, &mexpocc, &iexpocc, &psi, NULL, NULL, NULL)) != eslOK) goto ERROR;
/* Fill global and local truncation penalties in a single loop. We
* step through all nodes and set the truncation penalties for the
* MATP_MP, MATL_ML, MATR_MR, and BIF_B states and any parent
* inserts (i1, i2) of those states.
*/
g_5and3 = 2. / (cm->clen * (cm->clen+1)); /* for global mode: probability of all fragments if we're truncating 5' and 3' */
g_5or3 = 1. / cm->clen; /* for global mode: probability of all fragments if we're only truncating 5' or 3' */
prv5 = prv3 = prv53 = 0.; /* initialize 'previous' probability values used for calc'ing local truncation penalties */
for(nd = 0; nd < cm->nodes; nd++) {
lpos = (cm->ndtype[nd] == MATP_nd || cm->ndtype[nd] == MATL_nd) ? cm->emap->lpos[nd] : cm->emap->lpos[nd] + 1;
rpos = (cm->ndtype[nd] == MATP_nd || cm->ndtype[nd] == MATR_nd) ? cm->emap->rpos[nd] : cm->emap->rpos[nd] - 1;
/* now set penalties for match and insert states m, i1 and maybe i2 (if we're a MATP_MP or BIF_B) */
if(cm->ndtype[nd] == END_nd) {
prv5 = prv3 = prv53 = 0.;
}
else if(cm->ndtype[nd] == BEGL_nd || cm->ndtype[nd] == BEGR_nd) {
prv5 = (cm->ndtype[nd] == BEGL_nd) ? 0. : trp->l_ptyAA[TRPENALTY_5P_ONLY][cm->plast[cm->nodemap[nd]]]; /* parent BIF_B's probability */;
prv3 = (cm->ndtype[nd] == BEGR_nd) ? 0. : trp->l_ptyAA[TRPENALTY_3P_ONLY][cm->plast[cm->nodemap[nd]]]; /* parent BIF_B's probability */;
prv53 = trp->l_ptyAA[TRPENALTY_5P_AND_3P][cm->plast[cm->nodemap[nd]]]; /* parent BIF_B's probability */
}
else if(cm->ndtype[nd] == MATP_nd || cm->ndtype[nd] == MATL_nd || cm->ndtype[nd] == MATR_nd || cm->ndtype[nd] == BIF_nd) {
/* determine match states and insert states that pertain to this node */
m = cm->nodemap[nd]; /* MATP_MP, MATL_ML, MATR_MR, or BIF_B */
InsertsGivenNodeIndex(cm, nd-1, &i1, &i2);
m_psi = psi[m];
if(cm->ndtype[nd] == MATP_MP) { m_psi += (psi[m+1] + psi[m+2]); } /* include MATP_ML and MATP_MR psi */
i1_psi = (i1 == -1) ? 0. : psi[i1];
i2_psi = (i2 == -1) ? 0. : psi[i2];
summed_psi = m_psi + i1_psi + i2_psi;
if(ignore_inserts) {
i1_psi = i2_psi = 0.;
summed_psi = m_psi;
}
/* Global penalties */
/* sanity check, we should only set truncation penalty once per state */
if(NOT_IMPOSSIBLE(trp->g_ptyAA[TRPENALTY_5P_AND_3P][m])) goto ERROR;
if((i1 != -1) && NOT_IMPOSSIBLE(trp->g_ptyAA[TRPENALTY_5P_AND_3P][i1])) goto ERROR;
if((i2 != -1) && NOT_IMPOSSIBLE(trp->g_ptyAA[TRPENALTY_5P_AND_3P][i2])) goto ERROR;
/* divide up the probability g_5and3 amongst relevant states m, i1, i2, weighted by psi */
trp->g_ptyAA[TRPENALTY_5P_AND_3P][m] = (m_psi / summed_psi) * g_5and3;
if(i1 != -1) trp->g_ptyAA[TRPENALTY_5P_AND_3P][i1] = (i1_psi / summed_psi) * g_5and3;
if(i2 != -1) trp->g_ptyAA[TRPENALTY_5P_AND_3P][i2] = (i2_psi / summed_psi) * g_5and3;
/* same thing, for 5P only and 3P only */
if(rpos == cm->clen) { /* else it will remain IMPOSSIBLE */
