static void
emit_alignment(ESL_GETOPTS *go, FILE *ofp, int outfmt, ESL_RANDOMNESS *r, P7_HMM *hmm)
{
ESL_MSA *msa = NULL;
ESL_SQ **sq = NULL;
P7_TRACE **tr = NULL;
int N = esl_opt_GetInteger(go, "-N");
int optflags = p7_ALL_CONSENSUS_COLS;
int i;
if ((tr = malloc(sizeof(P7_TRACE *) * N)) == NULL) esl_fatal("failed to allocate trace array");
if ((sq = malloc(sizeof(ESL_SQ *) * N)) == NULL) esl_fatal("failed to allocate seq array");
for (i = 0; i < N; i++)
{
if ((sq[i] = esl_sq_CreateDigital(hmm->abc)) == NULL) esl_fatal("failed to allocate seq");
if ((tr[i] = p7_trace_Create()) == NULL) esl_fatal("failed to allocate trace");
}
for (i = 0; i < N; i++)
{
if (p7_CoreEmit(r, hmm, sq[i], tr[i]) != eslOK) esl_fatal("Failed to emit sequence");
if (esl_sq_FormatName(sq[i], "%s-sample%d", hmm->name, i+1) != eslOK) esl_fatal("Failed to set sequence name\n");
}
p7_tracealign_Seqs(sq, tr, N, hmm->M, optflags, hmm, &msa);
eslx_msafile_Write(ofp, msa, outfmt);
for (i = 0; i < N; i++) p7_trace_Destroy(tr[i]); free(tr);
for (i = 0; i < N; i++) esl_sq_Destroy(sq[i]); free(sq);
esl_msa_Destroy(msa);
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 *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 (p7_GViterbi(dsq, L, gm, gx, &vsc) != eslOK) esl_fatal("viterbi 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) esl_fatal("sampled trace has score > optimal Viterbi path; not possible");
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_omx_Destroy(ox);
p7_gmx_Destroy(gx);
}
int
main(int argc, char **argv)
{
ESL_GETOPTS *go = p7_CreateDefaultApp(options, 1, argc, argv, banner, usage);
ESL_RANDOMNESS *rng = esl_randomness_CreateFast(esl_opt_GetInteger(go, "-s"));
char *hmmfile = esl_opt_GetArg(go, 1);
int L = esl_opt_GetInteger(go, "-L");
int N = esl_opt_GetInteger(go, "-N");
ESL_ALPHABET *abc = NULL;
P7_HMMFILE *hfp = NULL;
P7_HMM *hmm = NULL;
P7_BG *bg = NULL;
P7_PROFILE *gm = NULL;
P7_TRACE *tr = p7_trace_Create();
ESL_SQ *sq = NULL;
char errbuf[eslERRBUFSIZE];
int i;
int status;
status = p7_hmmfile_OpenE(hmmfile, NULL, &hfp, errbuf);
if (status == eslENOTFOUND) p7_Fail("File existence/permissions problem in trying to open HMM file %s.\n%s\n", hmmfile, errbuf);
else if (status == eslEFORMAT) p7_Fail("File format problem in trying to open HMM file %s.\n%s\n", hmmfile, errbuf);
else if (status != eslOK) p7_Fail("Unexpected error %d in opening HMM file %s.\n%s\n", status, hmmfile, errbuf);
status = p7_hmmfile_Read(hfp, &abc, &hmm);
if (status == eslEFORMAT) p7_Fail("Bad file format in HMM file %s:\n%s\n", hfp->fname, hfp->errbuf);
else if (status == eslEINCOMPAT) p7_Fail("HMM in %s is not in the expected %s alphabet\n", hfp->fname, esl_abc_DecodeType(abc->type));
else if (status == eslEOF) p7_Fail("Empty HMM file %s? No HMM data found.\n", hfp->fname);
else if (status != eslOK) p7_Fail("Unexpected error in reading HMMs from %s\n", hfp->fname);
p7_hmmfile_Close(hfp);
bg = p7_bg_Create(abc); p7_bg_SetLength(bg, L);
gm = p7_profile_Create(hmm->M, abc); p7_ProfileConfig(hmm, bg, gm, L, p7_LOCAL);
sq = esl_sq_CreateDigital(abc);
for (i = 0; i < N; i++)
{
p7_ProfileEmit(rng, hmm, gm, bg, sq, tr);
esl_sq_FormatName(sq, "%s-sample%d", hmm->name, i);
