/* build_model():
* Given <msa>, choose HMM architecture, collect counts;
* upon return, <*ret_hmm> is newly allocated and contains
* relative-weighted observed counts.
* Optionally, caller can request an array of inferred traces for
* the <msa> too.
*/
static int
build_model(P7_BUILDER *bld, ESL_MSA *msa, P7_HMM **ret_hmm, P7_TRACE ***opt_tr)
{
int status;
if (bld->arch_strategy == p7_ARCH_FAST)
{
status = p7_Fastmodelmaker( msa, bld->symfrac, bld, ret_hmm, opt_tr);
if (status == eslENORESULT) ESL_XFAIL(status, bld->errbuf, "Alignment %s has no consensus columns w/ > %d%% residues - can't build a model.\n", msa->name != NULL ? msa->name : "", (int) (100 * bld->symfrac));
else if (status == eslEMEM) ESL_XFAIL(status, bld->errbuf, "Memory allocation failure in model construction.\n");
else if (status != eslOK) ESL_XFAIL(status, bld->errbuf, "internal error in model construction.\n");
}
else if (bld->arch_strategy == p7_ARCH_HAND)
{
status = p7_Handmodelmaker( msa, bld, ret_hmm, opt_tr);
if (status == eslENORESULT) ESL_XFAIL(status, bld->errbuf, "Alignment %s has no annotated consensus columns - can't build a model.\n", msa->name != NULL ? msa->name : "");
else if (status == eslEFORMAT) ESL_XFAIL(status, bld->errbuf, "Alignment %s has no reference annotation line\n", msa->name != NULL ? msa->name : "");
else if (status == eslEMEM) ESL_XFAIL(status, bld->errbuf, "Memory allocation failure in model construction.\n");
else if (status != eslOK) ESL_XFAIL(status, bld->errbuf, "internal error in model construction.\n");
}
return eslOK;
ERROR:
return status;
}
int
main(int argc, char **argv)
{
ESL_GETOPTS *go = p7_CreateDefaultApp(options, 1, argc, argv, banner, usage);
char *msafile = esl_opt_GetArg(go, 1);
int fmt = eslMSAFILE_UNKNOWN;
ESL_ALPHABET *abc = NULL;
ESL_MSAFILE *afp = NULL;
ESL_MSA *msa = NULL;
P7_HMM *hmm = NULL;
P7_PRIOR *prior = NULL;
P7_TRACE **trarr = NULL;
P7_BG *bg = NULL;
P7_PROFILE *gm = NULL;
ESL_MSA *postmsa = NULL;
int i;
int status;
/* Standard idioms for opening and reading a digital MSA. (See esl_msafile.c example). */
if (esl_opt_GetBoolean(go, "--rna")) abc = esl_alphabet_Create(eslRNA);
else if (esl_opt_GetBoolean(go, "--dna")) abc = esl_alphabet_Create(eslDNA);
else if (esl_opt_GetBoolean(go, "--amino")) abc = esl_alphabet_Create(eslAMINO);
if ((status = esl_msafile_Open(&abc, msafile, NULL, fmt, NULL, &afp)) != eslOK)
esl_msafile_OpenFailure(afp, status);
bg = p7_bg_Create(abc);
switch (abc->type) {
case eslAMINO: prior = p7_prior_CreateAmino(); break;
case eslDNA: prior = p7_prior_CreateNucleic(); break;
case eslRNA: prior = p7_prior_CreateNucleic(); break;
default: prior = p7_prior_CreateLaplace(abc); break;
}
if (prior == NULL) esl_fatal("Failed to initialize prior");
while ((status = esl_msafile_Read(afp, &msa)) != eslEOF)
{
if (status != eslOK) esl_msafile_ReadFailure(afp, status);
/* The modelmakers collect counts in an HMM structure */
status = p7_Handmodelmaker(msa, NULL, &hmm, &trarr);
if (status == eslENORESULT) esl_fatal("no consensus columns in alignment %s\n", msa->name);
else if (status != eslOK) esl_fatal("failed to build HMM from alignment %s\n", msa->name);
printf("COUNTS:\n");
p7_hmm_Dump(stdout, hmm);
/* These counts, in combination with a prior, are converted to probability parameters */
status = p7_ParameterEstimation(hmm, prior);
if (status != eslOK) esl_fatal("failed to parameterize HMM for %s", msa->name);
printf("PROBABILITIES:\n");
p7_hmm_Dump(stdout, hmm);
/* Just so we can dump a more informatively annotated trace - build a profile */
gm = p7_profile_Create(hmm->M, abc);
p7_profile_Config (gm, hmm, bg);
p7_profile_SetLength(gm, 400);
/* Dump the individual traces */
for (i = 0; i < msa->nseq; i++)