trp->g_ptyAA[TRPENALTY_5P_ONLY][m] = (m_psi / summed_psi) * g_5or3;
if(i1 != -1) trp->g_ptyAA[TRPENALTY_5P_ONLY][i1] = (i1_psi / summed_psi) * g_5or3;
if(i2 != -1) trp->g_ptyAA[TRPENALTY_5P_ONLY][i2] = (i2_psi / summed_psi) * g_5or3;
}
if(lpos == 1) { /* else it will remain IMPOSSIBLE */
trp->g_ptyAA[TRPENALTY_3P_ONLY][m] = (m_psi / summed_psi) * g_5or3;
if(i1 != -1) trp->g_ptyAA[TRPENALTY_3P_ONLY][i1] = (i1_psi / summed_psi) * g_5or3;
if(i2 != -1) trp->g_ptyAA[TRPENALTY_3P_ONLY][i2] = (i2_psi / summed_psi) * g_5or3;
}
/* Local penalties */
subtree_clen = rpos - lpos + 1;
nfrag5 = subtree_clen;
nfrag3 = subtree_clen;
nfrag53 = (subtree_clen * (subtree_clen+1)) / 2;
/* determine probability of observing a fragment aligned at
* state m (here, m is what I call t above and in notes) and
* partition that probability between m and i1 and/or i2 by
* relative occupancy of match versus inserts
*/
cur5 = begin[m] / (float) nfrag5 + prv5;
cur3 = begin[m] / (float) nfrag3 + prv3;
cur53 = begin[m] / (float) nfrag53 + prv53;
/* sanity check, we should only set truncation penalty once per state */
if(NOT_IMPOSSIBLE(trp->l_ptyAA[TRPENALTY_5P_AND_3P][m])) goto ERROR;
if((i1 != -1) && NOT_IMPOSSIBLE(trp->l_ptyAA[TRPENALTY_5P_AND_3P][i1])) goto ERROR;
if((i2 != -1) && NOT_IMPOSSIBLE(trp->l_ptyAA[TRPENALTY_5P_AND_3P][i2])) goto ERROR;
trp->l_ptyAA[TRPENALTY_5P_AND_3P][m] = (m_psi / summed_psi) * cur53;
if(i1 != -1) trp->l_ptyAA[TRPENALTY_5P_AND_3P][i1] = (i1_psi / summed_psi) * cur53;
if(i2 != -1) trp->l_ptyAA[TRPENALTY_5P_AND_3P][i2] = (i2_psi / summed_psi) * cur53;
trp->l_ptyAA[TRPENALTY_5P_ONLY][m] = (m_psi / summed_psi) * cur5;
if(i1 != -1) trp->l_ptyAA[TRPENALTY_5P_ONLY][i1] = (i1_psi / summed_psi) * cur5;
if(i2 != -1) trp->l_ptyAA[TRPENALTY_5P_ONLY][i2] = (i2_psi / summed_psi) * cur5;
trp->l_ptyAA[TRPENALTY_3P_ONLY][m] = (m_psi / summed_psi) * cur3;
if(i1 != -1) trp->l_ptyAA[TRPENALTY_3P_ONLY][i1] = (i1_psi / summed_psi) * cur3;
if(i2 != -1) trp->l_ptyAA[TRPENALTY_3P_ONLY][i2] = (i2_psi / summed_psi) * cur3;
prv5 = (cm->ndtype[nd] == MATL_nd) ? cur5 : 0.;
prv3 = (cm->ndtype[nd] == MATR_nd) ? cur3 : 0.;
prv53 = cur53;
}
}
/* all penalties are currently probabilities, convert them to log
* probs and set integer penalties (careful, we have to check if
* IMPOSSIBLE first)
*/
for(v = 0; v < cm->M; v++)
{
if((cm->stid[v] == MATP_MP || cm->stid[v] == MATL_ML || cm->stid[v] == MATR_MR || cm->stid[v] == BIF_B) ||
((cm->sttype[v] == IL_st || cm->sttype[v] == IR_st) && (! StateIsDetached(cm, v))))
{
/* Check for rare special case: if we're a MATP_IL and next
* two states are MATP_IR and END_E, then we won't have set
* a trunction penalty. This state will keep an impossible
* truncated begin score, if we did a truncated begin into
* it we'd just emit from the MATP_IL and then go to the
* END_E anyway (the MATP_IR will be detached.