esl_sqio_Write(stdout, sq, eslSQFILE_FASTA, FALSE);
if (p7_trace_Validate(tr, abc, sq->dsq, errbuf) != eslOK) esl_fatal(errbuf);
esl_sq_Reuse(sq);
p7_trace_Reuse(tr);
}
esl_sq_Destroy(sq);
p7_trace_Destroy(tr);
p7_profile_Destroy(gm);
p7_bg_Destroy(bg);
p7_hmm_Destroy(hmm);
esl_alphabet_Destroy(abc);
esl_randomness_Destroy(rng);
esl_getopts_Destroy(go);
return 0;
}
/* 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;
}
int
main(int argc, char **argv)
{
char *hmmfile = argv[1]; /* name of HMM file to read one HMM from */
ESL_ALPHABET *abc = NULL; /* sequence alphabet */
ESL_RANDOMNESS *r = NULL; /* source of randomness */
P7_HMMFILE *hfp = NULL; /* open hmmfile */
P7_HMM *hmm = NULL; /* HMM to emit from */
P7_PROFILE *gm = NULL; /* profile HMM (scores) */
P7_BG *bg = NULL; /* null model */
P7_TRACE *tr = NULL; /* sampled trace */
ESL_SQ *sq = NULL; /* sampled digital sequence */
int n = 1000;
int counts[p7T_NSTATETYPES];
int i;
float sc;
float nullsc;
double bitscore;
r = esl_randomness_CreateFast(0);
tr = p7_trace_Create();
if (p7_hmmfile_OpenE(hmmfile, NULL, &hfp, NULL) != eslOK) p7_Fail("failed to open %s", hmmfile);
if (p7_hmmfile_Read(hfp, &abc, &hmm) != eslOK) p7_Fail("failed to read HMM");
sq = esl_sq_CreateDigital(abc);
bg = p7_bg_Create(abc);
gm = p7_profile_Create(hmm->M, abc);
p7_ProfileConfig(hmm, bg, gm, sq->n, p7_LOCAL);
for (i = 0; i < n; i++)
{
p7_ProfileEmit(r, hmm, gm, bg, sq, tr);
p7_trace_GetStateUseCounts(tr, counts);
p7_ReconfigLength(gm, sq->n);
p7_bg_SetLength(bg, sq->n);
p7_trace_Score(tr, sq->dsq, gm, &sc);
p7_bg_NullOne (bg, sq->dsq, sq->n, &nullsc);
bitscore = (sc - nullsc)/ eslCONST_LOG2;
printf("%d %8.4f\n",
counts[p7T_M] + (counts[p7T_I] + counts[p7T_D])/2,
bitscore);
}
p7_profile_Destroy(gm);
esl_sq_Destroy(sq);
p7_trace_Destroy(tr);
esl_randomness_Destroy(r);
esl_alphabet_Destroy(abc);
p7_hmmfile_Close(hfp);
p7_hmm_Destroy(hmm);
return eslOK;
}
static void
emit_sequences(ESL_GETOPTS *go, FILE *ofp, int outfmt, ESL_RANDOMNESS *r, P7_HMM *hmm)
{
ESL_SQ *sq = NULL;
P7_TRACE *tr = NULL;
P7_BG *bg = NULL;
P7_PROFILE *gm = NULL;
int do_profile = esl_opt_GetBoolean(go, "-p");
int N = esl_opt_GetInteger(go, "-N");
int L = esl_opt_GetInteger(go, "-L");
int mode = p7_LOCAL;
int nseq;
int status;
if (esl_opt_GetBoolean(go, "--local")) mode = p7_LOCAL;
else if (esl_opt_GetBoolean(go, "--unilocal")) mode = p7_UNILOCAL;
else if (esl_opt_GetBoolean(go, "--glocal")) mode = p7_GLOCAL;
else if (esl_opt_GetBoolean(go, "--uniglocal")) mode = p7_UNIGLOCAL;
if ((sq = esl_sq_CreateDigital(hmm->abc)) == NULL) esl_fatal("failed to allocate sequence");
if ((tr = p7_trace_Create()) == NULL) esl_fatal("failed to allocate trace");
if ((bg = p7_bg_Create(hmm->abc)) == NULL) esl_fatal("failed to create null model");
if ((gm = p7_profile_Create(hmm->M, hmm->abc)) == NULL) esl_fatal("failed to create profile");
if (p7_ProfileConfig(hmm, bg, gm, L, mode) != eslOK) esl_fatal("failed to configure profile");
if (p7_bg_SetLength(bg, L) != eslOK) esl_fatal("failed to reconfig null model length");
if (p7_hmm_Validate (hmm, NULL, 0.0001) != eslOK) esl_fatal("whoops, HMM is bad!");
