/* The "basic" utest is a minimal driver for making a small DNA profile and a small DNA sequence,
* then running Viterbi and Forward. It's useful for dumping DP matrices and profiles for debugging.
*/
static void
utest_basic(ESL_GETOPTS *go)
{
char *query= "# STOCKHOLM 1.0\n\nseq1 GAATTC\nseq2 GAATTC\n//\n";
int fmt = eslMSAFILE_STOCKHOLM;
char *targ = "GAATTC";
ESL_ALPHABET *abc = NULL;
ESL_MSA *msa = NULL;
P7_HMM *hmm = NULL;
P7_PROFILE *gm = NULL;
P7_BG *bg = NULL;
P7_PRIOR *pri = NULL;
ESL_DSQ *dsq = NULL;
P7_GMX *gx = NULL;
P7_TRACE *tr = NULL;
int L = strlen(targ);
float vsc, vsc2, fsc;
if ((abc = esl_alphabet_Create(eslDNA)) == NULL) esl_fatal("failed to create alphabet");
if ((pri = p7_prior_CreateNucleic()) == NULL) esl_fatal("failed to create prior");
if ((msa = esl_msa_CreateFromString(query, fmt)) == NULL) esl_fatal("failed to create MSA");
if (esl_msa_Digitize(abc, msa, NULL) != eslOK) esl_fatal("failed to digitize MSA");
if (p7_Fastmodelmaker(msa, 0.5, NULL, &hmm, NULL) != eslOK) esl_fatal("failed to create GAATTC model");
if (p7_ParameterEstimation(hmm, pri) != eslOK) esl_fatal("failed to parameterize GAATTC model");
if (p7_hmm_SetConsensus(hmm, NULL) != eslOK) esl_fatal("failed to make consensus");
if ((bg = p7_bg_Create(abc)) == NULL) esl_fatal("failed to create DNA null model");
if ((gm = p7_profile_Create(hmm->M, abc)) == NULL) esl_fatal("failed to create GAATTC profile");
if (p7_ProfileConfig(hmm, bg, gm, L, p7_UNILOCAL)!= eslOK) esl_fatal("failed to config profile");
if (p7_profile_Validate(gm, NULL, 0.0001) != eslOK) esl_fatal("whoops, profile is bad!");
if (esl_abc_CreateDsq(abc, targ, &dsq) != eslOK) esl_fatal("failed to create GAATTC digital sequence");
if ((gx = p7_gmx_Create(gm->M, L)) == NULL) esl_fatal("failed to create DP matrix");
if ((tr = p7_trace_Create()) == NULL) esl_fatal("trace creation failed");
p7_GViterbi (dsq, L, gm, gx, &vsc);
if (esl_opt_GetBoolean(go, "-v")) printf("Viterbi score: %.4f\n", vsc);
if (esl_opt_GetBoolean(go, "-v")) p7_gmx_Dump(stdout, gx, p7_DEFAULT);
p7_GTrace (dsq, L, gm, gx, tr);
p7_trace_Score(tr, dsq, gm, &vsc2);
if (esl_opt_GetBoolean(go, "-v")) p7_trace_Dump(stdout, tr, gm, dsq);
if (esl_FCompare(vsc, vsc2, 1e-5) != eslOK) esl_fatal("trace score and Viterbi score don't agree.");
p7_GForward (dsq, L, gm, gx, &fsc);
if (esl_opt_GetBoolean(go, "-v")) printf("Forward score: %.4f\n", fsc);
if (esl_opt_GetBoolean(go, "-v")) p7_gmx_Dump(stdout, gx, p7_DEFAULT);
p7_trace_Destroy(tr);
p7_gmx_Destroy(gx);
free(dsq);
p7_profile_Destroy(gm);
p7_bg_Destroy(bg);
p7_hmm_Destroy(hmm);
esl_msa_Destroy(msa);
p7_prior_Destroy(pri);
esl_alphabet_Destroy(abc);
return;
}
/* tests:
* 1. each sampled trace must validate.
* 2. each trace must be <= viterbi trace score
* 3. in a large # of traces, one is "equal" to the viterbi trace score.
* (this of course is stochastic; but it's true for the particular
* choice of RNG seed used in tests here.)