*/
if(cm->stid[v] == MATP_IL && cm->ndtype[cm->ndidx[v]+1] == END_nd) continue;
/* glocal 5P AND 3P: all of these should have been set to a non-IMPOSSIBLE value */
if(! NOT_IMPOSSIBLE(trp->g_ptyAA[TRPENALTY_5P_AND_3P][v])) goto ERROR;
trp->ig_ptyAA[TRPENALTY_5P_AND_3P][v] = Prob2Score(trp->g_ptyAA[TRPENALTY_5P_AND_3P][v], 1.0);
trp->g_ptyAA[TRPENALTY_5P_AND_3P][v] = sreLOG2(trp->g_ptyAA[TRPENALTY_5P_AND_3P][v]);
/* glocal 5P only: some may be IMPOSSIBLE */
if(NOT_IMPOSSIBLE(trp->g_ptyAA[TRPENALTY_5P_ONLY][v])) {
trp->ig_ptyAA[TRPENALTY_5P_ONLY][v] = Prob2Score(trp->g_ptyAA[TRPENALTY_5P_ONLY][v], 1.0);
trp->g_ptyAA[TRPENALTY_5P_ONLY][v] = sreLOG2(trp->g_ptyAA[TRPENALTY_5P_ONLY][v]);
}
/* glocal 5P only: some may be IMPOSSIBLE */
if(NOT_IMPOSSIBLE(trp->g_ptyAA[TRPENALTY_3P_ONLY][v])) {
trp->ig_ptyAA[TRPENALTY_3P_ONLY][v] = Prob2Score(trp->g_ptyAA[TRPENALTY_3P_ONLY][v], 1.0);
trp->g_ptyAA[TRPENALTY_3P_ONLY][v] = sreLOG2(trp->g_ptyAA[TRPENALTY_3P_ONLY][v]);
}
/* local penalties all of these should have been set to a non-IMPOSSIBLE value */
if(! NOT_IMPOSSIBLE(trp->il_ptyAA[TRPENALTY_5P_AND_3P][v])) goto ERROR;
if(! NOT_IMPOSSIBLE(trp->il_ptyAA[TRPENALTY_5P_ONLY][v])) goto ERROR;
if(! NOT_IMPOSSIBLE(trp->il_ptyAA[TRPENALTY_3P_ONLY][v])) goto ERROR;
trp->il_ptyAA[TRPENALTY_5P_AND_3P][v] = Prob2Score(trp->l_ptyAA[TRPENALTY_5P_AND_3P][v], 1.0);
trp->il_ptyAA[TRPENALTY_5P_ONLY][v] = Prob2Score(trp->l_ptyAA[TRPENALTY_5P_ONLY][v], 1.0);
trp->il_ptyAA[TRPENALTY_3P_ONLY][v] = Prob2Score(trp->l_ptyAA[TRPENALTY_3P_ONLY][v], 1.0);
trp->l_ptyAA[TRPENALTY_5P_AND_3P][v] = sreLOG2(trp->l_ptyAA[TRPENALTY_5P_AND_3P][v]);
trp->l_ptyAA[TRPENALTY_5P_ONLY][v] = sreLOG2(trp->l_ptyAA[TRPENALTY_5P_ONLY][v]);
trp->l_ptyAA[TRPENALTY_3P_ONLY][v] = sreLOG2(trp->l_ptyAA[TRPENALTY_3P_ONLY][v]);
}
}
if(ignore_inserts) {
if((status = cm_tr_penalties_Validate(trp, cm, 0.0001, errbuf)) != eslOK) { printf("%s", errbuf); goto ERROR; }
}
/* cm_tr_penalties_Dump(stdout, cm, trp); */
if(mexpocc != NULL) free(mexpocc);
if(iexpocc != NULL) free(iexpocc);
if(psi != NULL) free(psi);
if(begin != NULL) free(begin);
return trp;
ERROR:
if(mexpocc != NULL) free(mexpocc);
if(iexpocc != NULL) free(iexpocc);
if(psi != NULL) free(psi);
if(begin != NULL) free(begin);
if(trp != NULL) cm_tr_penalties_Destroy(trp);
return NULL;
}
/* Function: P7Logoddsify()
*
* Purpose: Take an HMM with valid probabilities, and
* fill in the integer log-odds score section of the model.
*
* Notes on log-odds scores:
* type of parameter probability score
* ----------------- ----------- ------
* any emission p_x log_2 p_x/null_x
* N,J,C /assume/ p_x = null_x so /always/ score zero.
* transition to emitters t_x log_2 t_x/p1
* (M,I; N,C; J)
* NN and CC loops are often equal to p1, so usu. score zero.
* C->T transition t_x log_2 t_x/p2
* often zero, usu. C->T = p2.