if (p7_profile_Validate(gm, NULL, 0.0001) != eslOK) esl_fatal("whoops, profile is bad!");
for (nseq = 1; nseq <= N; nseq++)
{
if (do_profile) status = p7_ProfileEmit(r, hmm, gm, bg, sq, tr);
else status = p7_CoreEmit (r, hmm, sq, tr);
if (status) esl_fatal("Failed to emit sequence\n");
status = esl_sq_FormatName(sq, "%s-sample%d", hmm->name, nseq);
if (status) esl_fatal("Failed to set sequence name\n");
status = esl_sqio_Write(ofp, sq, outfmt, FALSE);
if (status != eslOK) esl_fatal("Failed to write sequence\n");
p7_trace_Reuse(tr);
esl_sq_Reuse(sq);
}
esl_sq_Destroy(sq);
p7_trace_Destroy(tr);
p7_bg_Destroy(bg);
p7_profile_Destroy(gm);
return;
}
/* Function: p7_SingleBuilder()
* Synopsis: Build a new HMM from a single sequence.
*
* Purpose: Take the sequence <sq> and a build configuration <bld>, and
* build a new HMM.
*
* The single sequence scoring system in the <bld>
* configuration must have been previously initialized by
* <p7_builder_SetScoreSystem()>.
*
* Args: bld - build configuration
* sq - query sequence
* bg - null model (needed to paramaterize insert emission probs)
* opt_hmm - optRETURN: new HMM
* opt_gm - optRETURN: profile corresponding to <hmm>
* opt_om - optRETURN: optimized profile corresponding to <gm>
*
* Returns: <eslOK> on success.
*
* Throws: <eslEMEM> on allocation error.
* <eslEINVAL> if <bld> isn't properly configured somehow.
*/
int
p7_SingleBuilder(P7_BUILDER *bld, ESL_SQ *sq, P7_BG *bg, P7_HMM **opt_hmm,
P7_TRACE **opt_tr, P7_PROFILE **opt_gm, P7_OPROFILE **opt_om)
{
P7_HMM *hmm = NULL;
P7_TRACE *tr = NULL;
int k;
int status;
bld->errbuf[0] = '\0';
if (! bld->Q) ESL_XEXCEPTION(eslEINVAL, "score system not initialized");
if ((status = p7_Seqmodel(bld->abc, sq->dsq, sq->n, sq->name, bld->Q, bg->f, bld->popen, bld->pextend, &hmm)) != eslOK) goto ERROR;
if ((status = p7_hmm_SetComposition(hmm)) != eslOK) goto ERROR;
if ((status = p7_hmm_SetConsensus(hmm, sq)) != eslOK) goto ERROR;
if ((status = calibrate(bld, hmm, bg, opt_gm, opt_om)) != eslOK) goto ERROR;
if ( bld->abc->type == eslDNA || bld->abc->type == eslRNA ) {
if (bld->w_len > 0) hmm->max_length = bld->w_len;
else if (bld->w_beta == 0.0) hmm->max_length = hmm->M *4;
else if ( (status = p7_Builder_MaxLength(hmm, bld->w_beta)) != eslOK) goto ERROR;
}
/* build a faux trace: relative to core model (B->M_1..M_L->E) */
if (opt_tr != NULL)
{
if ((tr = p7_trace_Create()) == NULL) goto ERROR;
if ((status = p7_trace_Append(tr, p7T_B, 0, 0)) != eslOK) goto ERROR;
for (k = 1; k <= sq->n; k++)
if ((status = p7_trace_Append(tr, p7T_M, k, k)) != eslOK) goto ERROR;
if ((status = p7_trace_Append(tr, p7T_E, 0, 0)) != eslOK) goto ERROR;
tr->M = sq->n;
tr->L = sq->n;
}
/* note that <opt_gm> and <opt_om> were already set by calibrate() call above. */
if (opt_hmm != NULL) *opt_hmm = hmm; else p7_hmm_Destroy(hmm);
if (opt_tr != NULL) *opt_tr = tr;
return eslOK;
ERROR:
p7_hmm_Destroy(hmm);
if (tr != NULL) p7_trace_Destroy(tr);
if (opt_gm != NULL) p7_profile_Destroy(*opt_gm);
if (opt_om != NULL) p7_oprofile_Destroy(*opt_om);
return status;
}
/* 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;
}
/* The "generation" test scores sequences generated by the same profile.