*/
static void
utest_stotrace(ESL_GETOPTS *go, ESL_RANDOMNESS *rng, ESL_ALPHABET *abc, P7_PROFILE *gm, P7_OPROFILE *om, ESL_DSQ *dsq, int L, int ntrace)
{
P7_GMX *gx = NULL;
P7_OMX *ox = NULL;
P7_TRACE *vtr = NULL;
P7_TRACE *tr = NULL;
char errbuf[eslERRBUFSIZE];
int idx;
float maxsc = -eslINFINITY;
float vsc, sc;
if ((gx = p7_gmx_Create(gm->M, L)) == NULL) esl_fatal("generic DP matrix creation failed");
if ((ox = p7_omx_Create(gm->M, L, L)) == NULL) esl_fatal("optimized DP matrix create failed");
if ((tr = p7_trace_Create()) == NULL) esl_fatal("trace creation failed");
if ((vtr = p7_trace_Create()) == NULL) esl_fatal("trace creation failed");
if (p7_GViterbi(dsq, L, gm, gx, &vsc) != eslOK) esl_fatal("viterbi failed");
if (p7_GTrace (dsq, L, gm, gx, vtr) != eslOK) esl_fatal("viterbi trace failed");
if (p7_Forward (dsq, L, om, ox, NULL) != eslOK) esl_fatal("forward failed");
for (idx = 0; idx < ntrace; idx++)
{
if (p7_StochasticTrace(rng, dsq, L, om, ox, tr) != eslOK) esl_fatal("stochastic trace failed");
if (p7_trace_Validate(tr, abc, dsq, errbuf) != eslOK) esl_fatal("trace invalid:\n%s", errbuf);
if (p7_trace_Score(tr, dsq, gm, &sc) != eslOK) esl_fatal("trace scoring failed");
maxsc = ESL_MAX(sc, maxsc);
if (sc > vsc + 0.001){ /* need a little tolerance of floating point math here */
//p7_trace_Dump(stdout, vtr, gm, dsq);
//p7_trace_Dump(stdout, tr, gm, dsq);
esl_fatal("sampled trace has score > optimal Viterbi path; not possible (%f > %f)", sc, vsc);
}
p7_trace_Reuse(tr);
}
if (esl_FCompare(maxsc, vsc, 0.1) != eslOK) esl_fatal("stochastic trace failed to sample the Viterbi path");
p7_trace_Destroy(tr);
p7_trace_Destroy(vtr);
p7_omx_Destroy(ox);
p7_gmx_Destroy(gx);
}
/* Viterbi validation is done by comparing the returned score
* to the score of the optimal trace. Not foolproof, but catches
* many kinds of errors.
*
* Another check is that the average score should be <= 0,
* since the random sequences are drawn from the null model.
*/
static void
utest_viterbi(ESL_GETOPTS *go, ESL_RANDOMNESS *r, ESL_ALPHABET *abc, P7_BG *bg, P7_PROFILE *gm, int nseq, int L)
{
float avg_sc = 0.;
char errbuf[eslERRBUFSIZE];
ESL_DSQ *dsq = NULL;
P7_GMX *gx = NULL;
P7_TRACE *tr = NULL;
int idx;
float sc1, sc2;
if ((dsq = malloc(sizeof(ESL_DSQ) *(L+2))) == NULL) esl_fatal("malloc failed");
if ((tr = p7_trace_Create()) == NULL) esl_fatal("trace creation failed");
if ((gx = p7_gmx_Create(gm->M, L)) == NULL) esl_fatal("matrix creation failed");
for (idx = 0; idx < nseq; idx++)
{
if (esl_rsq_xfIID(r, bg->f, abc->K, L, dsq) != eslOK) esl_fatal("seq generation failed");
if (p7_GViterbi(dsq, L, gm, gx, &sc1) != eslOK) esl_fatal("viterbi failed");
if (p7_GTrace (dsq, L, gm, gx, tr) != eslOK) esl_fatal("trace failed");
if (p7_trace_Validate(tr, abc, dsq, errbuf) != eslOK) esl_fatal("trace invalid:\n%s", errbuf);