* all other transitions t_x log_2 t_x
* (no null model counterpart, so null prob is 1)
*
* Notes on entry/exit scores, B->M and M->E:
* The probability form model includes delete states 1 and M.
* these states are removed from a search form model to
* prevent B->D...D->E->J->B mute cycles, which would complicate
* dynamic programming algorithms. The data-independent
* S/W B->M and M->E transitions are folded together with
* data-dependent B->D...D->M and M->D...D->E paths.
*
* This process is referred to in the code as "wing folding"
* or "wing retraction"... the analogy is to a swept-wing
* fighter in landing vs. high speed flight configuration.
*
* Note on Viterbi vs. forward flag:
* Wing retraction must take forward vs. Viterbi
* into account. If forward, sum two paths; if Viterbi, take
* max. I tried to slide this by as a sum, without
* the flag, but Alex detected it as a bug, because you can
* then find cases where the Viterbi score doesn't match
* the P7TraceScore().
*
* Args: hmm - the hmm to calculate scores in.
* viterbi_mode - TRUE to fold wings in Viterbi configuration.
*
* Return: (void)
* hmm scores are filled in.
*/
void
P7Logoddsify(struct plan7_s *hmm, int viterbi_mode)
{
int k; /* counter for model position */
int x; /* counter for symbols */
float accum;
float tbm, tme;
if (hmm->flags & PLAN7_HASBITS) return;
/* Symbol emission scores
*/
for (k = 1; k <= hmm->M; k++)
{
/* match/insert emissions in main model */
for (x = 0; x < Alphabet_size; x++)
{
hmm->msc[x][k] = Prob2Score(hmm->mat[k][x], hmm->null[x]);
if (k < hmm->M)
hmm->isc[x][k] = Prob2Score(hmm->ins[k][x], hmm->null[x]);
}
/* degenerate match/insert emissions */
for (x = Alphabet_size; x < Alphabet_iupac; x++)
{
hmm->msc[x][k] = DegenerateSymbolScore(hmm->mat[k], hmm->null, x);
if (k < hmm->M)
hmm->isc[x][k] = DegenerateSymbolScore(hmm->ins[k], hmm->null, x);
}
}
/* State transitions.
*
* A note on "folding" of D_1 and D_M.
* These two delete states are folded out of search form models
* in order to prevent null cycles in the dynamic programming
* algorithms (see code below). However, we use their log transitions
* when we save the model! So the following log transition probs
* are used *only* in save files, *never* in search algorithms:
* log (tbd1), D1 -> M2, D1 -> D2
* Mm-1 -> Dm, Dm-1 -> Dm
*
* In a search algorithm, these have to be interpreted as -INFTY
* because their contributions are folded into bsc[] and esc[]
* entry/exit scores. They can't be set to -INFTY here because
* we need them in save files.
*/
for (k = 1; k < hmm->M; k++)
{
hmm->tsc[k][TMM] = Prob2Score(hmm->t[k][TMM], hmm->p1);
hmm->tsc[k][TMI] = Prob2Score(hmm->t[k][TMI], hmm->p1);
hmm->tsc[k][TMD] = Prob2Score(hmm->t[k][TMD], 1.0);
hmm->tsc[k][TIM] = Prob2Score(hmm->t[k][TIM], hmm->p1);
hmm->tsc[k][TII] = Prob2Score(hmm->t[k][TII], hmm->p1);
hmm->tsc[k][TDM] = Prob2Score(hmm->t[k][TDM], hmm->p1);
hmm->tsc[k][TDD] = Prob2Score(hmm->t[k][TDD], 1.0);
}
/* B->M entry transitions. Note how D_1 is folded out.
* M1 is just B->M1
* M2 is sum (or max) of B->M2 and B->D1->M2
* M_k is sum (or max) of B->M_k and B->D1...D_k-1->M_k
* These have to be done in log space, else you'll get
* underflow errors; and we also have to watch for log(0).
* A little sloppier than it probably has to be; historically,
* doing in this in log space was in response to a bug report.
*/
accum = hmm->tbd1 > 0.0 ? log(hmm->tbd1) : -9999.;
for (k = 1; k <= hmm->M; k++)
{
tbm = hmm->begin[k] > 0. ? log(hmm->begin[k]) : -9999.; /* B->M_k part */
/* B->D1...D_k-1->M_k part we get from accum*/
if (k > 1 && accum > -9999.)