* Each Viterbi and Forward score should be >= the trace score of the emitted seq.
* The expectation of Forward scores should be positive.
*/
static void
utest_generation(ESL_GETOPTS *go, ESL_RANDOMNESS *r, ESL_ALPHABET *abc,
P7_PROFILE *gm, P7_HMM *hmm, P7_BG *bg, int nseq)
{
ESL_SQ *sq = esl_sq_CreateDigital(abc);
P7_GMX *gx = p7_gmx_Create(gm->M, 100);
P7_TRACE *tr = p7_trace_Create();
float vsc, fsc, nullsc, tracesc;
float avg_fsc;
int idx;
avg_fsc = 0.0;
for (idx = 0; idx < nseq; idx++)
{
if (p7_ProfileEmit(r, hmm, gm, bg, sq, tr) != eslOK) esl_fatal("profile emission failed");
if (p7_gmx_GrowTo(gx, gm->M, sq->n) != eslOK) esl_fatal("failed to reallocate gmx");
if (p7_GViterbi(sq->dsq, sq->n, gm, gx, &vsc) != eslOK) esl_fatal("viterbi failed");
if (p7_GForward(sq->dsq, sq->n, gm, gx, &fsc) != eslOK) esl_fatal("forward failed");
if (p7_trace_Score(tr, sq->dsq, gm, &tracesc) != eslOK) esl_fatal("trace score failed");
if (p7_bg_NullOne(bg, sq->dsq, sq->n, &nullsc) != eslOK) esl_fatal("null score failed");
if (vsc < tracesc) esl_fatal("viterbi score is less than trace");
if (fsc < tracesc) esl_fatal("forward score is less than trace");
if (vsc > fsc) esl_fatal("viterbi score is greater than forward");
if (esl_opt_GetBoolean(go, "--vv"))
printf("generated: len=%d v=%8.4f f=%8.4f t=%8.4f\n", (int) sq->n, vsc, fsc, tracesc);
avg_fsc += (fsc - nullsc);
}
avg_fsc /= (float) nseq;
if (avg_fsc < 0.) esl_fatal("generation: Forward scores have negative expectation (%f nats)", avg_fsc);
p7_gmx_Destroy(gx);
p7_trace_Destroy(tr);
esl_sq_Destroy(sq);
}
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;
}
/* profile_local_endpoints()
*
* Purpose: Wrapper around <p7_ProfileEmit()>, sampling a local
* alignment fragment from the profile's probabilistic model
* (which may be the implicit model of HMMER3, or the
* Plan7 model of HMMER2), and reporting coordinates
* of the fragment w.r.t. both model and sequence.
*
* To simplify the implementation, the profile must be in
* <p7_UNILOCAL> mode, not <p7_LOCAL> mode, so we know we
* only have to deal with a single hit per sampled
* sequence.
*
* We want <i1..i2> to be relative to the sequence coords
* of a complete (global) sampled sequence that we could
* have sampled this local alignment from; but the <i1..i2>
* we initially get are relative to our profile-sampled
* trace, so they are offset both by N-generated residues
* that occur in the profile and by residues that the
* profile's local entry skipped. To translate from
* profile/sequence coords to core model/sequence coords,
* we use rejection sampling: sample traces from the core
* model until we find one that uses the same statetypes
* at *initial* entry/exit points <k1>,<k2>, then use
* that sample's sequence to determine offsets and correct
* <i1..i2> reference frame.
*
* Local alignment endpoints are defined to be
* match-delimited. However, an H3 model allows exit on
* either a D or M state. Thus, the initially sampled end
* point k2 may need to be rolled back to last M state, to
* satisfy local alignment endpoint definition. Entries are
* not a problem; both H2 and H3 profiles can only enter on
* a M state. (This rollback has to occur after we've
* matched a core trace to the profile trace to determine
* i offsets.)
*
* Then, sampling from both the core model and the profile
* in the same routine introduces a complication:
* conceivably, profile configuration alters the transition
* probabilities in the core model (by adding <M->E>
* transitions and renormalizing the M transition
* distributions, for example; H2 configuration does this,
* though H3 does not). So you can't <CoreSample()> the
* <gm->hmm> safely. To avoid such things, the caller
* provides a clean copy of the core model in <core>.