{
printf("Trace %d: %s\n", i+1, msa->sqname[i]);
p7_trace_DumpAnnotated(stdout, trarr[i], gm, msa->ax[i]);
}
/* Create an MSA from the individual traces */
status = p7_tracealign_MSA(msa, trarr, hmm->M, p7_DEFAULT, &postmsa);
if (status != eslOK) esl_fatal("failed to create new MSA from traces\n");
esl_msafile_Write(stdout, postmsa, eslMSAFILE_PFAM);
p7_profile_Destroy(gm);
p7_hmm_Destroy(hmm);
p7_trace_DestroyArray(trarr, msa->nseq);
esl_msa_Destroy(postmsa);
esl_msa_Destroy(msa);
}
esl_msafile_Close(afp);
p7_bg_Destroy(bg);
esl_alphabet_Destroy(abc);
esl_getopts_Destroy(go);
return 0;
}
/* utest_fragments()
* This exercises the building code that deals with fragments,
* creating traces with B->X->{MDI}k and {MDI}k->X->E
* transitions, and making sure we can make MSAs correctly
* from them using p7_tracealign_MSA(). This code was initially
* buggy when first written; bugs first detected by Elena,
* Nov 2009
*/
static void
utest_fragments(void)
{
char *failmsg = "failure in build.c::utest_fragments() unit test";
char msafile[16] = "p7tmpXXXXXX"; /* tmpfile name template */
FILE *ofp = NULL;
ESL_ALPHABET *abc = esl_alphabet_Create(eslAMINO);
ESL_MSAFILE *afp = NULL;
ESL_MSA *msa = NULL;
ESL_MSA *dmsa = NULL;
ESL_MSA *postmsa = NULL;
P7_HMM *hmm = NULL;
P7_TRACE **trarr = NULL;
int i;
/* Write an MSA that tests fragment/missing data transitions.
* When built with Handmodelmaker (using the RF line):
* seq1 forces B->X->Mk and Mk->X->E missing data transitions;
* seq2 forces B->X->Ik and Ik->X->E missing data transitions;
* seq3 forces B->X->Dk and Dk->X->E missing data transitions.
*
* The first two cases can arise from fragment definition in
* model construction, or in an input file.
*
* The X->Dk and Dk->X cases should never happen, but we don't
* prohibit them. They can only arise in an input file, because
* esl_msa_MarkFragments() converts everything before/after
* first/last residue to ~, and won't leave a gap character in
* between.
*
* There's nothing being tested by seq4 and seq5; they're just there.
*/
if (esl_tmpfile_named(msafile, &ofp) != eslOK) esl_fatal(failmsg);
fprintf(ofp, "# STOCKHOLM 1.0\n");
fprintf(ofp, "#=GC RF xxxxx.xxxxxxxxxxxx.xxx\n");
fprintf(ofp, "seq1 ~~~~~~GHIKLMNPQRST~~~~\n");
fprintf(ofp, "seq2 ~~~~~aGHIKLMNPQRSTa~~~\n");
fprintf(ofp, "seq3 ~~~~~~~HIKLMNPQRS~~~~~\n");
fprintf(ofp, "seq4 ACDEF.GHIKLMNPQRST.VWY\n");
fprintf(ofp, "seq5 ACDEF.GHIKLMNPQRST.VWY\n");
fprintf(ofp, "//\n");
fclose(ofp);
/* Read the original as text for comparison to postmsa. Make a digital copy for construction */
if (esl_msafile_Open(NULL, msafile, NULL, eslMSAFILE_UNKNOWN, NULL, &afp)!= eslOK) esl_fatal(failmsg);
if (esl_msafile_Read(afp, &msa) != eslOK) esl_fatal(failmsg);
if ((dmsa = esl_msa_Clone(msa)) == NULL) esl_fatal(failmsg);
if (esl_msa_Digitize(abc, dmsa, NULL) != eslOK) esl_fatal(failmsg);
if (p7_Handmodelmaker(dmsa, NULL, &hmm, &trarr) != eslOK) esl_fatal(failmsg);
for (i = 0; i < dmsa->nseq; i++)
if (p7_trace_Validate(trarr[i], abc, dmsa->ax[i], NULL) != eslOK) esl_fatal(failmsg);
/* The example is contrived such that the traces should give exactly the
* same (text) alignment as the input alignment; no tracedoctoring.
* Not a trivial test; for example, sequence 2 has a B->X->I transition that
* can be problematic to handle.
*/
if (p7_tracealign_MSA(dmsa, trarr, hmm->M, p7_DEFAULT, &postmsa) != eslOK) esl_fatal(failmsg);
for (i = 0; i < msa->nseq; i++)
if (strcmp(msa->aseq[i], postmsa->aseq[i]) != 0) esl_fatal(failmsg);
p7_trace_DestroyArray(trarr, msa->nseq);
p7_hmm_Destroy(hmm);
esl_msa_Destroy(msa);
esl_msa_Destroy(dmsa);
esl_msa_Destroy(postmsa);
esl_msafile_Close(afp);
esl_alphabet_Destroy(abc);
remove(msafile);
return;
}