{
if (hmm->t[k-1][TDM] > 0.0)
{
if (viterbi_mode) tbm = MAX(tbm, accum + log(hmm->t[k-1][TDM]));
else tbm = LogSum(tbm, accum + log(hmm->t[k-1][TDM]));
}
accum = (hmm->t[k-1][TDD] > 0.0) ? accum + log(hmm->t[k-1][TDD]) : -9999.;
}
/* Convert from log_e to scaled integer log_2 odds. */
if (tbm > -9999.)
hmm->bsc[k] = (int) floor(0.5 + INTSCALE * 1.44269504 * (tbm - log(hmm->p1)));
else
hmm->bsc[k] = -INFTY;
}
/* M->E exit transitions. Note how D_M is folded out.
* M_M is 1 by definition
* M_M-1 is sum of M_M-1->E and M_M-1->D_M->E, where D_M->E is 1 by definition
* M_k is sum of M_k->E and M_k->D_k+1...D_M->E
* Must be done in log space to avoid underflow errors.
* A little sloppier than it probably has to be; historically,
* doing in this in log space was in response to a bug report.
*/
hmm->esc[hmm->M] = 0;
accum = 0.;
for (k = hmm->M-1; k >= 1; k--)
{
tme = hmm->end[k] > 0. ? log(hmm->end[k]) : -9999.;
if (accum > -9999.)
{
if (hmm->t[k][TMD] > 0.0)
{
if (viterbi_mode) tme = MAX(tme, accum + log(hmm->t[k][TMD]));
else tme = LogSum(tme, accum + log(hmm->t[k][TMD]));
}
accum = (hmm->t[k][TDD] > 0.0) ? accum + log(hmm->t[k][TDD]) : -9999.;
}
/* convert from log_e to scaled integer log odds. */
hmm->esc[k] = (tme > -9999.) ? (int) floor(0.5 + INTSCALE * 1.44269504 * tme) : -INFTY;
}
/* special transitions */
hmm->xsc[XTN][LOOP] = Prob2Score(hmm->xt[XTN][LOOP], hmm->p1);
hmm->xsc[XTN][MOVE] = Prob2Score(hmm->xt[XTN][MOVE], 1.0);
hmm->xsc[XTE][LOOP] = Prob2Score(hmm->xt[XTE][LOOP], 1.0);
hmm->xsc[XTE][MOVE] = Prob2Score(hmm->xt[XTE][MOVE], 1.0);
hmm->xsc[XTC][LOOP] = Prob2Score(hmm->xt[XTC][LOOP], hmm->p1);
hmm->xsc[XTC][MOVE] = Prob2Score(hmm->xt[XTC][MOVE], 1.-hmm->p1);
hmm->xsc[XTJ][LOOP] = Prob2Score(hmm->xt[XTJ][LOOP], hmm->p1);
hmm->xsc[XTJ][MOVE] = Prob2Score(hmm->xt[XTJ][MOVE], 1.0);
hmm->flags |= PLAN7_HASBITS; /* raise the log-odds ready flag */
}
/* Function: CP9_reconfig2sub()
* EPN 10.16.06
*
* Purpose: Given a CM Plan 9 HMM and a start position
* (spos) and end position (epos) that a sub CM models,
* reconfigure the HMM so that it can only start in the
* node that models spos (spos_nd) end in the node that
* models epos (epos_nd).
*
* If we're reconfiguring a CP9 HMM that ONLY models the
* consensus columns spos to epos, then spos_nd == 1
* and epos_nd == hmm->M, but this is not necessarily true.
* We may be reconfiguring a CP9 HMM that models the
* full alignment including positions before and/or after
* spos and epos. In this case spos_nd == spos and
* epos_nd == epos;
*
* Args: hmm - the CP9 model w/ data-dep prob's valid
* spos - first consensus column modelled by some original
* full length, template CP9 HMM that 'hmm' models.
* epos - final consensus column modelled by some original
* CP9 HMM that 'hmm' models.
* spos_nd - the node of 'hmm' that models spos.
* (1 if 'hmm' only has (epos-spos+1) nodes
* (spos if 'hmm' has a node for each column of original aln)
* epos_nd - the node of the 'hmm' in that models epos.
* (hmm->M if 'hmm' only has (epos-spos+1) nodes
* (epos if 'hmm' has a node for each column of original aln)
* orig_phi - the 2D phi array for the original CP9 HMM.
* Return: (void)
* HMM probabilities are modified.