*
* i endpoints are normalized/discretized to 1..<Lbins>, so
* we can collate i statistics from sampled sequences of
* varying L. Note this causes discretization artifacts,
* leading to underrepresentation of j=M and
* overrepresentation of i=1.
*
* Returns: <eslOK> on success; returns normalized sequence coords in
* <*ret_i1> and <*ret_i2>, and the model entry/exit coords
* in <*ret_k1> and <*ret_k2>.
*
* Xref: STL11/142-143
*/
static int
profile_local_endpoints(ESL_RANDOMNESS *r, P7_HMM *core, P7_PROFILE *gm, ESL_SQ *sq, P7_TRACE *tr, int Lbins,
int *ret_i1, int *ret_i2, int *ret_k1, int *ret_k2)
{
int status;
int i1,i2;
int k1,k2;
int t1,t2; /* entry/exit positions in local trace, tr */
int tg1, tg2; /* entry/exit positions in global trace, tr2 */
int tpos;
int nterm, cterm; /* offsets at N, C terminus. */
int L; /* inferred length from 3-part patching */
ESL_SQ *sq2 = NULL;
P7_TRACE *tr2 = NULL;
int failsafe = 0;
if (gm->mode != p7_UNILOCAL) ESL_XEXCEPTION(eslEINVAL, "profile must be unilocal");
if ((sq2 = esl_sq_CreateDigital(gm->abc)) == NULL) { status = eslEMEM; goto ERROR; }
if ((tr = p7_trace_Create()) == NULL) { status = eslEMEM; goto ERROR; }
/* sample local alignment from the implicit model */
if (gm->h2_mode) {
if ((status = p7_H2_ProfileEmit(r, gm, sq, tr)) != eslOK) goto ERROR;
} else {
if ((status = p7_ProfileEmit(r, gm, sq, tr)) != eslOK) goto ERROR;
}
/* Get initial trace coords */
for (tpos = 0; tpos < tr->N; tpos++) if (tr->st[tpos] == p7T_B) { t1 = tpos+1; break; }
for (tpos = tr->N-1; tpos >= 0; tpos--) if (tr->st[tpos] == p7T_E) { t2 = tpos-1; break; }
/* Match a core trace to this local trace by rejection sampling;
* this is to let us calculate sequence offsets; see comments above in preamble
*/
do {
if (failsafe++ == 100000) ESL_XEXCEPTION(eslENOHALT, "failed to match core,local traces in %d tries\n", failsafe);
if ((status = p7_CoreEmit(r, core, sq2, tr2)) != eslOK) goto ERROR;
for (tpos = 0; tpos < tr2->N; tpos++)
if (tr2->k[tpos] == tr->k[t1]) { tg1 = tpos; break; }
for (tpos = tr2->N-1; tpos >= 0; tpos--)
if (tr2->k[tpos] == tr->k[t2]) { tg2 = tpos; break; }
} while (tr2->st[tg1] != tr->st[t1] && tr2->st[tg2] != tr->st[t2]);
/* tg1..tg2 in core trace is now matched to t1..t2 in the profile trace.
* Calculate # of residues preceding tg1 and following tg2 in the core trace.
* A core trace can only generate residues from M or I states.
