/* Function: p7_GOATrace()
* Synopsis: Optimal accuracy decoding: traceback.
* Incept: SRE, Fri Feb 29 12:59:11 2008 [Janelia]
*
* Purpose: The traceback stage of the optimal accuracy decoding algorithm
* \citep{Kall05}.
*
* Caller provides the OA DP matrix <gx> that was just
* calculated by <p7_GOptimalAccuracy()>, as well as the
* posterior decoding matrix <pp>, which was calculated by
* Forward/Backward on a target sequence of length <L>
* using the query model <gm>.
*
* Caller provides an empty traceback structure <tr> to
* hold the result, allocated to hold optional posterior
* probability annotation on residues (with
* <p7_trace_CreateWithPP()>, generally). This will be
* internally reallocated as needed for larger traces.
*
* Args: gm - query profile
* pp - posterior decoding matrix created by <p7_PosteriorDecoding()>
* gx - OA DP matrix calculated by <p7_OptimalAccuracyDP()>
* tr - RESULT: OA traceback, allocated with posterior probs
*
* Returns: <eslOK> on success, and <tr> contains the OA traceback.
*
* Throws: <eslEMEM> on allocation error.
*/
int
p7_GOATrace(const P7_PROFILE *gm, const P7_GMX *pp, const P7_GMX *gx, P7_TRACE *tr)
{
int i = gx->L; /* position in seq (1..L) */
int k = 0; /* position in model (1..M) */
float postprob;
int sprv, scur;
int status;
#ifdef p7_DEBUGGING
if (tr->N != 0) ESL_EXCEPTION(eslEINVAL, "trace isn't empty: forgot to Reuse()?");
#endif
if ((status = p7_trace_AppendWithPP(tr, p7T_T, k, i, 0.0)) != eslOK) return status;
if ((status = p7_trace_AppendWithPP(tr, p7T_C, k, i, 0.0)) != eslOK) return status;
sprv = p7T_C;
while (sprv != p7T_S)
{
switch (sprv) {
case p7T_M: scur = select_m(gm, gx, i, k); k--; i--; break;
case p7T_D: scur = select_d(gm, gx, i, k); k--; break;
case p7T_I: scur = select_i(gm, gx, i, k); i--; break;
case p7T_N: scur = select_n(i); break;
case p7T_C: scur = select_c(gm, pp, gx, i); break;
case p7T_J: scur = select_j(gm, pp, gx, i); break;
case p7T_E: scur = select_e(gm, gx, i, &k); break;
case p7T_B: scur = select_b(gm, gx, i); break;
default: ESL_EXCEPTION(eslEINVAL, "bogus state in traceback");
}
if (scur == -1) ESL_EXCEPTION(eslEINVAL, "OA traceback choice failed");
postprob = get_postprob(pp, scur, sprv, k, i);
if ((status = p7_trace_AppendWithPP(tr, scur, k, i, postprob)) != eslOK) return status;
/* For NCJ, we had to defer i decrement. */
if ( (scur == p7T_N || scur == p7T_J || scur == p7T_C) && scur == sprv) i--;
sprv = scur;
}
tr->M = gm->M;
tr->L = gx->L;
return p7_trace_Reverse(tr);
}
/* Function: p7_OATrace()
* Synopsis: Optimal accuracy decoding: traceback.
* Incept: SRE, Mon Aug 18 13:53:33 2008 [Janelia]
*
* Purpose: The traceback stage of the optimal accuracy decoding algorithm
* \citep{Kall05}.
*
* Caller provides the OA DP matrix <ox> that was just
* calculated by <p7_OptimalAccuracyDP()>, as well as the
* posterior decoding matrix <pp>, which was calculated by
* Forward/Backward on a target sequence using the query
* model <gm>. Because the calculation depends only on
* <pp>, the target sequence itself need not be provided.
*
* The resulting optimal accuracy decoding traceback is put
* in a caller-provided traceback structure <tr>, which the
* caller has allocated for optional posterior probability
* annotation on residues (with <p7_trace_CreateWithPP()>,
* generally). This structure will be reallocated
* internally if necessary.
*
* Args: om - profile
* pp - posterior probability matrix
* ox - OA matrix to trace, LxM
* tr - storage for the recovered traceback
*
* Returns: <eslOK> on success.
*
* Throws: <eslEMEM> on allocation error.
* <eslEINVAL> if the trace <tr> isn't empty (needs to be Reuse()'d).
*/
int
p7_OATrace(const P7_OPROFILE *om, const P7_OMX *pp, const P7_OMX *ox, P7_TRACE *tr)
{
int i = ox->L; /* position in sequence 1..L */
int k = 0; /* position in model 1..M */
int s0, s1; /* choice of a state */
float postprob;
int status;
if (tr->N != 0) ESL_EXCEPTION(eslEINVAL, "trace not empty; needs to be Reuse()'d?");
if ((status = p7_trace_AppendWithPP(tr, p7T_T, k, i, 0.0)) != eslOK) return status;
if ((status = p7_trace_AppendWithPP(tr, p7T_C, k, i, 0.0)) != eslOK) return status;
s0 = tr->st[tr->N-1];
while (s0 != p7T_S)
{
switch (s0) {
case p7T_M: s1 = select_m(om, ox, i, k); k--; i--; break;
case p7T_D: s1 = select_d(om, ox, i, k); k--; break;
case p7T_I: s1 = select_i(om, ox, i, k); i--; break;
case p7T_N: s1 = select_n(i); break;
case p7T_C: s1 = select_c(om, pp, ox, i); break;
case p7T_J: s1 = select_j(om, pp, ox, i); break;
case p7T_E: s1 = select_e(om, ox, i, &k); break;
case p7T_B: s1 = select_b(om, ox, i); break;
default: ESL_EXCEPTION(eslEINVAL, "bogus state in traceback");
}
if (s1 == -1) ESL_EXCEPTION(eslEINVAL, "OA traceback choice failed");
postprob = get_postprob(pp, s1, s0, k, i);
if ((status = p7_trace_AppendWithPP(tr, s1, k, i, postprob)) != eslOK) return status;
if ( (s1 == p7T_N || s1 == p7T_J || s1 == p7T_C) && s1 == s0) i--;
s0 = s1;
} /* end traceback, at S state */
tr->M = om->M;
tr->L = ox->L;
return p7_trace_Reverse(tr);
}
/* 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;
}