*/
void
CP9_reconfig2sub(CP9_t *hmm, int spos, int epos, int spos_nd,
int epos_nd, double **orig_phi)
{
/* Make the necessary modifications. Since in cmalign --sub mode this
* function will be called potentially once for each sequence, we
* don't want to call CP9Logoddsify(), but rather only logoddsify
* the parameters that are different.
*/
/* Configure entry.
* Exactly 3 ways to start, B->M_1 (hmm->begin[1]), B->I_0 (hmm->t[0][CTMI]),
* and B->D_1 (hmm->t[0][CTMD])
*/
/* prob of starting in M_spos is (1. - prob of starting in I_spos-1) as there is no D_spos-1 -> M_spos trans */
if(spos > 1)
{
hmm->begin[spos_nd] = 1.-((orig_phi[spos-1][HMMINSERT] * (1. - hmm->t[spos-1][CTII])) +
(orig_phi[spos ][HMMDELETE] - (orig_phi[spos-1][HMMINSERT] * hmm->t[spos-1][CTID])));
hmm->t[spos_nd-1][CTMI] = (orig_phi[spos-1][HMMINSERT] * (1. - hmm->t[spos-1][CTII]));
hmm->t[spos_nd-1][CTMD] = orig_phi[spos ][HMMDELETE] - (orig_phi[spos-1][HMMINSERT] * hmm->t[spos-1][CTID]);
hmm->t[spos_nd-1][CTMM] = 0.; /* probability of going from B(M_0) to M_1 is begin[1] */
hmm->t[spos_nd-1][CTMEL] = 0.; /* can't go to EL from B(M_0) */
hmm->t[spos_nd-1][CTDM] = 0.; /* D_0 doesn't exist */
hmm->t[spos_nd-1][CTDI] = 0.; /* D_0 doesn't exist */
hmm->t[spos_nd-1][CTDD] = 0.; /* D_0 doesn't exist */
hmm->bsc[spos_nd] = Prob2Score(hmm->begin[1], 1.0);
hmm->tsc[CTMM][spos_nd-1] = -INFTY; /* probability of going from B(M_0) to M_1 is begin[1] */
hmm->tsc[CTMEL][spos_nd-1] = -INFTY;
hmm->tsc[CTDM][spos_nd-1] = -INFTY; /* D_0 doesn't exist */
hmm->tsc[CTDI][spos_nd-1] = -INFTY; /* D_0 doesn't exist */
hmm->tsc[CTDD][spos_nd-1] = -INFTY; /* D_0 doesn't exist */
hmm->tsc[CTMI][spos_nd-1] = Prob2Score(hmm->t[spos_nd-1][CTMI], 1.0);
hmm->tsc[CTMD][spos_nd-1] = Prob2Score(hmm->t[spos_nd-1][CTMD], 1.0);
}
if(epos < hmm->M)
{
hmm->end[epos_nd] = hmm->t[epos][CTMM] + hmm->t[epos][CTMD];
hmm->t[epos_nd][CTDM] += hmm->t[epos][CTDD];
hmm->t[epos_nd][CTIM] += hmm->t[epos][CTID];
hmm->t[epos_nd][CTMM] = 0.; /* M->E is actually end[M] */
hmm->t[epos_nd][CTMEL] = 0.;
hmm->t[epos_nd][CTMD] = 0.; /* D_M+1 doesn't exist */
hmm->t[epos_nd][CTDD] = 0.; /* D_M+1 doesn't exist */
hmm->t[epos_nd][CTID] = 0.; /* D_M+1 doesn't exist */
hmm->esc[epos_nd] = Prob2Score(hmm->end[epos_nd], 1.0);
hmm->tsc[CTDM][epos_nd] = Prob2Score(hmm->t[epos_nd][CTDM], 1.0);
hmm->tsc[CTIM][epos_nd] = Prob2Score(hmm->t[epos_nd][CTIM], 1.0);
hmm->tsc[CTMM][epos_nd] = -INFTY; /* M->E is actually end[M] */
hmm->tsc[CTMEL][epos_nd] = -INFTY;
hmm->tsc[CTMD][epos_nd] = -INFTY; /* D_M+1 doesn't exist */
hmm->tsc[CTDD][epos_nd] = -INFTY; /* D_M+1 doesn't exist */
hmm->tsc[CTID][epos_nd] = -INFTY; /* D_M+1 doesn't exist */
}
hmm->flags |= CPLAN9_HASBITS; /* raise the log-odds ready flag */
return;
}
/* Function: CPlan9InitEL()
* Incept: EPN, Tue Jun 19 13:10:56 2007
*
* Purpose: Initialize a CP9 HMM for possible EL local ends
* by determining how the EL states should be connected
* based on the CM node topology.