*/
for (nterm = 0, tpos = 0; tpos < tg1; tpos++)
if (tr2->st[tpos] == p7T_M || tr2->st[tpos] == p7T_I) nterm++;
for (cterm = 0, tpos = tr2->N-1; tpos > tg2; tpos--)
if (tr2->st[tpos] == p7T_M || tr2->st[tpos] == p7T_I) cterm++;
/* rectify the t2 endpoint, rolling back any trailing D path
*/
for (; t2 >= 0; t2--) if (tr->st[t2] == p7T_M) break;
if (t2 < t1) ESL_XEXCEPTION(eslEINCONCEIVABLE, "this only happens on an all-D path through profile");
/* determine initial endpoint coords from t1 and t2 */
i1 = tr->i[t1]; i2 = tr->i[t2];
k1 = tr->k[t1]; k2 = tr->k[t2];
/* offset the i coords. */
L = (i2-i1+1) + nterm + cterm;
i2 = (i2-i1+1) + nterm;
i1 = nterm+1;
/* normalize the i coords into range 1..Lbins, instead of 1..L */
i1 = ((i1-1) * Lbins / L) + 1;
i2 = ((i2-1) * Lbins / L) + 1;
*ret_i1 = i1;
*ret_i2 = i2;
*ret_k1 = k1;
*ret_k2 = k2;
p7_trace_Destroy(tr2);
esl_sq_Destroy(sq2);
return eslOK;
ERROR:
if (sq2 != NULL) esl_sq_Destroy(sq2);
if (tr2 != NULL) p7_trace_Destroy(tr2);
*ret_i1 = 0.;
*ret_i2 = 0.;
*ret_k1 = 0;
*ret_k2 = 0;
return status;
}
int
main(int argc, char **argv)
{
ESL_ALPHABET *abc = NULL; /* sequence alphabet */
ESL_GETOPTS *go = NULL; /* command line processing */
ESL_RANDOMNESS *r = NULL; /* source of randomness */
P7_HMM *hmm = NULL; /* sampled HMM to emit from */
P7_HMM *core = NULL; /* safe copy of the HMM, before config */
P7_BG *bg = NULL; /* null model */
ESL_SQ *sq = NULL; /* sampled sequence */
P7_TRACE *tr = NULL; /* sampled trace */
P7_PROFILE *gm = NULL; /* profile */
int i,j;
int i1,i2;
int k1,k2;
int iseq;
FILE *fp = NULL;
double expected;
int do_ilocal;
char *hmmfile = NULL;
int nseq;
int do_swlike;
int do_ungapped;
int L;
int M;
int do_h2;
char *ipsfile = NULL;
char *kpsfile = NULL;
ESL_DMATRIX *imx = NULL;
ESL_DMATRIX *kmx = NULL;
ESL_DMATRIX *iref = NULL; /* reference matrix: expected i distribution under ideality */
int Lbins;
int status;
char errbuf[eslERRBUFSIZE];
/*****************************************************************
* Parse the command line
*****************************************************************/
go = esl_getopts_Create(options);
if (esl_opt_ProcessCmdline(go, argc, argv) != eslOK) esl_fatal("Failed to parse command line: %s\n", go->errbuf);
if (esl_opt_VerifyConfig(go) != eslOK) esl_fatal("Failed to parse command line: %s\n", go->errbuf);
if (esl_opt_GetBoolean(go, "-h") == TRUE) {
puts(usage);
puts("\n where options are:\n");
esl_opt_DisplayHelp(stdout, go, 0, 2, 80); /* 0=all docgroups; 2 = indentation; 80=textwidth*/
return eslOK;
}
do_ilocal = esl_opt_GetBoolean(go, "-i");
hmmfile = esl_opt_GetString (go, "-m");
nseq = esl_opt_GetInteger(go, "-n");
do_swlike = esl_opt_GetBoolean(go, "-s");
do_ungapped = esl_opt_GetBoolean(go, "-u");
L = esl_opt_GetInteger(go, "-L");
M = esl_opt_GetInteger(go, "-M");
do_h2 = esl_opt_GetBoolean(go, "-2");
ipsfile = esl_opt_GetString (go, "--ips");
kpsfile = esl_opt_GetString (go, "--kps");
if (esl_opt_ArgNumber(go) != 0) {
puts("Incorrect number of command line arguments.");
printf("Usage: %s [options]\n", argv[0]);
return eslFAIL;
}
r = esl_randomness_CreateFast(0);
if (hmmfile != NULL)
{ /* Read the HMM (and get alphabet from it) */
P7_HMMFILE *hfp = NULL;