if (p7_trace_Score(tr, dsq, gm, &sc2) != eslOK) esl_fatal("trace score failed");
if (esl_FCompare(sc1, sc2, 1e-6) != eslOK) esl_fatal("Trace score != Viterbi score");
if (p7_bg_NullOne(bg, dsq, L, &sc2) != eslOK) esl_fatal("null score failed");
avg_sc += (sc1 - sc2);
if (esl_opt_GetBoolean(go, "--vv"))
printf("utest_viterbi: Viterbi score: %.4f (null %.4f) (total so far: %.4f)\n", sc1, sc2, avg_sc);
p7_trace_Reuse(tr);
}
avg_sc /= (float) nseq;
if (avg_sc > 0.) esl_fatal("Viterbi scores have positive expectation (%f nats)", avg_sc);
p7_gmx_Destroy(gx);
p7_trace_Destroy(tr);
free(dsq);
return;
}
int
main(int argc, char **argv)
{
ESL_GETOPTS *go = esl_getopts_CreateDefaultApp(options, 2, argc, argv, banner, usage);
char *hmmfile = esl_opt_GetArg(go, 1);
char *seqfile = esl_opt_GetArg(go, 2);
ESL_ALPHABET *abc = NULL;
P7_HMMFILE *hfp = NULL;
P7_HMM *hmm = NULL;
P7_BG *bg = NULL;
P7_PROFILE *gm = NULL;
P7_GMX *fwd = NULL;
ESL_SQ *sq = NULL;
ESL_SQFILE *sqfp = NULL;
P7_TRACE *tr = NULL;
int format = eslSQFILE_UNKNOWN;
char errbuf[eslERRBUFSIZE];
float sc;
int d;
int status;
/* Read in one HMM */
if (p7_hmmfile_Open(hmmfile, NULL, &hfp) != eslOK) p7_Fail("Failed to open HMM file %s", hmmfile);
if (p7_hmmfile_Read(hfp, &abc, &hmm) != eslOK) p7_Fail("Failed to read HMM");
p7_hmmfile_Close(hfp);
/* Read in one sequence */
sq = esl_sq_CreateDigital(abc);
status = esl_sqfile_Open(seqfile, format, NULL, &sqfp);
if (status == eslENOTFOUND) p7_Fail("No such file.");
else if (status == eslEFORMAT) p7_Fail("Format unrecognized.");
else if (status == eslEINVAL) p7_Fail("Can't autodetect stdin or .gz.");
else if (status != eslOK) p7_Fail("Open failed, code %d.", status);
if (esl_sqio_Read(sqfp, sq) != eslOK) p7_Fail("Failed to read sequence");
esl_sqfile_Close(sqfp);
/* Configure a profile from the HMM */
bg = p7_bg_Create(abc);
p7_bg_SetLength(bg, sq->n);
gm = p7_profile_Create(hmm->M, abc);
p7_ProfileConfig(hmm, bg, gm, sq->n, p7_LOCAL);
/* Allocate matrix and a trace */
fwd = p7_gmx_Create(gm->M, sq->n);
tr = p7_trace_Create();
/* Run Viterbi; do traceback */
p7_GViterbi (sq->dsq, sq->n, gm, fwd, &sc);
p7_GTrace (sq->dsq, sq->n, gm, fwd, tr);
/* Dump and validate the trace. */
p7_trace_Dump(stdout, tr, gm, sq->dsq);
if (p7_trace_Validate(tr, abc, sq->dsq, errbuf) != eslOK) p7_Die("trace fails validation:\n%s\n", errbuf);
/* Domain info in the trace. */
p7_trace_Index(tr);
printf("# Viterbi: %d domains : ", tr->ndom);
for (d = 0; d < tr->ndom; d++) printf("%6d %6d %6d %6d ", tr->sqfrom[d], tr->sqto[d], tr->hmmfrom[d], tr->hmmto[d]);
printf("\n");
/* Cleanup */
p7_trace_Destroy(tr);
p7_gmx_Destroy(fwd);
p7_profile_Destroy(gm);
p7_bg_Destroy(bg);
p7_hmm_Destroy(hmm);
esl_sq_Destroy(sq);
esl_alphabet_Destroy(abc);
esl_getopts_Destroy(go);
return 0;
}
int