*
* Args: cm - the CM
* cp9 - the CP9 HMM, built from cm
*
* Return: (void)
*/
void
CPlan9InitEL(CM_t *cm, CP9_t *cp9)
{
int status;
CMEmitMap_t *emap; /* consensus emit map for the CM */
int k; /* counter over HMM nodes */
int nd;
int *tmp_el_from_ct;
/* First copy the CM el self transition score/probability: */
cp9->el_self = sreEXP2(cm->el_selfsc);
cp9->el_selfsc = Prob2Score(cp9->el_self, 1.0);
/* For each HMM node k, we can transit FROM >= 0 EL states from
* HMM nodes kp. Determine how many such valid transitions exist
* from each node, then allocate and fill cp9->el_from_idx[k] and
* cp9->el_from_cmnd arrays based on that.
* This two-pass method saves memory b/c we only allocate for
* what we'll need.
*/
emap = CreateEmitMap(cm);
/* Initialize to 0 */
for(k = 0; k <= cp9->M; k++)
{
cp9->el_from_ct[k] = 0;
cp9->has_el[k] = FALSE;
}
cp9->el_from_ct[(cp9->M+1)] = 0; /* special case, we can get to E state from EL states */
/* first pass to get number of valid transitions */
for(nd = 0; nd < cm->nodes; nd++)
{
if ((cm->ndtype[nd] == MATP_nd || cm->ndtype[nd] == MATL_nd ||
cm->ndtype[nd] == MATR_nd || cm->ndtype[nd] == BEGL_nd ||
cm->ndtype[nd] == BEGR_nd) &&
cm->ndtype[nd+1] != END_nd)
{
/*printf("HMM node %d can be reached from HMM node %d's EL state\n", emap->rpos[nd], emap->lpos[nd]);*/
cp9->el_from_ct[emap->rpos[nd]]++;
cp9->has_el[emap->lpos[nd]] = TRUE;
}
}
/* allocate cp9->el_from_idx[k], cp9->el_from_cmnd for all k */
for(k = 0; k <= (cp9->M+1); k++)
{
if(cp9->el_from_idx[k] != NULL) /* if !NULL we already filled it, shouldn't happen */
cm_Fail("ERROR in CPlan9InitEL() el_from_idx has already been initialized\n");
if(cp9->el_from_ct[k] > 0)
{
ESL_ALLOC(cp9->el_from_idx[k], sizeof(int) * cp9->el_from_ct[k]);
ESL_ALLOC(cp9->el_from_cmnd[k],sizeof(int) * cp9->el_from_ct[k]);
}
/* else it remains NULL */
}
/* now fill in cp9->el_from_idx, we need a new counter array */
ESL_ALLOC(tmp_el_from_ct, sizeof(int) * (cp9->M+2));
for(k = 0; k <= (cp9->M+1); k++)
tmp_el_from_ct[k] = 0;
for(nd = 0; nd < cm->nodes; nd++)
{
if ((cm->ndtype[nd] == MATP_nd || cm->ndtype[nd] == MATL_nd ||
cm->ndtype[nd] == MATR_nd || cm->ndtype[nd] == BEGL_nd ||
cm->ndtype[nd] == BEGR_nd) &&
cm->ndtype[nd+1] != END_nd)
{
k = emap->rpos[nd];
cp9->el_from_idx[k][tmp_el_from_ct[k]] = emap->lpos[nd];
cp9->el_from_cmnd[k][tmp_el_from_ct[k]] = nd;
tmp_el_from_ct[k]++;
}
}
/* Debugging printfs */
/* for(k = 0; k <= (cp9->M+1); k++)
{
for(c = 0; c < cp9->el_from_ct[k]; c++)
printf("cp9->el_from_idx[%3d][%2d]: %4d\n", k, c, cp9->el_from_idx[k][c]);
if(cp9->has_el[k])
printf("node k:%3d HAS an EL!\n", k);
}*/
/* Free memory and exit */
free(tmp_el_from_ct);
FreeEmitMap(emap);
return;
ERROR:
cm_Fail("Memory allocation error.");
}
/* Function: CP9Logoddsify()
*
* Purpose: Take an HMM with valid probabilities, and
* fill in the integer log-odds score section of the model.