status = p7_hmmfile_OpenE(hmmfile, NULL, &hfp, errbuf);
if (status == eslENOTFOUND) p7_Fail("File existence/permissions problem in trying to open HMM file %s.\n%s\n", hmmfile, errbuf);
else if (status == eslEFORMAT) p7_Fail("File format problem in trying to open HMM file %s.\n%s\n", hmmfile, errbuf);
else if (status != eslOK) p7_Fail("Unexpected error %d in opening HMM file %s.\n%s\n", status, hmmfile, errbuf);
if ((status = p7_hmmfile_Read(hfp, &abc, &hmm)) != eslOK) {
if (status == eslEOD) esl_fatal("read failed, HMM file %s may be truncated?", hmmfile);
else if (status == eslEFORMAT) esl_fatal("bad file format in HMM file %s", hmmfile);
else if (status == eslEINCOMPAT) esl_fatal("HMM file %s contains different alphabets", hmmfile);
else esl_fatal("Unexpected error in reading HMMs");
}
M = hmm->M;
p7_hmmfile_Close(hfp);
}
else
{ /* Or sample the HMM (create alphabet first) */
abc = esl_alphabet_Create(eslAMINO);
if (do_ungapped) p7_hmm_SampleUngapped(r, M, abc, &hmm);
else if (do_swlike) p7_hmm_SampleUniform (r, M, abc, 0.05, 0.5, 0.05, 0.2, &hmm); /* tmi, tii, tmd, tdd */
else p7_hmm_Sample (r, M, abc, &hmm);
}
Lbins = M;
imx = esl_dmatrix_Create(Lbins, Lbins);
iref = esl_dmatrix_Create(Lbins, Lbins);
kmx = esl_dmatrix_Create(M, M);
esl_dmatrix_SetZero(imx);
esl_dmatrix_SetZero(iref);
esl_dmatrix_SetZero(kmx);
tr = p7_trace_Create();
sq = esl_sq_CreateDigital(abc);
bg = p7_bg_Create(abc);
core = p7_hmm_Clone(hmm);
if (do_h2) {
gm = p7_profile_Create(hmm->M, abc);
p7_H2_ProfileConfig(hmm, bg, gm, p7_UNILOCAL);
} else {
gm = p7_profile_Create(hmm->M, abc);
p7_ProfileConfig(hmm, bg, gm, L, p7_UNILOCAL);
if (p7_hmm_Validate (hmm, NULL, 0.0001) != eslOK) esl_fatal("whoops, HMM is bad!");
if (p7_profile_Validate(gm, NULL, 0.0001) != eslOK) esl_fatal("whoops, profile is bad!");
}
/* Sample endpoints.
* Also sample an ideal reference distribution for i endpoints. i
* endpoints are prone to discretization artifacts, when emitted
* sequences have varying lengths. Taking log odds w.r.t. an ideal
* reference that is subject to the same discretization artifacts
* cancels out the effect.
*/
for (iseq = 0; iseq < nseq; iseq++)
{
if (do_ilocal) ideal_local_endpoints (r, core, sq, tr, Lbins, &i1, &i2, &k1, &k2);
else profile_local_endpoints(r, core, gm, sq, tr, Lbins, &i1, &i2, &k1, &k2);
imx->mx[i1-1][i2-1] += 1.;
kmx->mx[k1-1][k2-1] += 1.;
/* reference distribution for i */
ideal_local_endpoints (r, core, sq, tr, Lbins, &i1, &i2, &k1, &k2);
iref->mx[i1-1][i2-1] += 1.;
}
/* Adjust both mx's to log_2(obs/exp) ratio */
printf("Before normalization/log-odds:\n");
printf(" i matrix values range from %f to %f\n", dmx_upper_min(imx), dmx_upper_max(imx));
printf(" k matrix values range from %f to %f\n", dmx_upper_min(kmx), dmx_upper_max(kmx));
printf("iref matrix values range from %f to %f\n", dmx_upper_min(iref), dmx_upper_max(iref));
expected = (double) nseq * 2. / (double) (M*(M+1));
for (i = 0; i < kmx->m; i++)
for (j = i; j < kmx->n; j++)
kmx->mx[i][j] = log(kmx->mx[i][j] / expected) / log(2.0);
for (i = 0; i < imx->m; i++)
for (j = i; j < imx->m; j++)
if (iref->mx[i][j] == 0. && imx->mx[i][j] == 0.)
imx->mx[i][j] = 0.;
else if (iref->mx[i][j] == 0.)
imx->mx[i][j] = eslINFINITY;
else if (imx->mx[i][j] == 0.)