main(int argc, char **argv)
{
ESL_GETOPTS *go = p7_CreateDefaultApp(options, 2, argc, argv, banner, usage);
char *hmmfile = esl_opt_GetArg(go, 1);
char *seqfile = esl_opt_GetArg(go, 2);
ESL_ALPHABET *abc = NULL;
P7_HMMFILE *hfp = NULL;
P7_HMM *hmm = NULL;
P7_BG *bg = NULL;
P7_PROFILE *gm = NULL;
P7_GMX *gx1 = NULL;
P7_GMX *gx2 = NULL;
ESL_SQ *sq = NULL;
ESL_SQFILE *sqfp = NULL;
P7_TRACE *tr = NULL;
int format = eslSQFILE_UNKNOWN;
char errbuf[eslERRBUFSIZE];
float fsc, bsc, vsc;
float accscore;
int status;
/* Read in one HMM */
if (p7_hmmfile_OpenE(hmmfile, NULL, &hfp, NULL) != eslOK) p7_Fail("Failed to open HMM file %s", hmmfile);
if (p7_hmmfile_Read(hfp, &abc, &hmm) != eslOK) p7_Fail("Failed to read HMM");
p7_hmmfile_Close(hfp);
/* Read in one sequence */
sq = esl_sq_CreateDigital(abc);
status = esl_sqfile_OpenDigital(abc, seqfile, format, NULL, &sqfp);
if (status == eslENOTFOUND) p7_Fail("No such file.");
else if (status == eslEFORMAT) p7_Fail("Format unrecognized.");
else if (status == eslEINVAL) p7_Fail("Can't autodetect stdin or .gz.");
else if (status != eslOK) p7_Fail("Open failed, code %d.", status);
if (esl_sqio_Read(sqfp, sq) != eslOK) p7_Fail("Failed to read sequence");
esl_sqfile_Close(sqfp);
/* Configure a profile from the HMM */
bg = p7_bg_Create(abc);
p7_bg_SetLength(bg, sq->n);
gm = p7_profile_Create(hmm->M, abc);
p7_ProfileConfig(hmm, bg, gm, sq->n, p7_LOCAL); /* multihit local: H3 default */
/* Allocations */
gx1 = p7_gmx_Create(gm->M, sq->n);
gx2 = p7_gmx_Create(gm->M, sq->n);
tr = p7_trace_CreateWithPP();
p7_FLogsumInit();
/* Run Forward, Backward; do OA fill and trace */
p7_GForward (sq->dsq, sq->n, gm, gx1, &fsc);
p7_GBackward(sq->dsq, sq->n, gm, gx2, &bsc);
p7_GDecoding(gm, gx1, gx2, gx2); /* <gx2> is now the posterior decoding matrix */
p7_GOptimalAccuracy(gm, gx2, gx1, &accscore); /* <gx1> is now the OA matrix */
p7_GOATrace(gm, gx2, gx1, tr);
if (esl_opt_GetBoolean(go, "-d")) p7_gmx_Dump(stdout, gx2, p7_DEFAULT);
if (esl_opt_GetBoolean(go, "-m")) p7_gmx_Dump(stdout, gx1, p7_DEFAULT);
p7_trace_Dump(stdout, tr, gm, sq->dsq);
if (p7_trace_Validate(tr, abc, sq->dsq, errbuf) != eslOK) p7_Die("trace fails validation:\n%s\n", errbuf);
printf("fwd = %.4f nats\n", fsc);
printf("bck = %.4f nats\n", bsc);
printf("acc = %.4f (%.2f%%)\n", accscore, accscore * 100. / (float) sq->n);
p7_trace_Reuse(tr);
p7_GViterbi(sq->dsq, sq->n, gm, gx1, &vsc);
p7_GTrace (sq->dsq, sq->n, gm, gx1, tr);
p7_trace_SetPP(tr, gx2);
p7_trace_Dump(stdout, tr, gm, sq->dsq);
printf("vit = %.4f nats\n", vsc);
printf("acc = %.4f\n", p7_trace_GetExpectedAccuracy(tr));
/* Cleanup */
esl_sq_Destroy(sq);
p7_trace_Destroy(tr);
p7_gmx_Destroy(gx1);
p7_gmx_Destroy(gx2);
p7_profile_Destroy(gm);
p7_bg_Destroy(bg);
p7_hmm_Destroy(hmm);
esl_alphabet_Destroy(abc);
esl_getopts_Destroy(go);
return 0;
}