*
* Notes on log-odds scores (simplified from plan7.c):
* type of parameter probability score
* ----------------- ----------- ------
* any emission p_x log_2 p_x/null_x
* any transition t_x log_2 t_x
*
* Args: hmm - the hmm to calculate scores in.
*
* Return: (void)
* hmm scores are filled in.
*/
void
CP9Logoddsify(CP9_t *hmm)
{
int k; /* counter for model position */
int x; /* counter for symbols */
int sc[MAXDEGEN]; /* 17, NEED TO INCREASE FOR BIGGER ALPHABETS! */
if (hmm->flags & CPLAN9_HASBITS) return;
/* Symbol emission scores
*/
sc[hmm->abc->K] = -INFTY; /* gap character */
sc[hmm->abc->Kp-1] = -INFTY; /* missing data character */
/* Insert emission scores, relies on sc[K, Kp-1] initialization to -inf above */
for (k = 0; k <= hmm->M; k++) {
for (x = 0; x < hmm->abc->K; x++)
sc[x] = Prob2Score(hmm->ins[k][x], hmm->null[x]);
esl_abc_IExpectScVec(hmm->abc, sc, hmm->null);
for (x = 0; x < hmm->abc->Kp; x++) {
hmm->isc[x][k] = sc[x];
}
}
/* Match emission scores, 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] = Prob2Score(hmm->mat[k][x], hmm->null[x]);
esl_abc_IExpectScVec(hmm->abc, sc, hmm->null);
for (x = 0; x < hmm->abc->Kp; x++) {
hmm->msc[x][k] = sc[x];
}
}
for (k = 0; k <= hmm->M; k++)
{
hmm->tsc[CTMM][k] = Prob2Score(hmm->t[k][CTMM], 1.0);
hmm->tsc[CTMI][k] = Prob2Score(hmm->t[k][CTMI], 1.0);
hmm->tsc[CTMD][k] = Prob2Score(hmm->t[k][CTMD], 1.0);
hmm->tsc[CTMEL][k] = Prob2Score(hmm->t[k][CTMEL], 1.0);
hmm->tsc[CTIM][k] = Prob2Score(hmm->t[k][CTIM], 1.0);
hmm->tsc[CTII][k] = Prob2Score(hmm->t[k][CTII], 1.0);
hmm->tsc[CTID][k] = Prob2Score(hmm->t[k][CTID], 1.0);
if(k != 0)
{
hmm->tsc[CTDM][k] = Prob2Score(hmm->t[k][CTDM], 1.0);
hmm->tsc[CTDI][k] = Prob2Score(hmm->t[k][CTDI], 1.0);
hmm->tsc[CTDD][k] = Prob2Score(hmm->t[k][CTDD], 1.0);
}
else
{
hmm->tsc[CTDM][k] = -INFTY;
hmm->tsc[CTDD][k] = -INFTY; /*D_0 doesn't exist*/
hmm->tsc[CTDI][k] = -INFTY;
}
if(k != 0)
{
hmm->bsc[k] = Prob2Score(hmm->begin[k], 1.0);
//if(hmm->flags & CPLAN9_LOCAL_END) hmm->esc[k] = 0;
//else hmm->esc[k] = -INFTY;
hmm->esc[k] = Prob2Score(hmm->end[k], 1.0);
}
}
hmm->el_selfsc = Prob2Score(hmm->el_self, 1.0);
/* Finally, fill the efficiently reordered transition scores for this HMM. */
for (k = 0 ; k <= hmm->M; k++) {
int *otsc_k = hmm->otsc + k*cp9O_NTRANS;
otsc_k[cp9O_MM] = hmm->tsc[CTMM][k];
otsc_k[cp9O_MI] = hmm->tsc[CTMI][k];
otsc_k[cp9O_MD] = hmm->tsc[CTMD][k];
otsc_k[cp9O_IM] = hmm->tsc[CTIM][k];
otsc_k[cp9O_II] = hmm->tsc[CTII][k];
otsc_k[cp9O_DM] = hmm->tsc[CTDM][k];
otsc_k[cp9O_DD] = hmm->tsc[CTDD][k];
otsc_k[cp9O_ID] = hmm->tsc[CTID][k];
otsc_k[cp9O_DI] = hmm->tsc[CTDI][k];
otsc_k[cp9O_BM] = hmm->bsc[k];
otsc_k[cp9O_MEL]= hmm->tsc[CTMEL][k];
otsc_k[cp9O_ME] = hmm->esc[k];
}
hmm->flags |= CPLAN9_HASBITS; /* raise the log-odds ready flag */
}