imx->mx[i][j] = -eslINFINITY;
else
imx->mx[i][j] = log(imx->mx[i][j] / iref->mx[i][j]) / log(2.0);
/* Print ps files */
if (kpsfile != NULL) {
if ((fp = fopen(kpsfile, "w")) == NULL) esl_fatal("Failed to open output postscript file %s", kpsfile);
dmx_Visualize(fp, kmx, -4., 5.);
fclose(fp);
}
if (ipsfile != NULL) {
if ((fp = fopen(ipsfile, "w")) == NULL) esl_fatal("Failed to open output postscript file %s", ipsfile);
dmx_Visualize(fp, imx, -4., 5.);
/* dmx_Visualize(fp, imx, dmx_upper_min(imx), dmx_upper_max(imx)); */
fclose(fp);
}
printf("After normalization/log-odds:\n");
printf("i matrix values range from %f to %f\n", dmx_upper_min(imx), dmx_upper_max(imx));
printf("k matrix values range from %f to %f\n", dmx_upper_min(kmx), dmx_upper_max(kmx));
p7_profile_Destroy(gm);
p7_bg_Destroy(bg);
p7_hmm_Destroy(core);
p7_hmm_Destroy(hmm);
p7_trace_Destroy(tr);
esl_sq_Destroy(sq);
esl_dmatrix_Destroy(imx);
esl_dmatrix_Destroy(kmx);
esl_alphabet_Destroy(abc);
esl_randomness_Destroy(r);
esl_getopts_Destroy(go);
return eslOK;
}
int
main(int argc, char **argv)
{
ESL_GETOPTS *go = p7_CreateDefaultApp(options, 1, argc, argv, banner, usage);
char *hmmfile = esl_opt_GetArg(go, 1);
ESL_STOPWATCH *w = esl_stopwatch_Create();
ESL_RANDOMNESS *r = esl_randomness_CreateFast(esl_opt_GetInteger(go, "-s"));
ESL_ALPHABET *abc = NULL;
P7_HMMFILE *hfp = NULL;
P7_HMM *hmm = NULL;
P7_BG *bg = NULL;
P7_PROFILE *gm = NULL;
P7_OPROFILE *om = NULL;
P7_OMX *ox1 = NULL;
P7_OMX *ox2 = NULL;
int L = esl_opt_GetInteger(go, "-L");
int N = esl_opt_GetInteger(go, "-N");
ESL_DSQ *dsq = malloc(sizeof(ESL_DSQ) * (L+2));
float null2[p7_MAXCODE];
int i,j,d,pos;
int nsamples = 200;
float fsc, bsc;
double Mcs;
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");
bg = p7_bg_Create(abc); p7_bg_SetLength(bg, L);
gm = p7_profile_Create(hmm->M, abc); p7_ProfileConfig(hmm, bg, gm, L, p7_LOCAL);
om = p7_oprofile_Create(gm->M, abc); p7_oprofile_Convert(gm, om);
p7_oprofile_ReconfigLength(om, L);
ox1 = p7_omx_Create(gm->M, L, L);
ox2 = p7_omx_Create(gm->M, L, L);
esl_rsq_xfIID(r, bg->f, abc->K, L, dsq);
p7_Forward (dsq, L, om, ox1, &fsc);
if (esl_opt_GetBoolean(go, "-t"))
{
P7_TRACE *tr = p7_trace_Create();
float *n2sc = malloc(sizeof(float) * (L+1));
esl_stopwatch_Start(w);
for (i = 0; i < N; i++)
{ /* This is approximately what p7_domaindef.c::region_trace_ensemble() is doing: */
for (j = 0; j < nsamples; j++)
{
p7_StochasticTrace(r, dsq, L, om, ox1, tr);
p7_trace_Index(tr);
pos = 1;
for (d = 0; d < tr->ndom; d++)
{
p7_Null2_ByTrace(om, tr, tr->tfrom[d], tr->tto[d], ox2, null2);
for (; pos <= tr->sqfrom[d]; pos++) n2sc[pos] += 1.0;
for (; pos < tr->sqto[d]; pos++) n2sc[pos] += null2[dsq[pos]];
}
for (; pos <= L; pos++) n2sc[pos] += 1.0;
p7_trace_Reuse(tr);
}
for (pos = 1; pos <= L; pos++)
n2sc[pos] = logf(n2sc[pos] / nsamples);
}
esl_stopwatch_Stop(w);
free(n2sc);
p7_trace_Destroy(tr);
}
else
{
p7_Backward(dsq, L, om, ox1, ox2, &bsc);
p7_Decoding(om, ox1, ox2, ox2);
esl_stopwatch_Start(w);
for (i = 0; i < N; i++)
p7_Null2_ByExpectation(om, ox2, null2);
esl_stopwatch_Stop(w);
}
Mcs = (double) N * (double) L * (double) gm->M * 1e-6 / (double) w->user;
esl_stopwatch_Display(stdout, w, "# CPU time: ");
printf("# M = %d\n", gm->M);
printf("# %.1f Mc/s\n", Mcs);
free(dsq);
p7_omx_Destroy(ox1);
p7_omx_Destroy(ox2);
p7_oprofile_Destroy(om);
p7_profile_Destroy(gm);
p7_bg_Destroy(bg);
p7_hmm_Destroy(hmm);
p7_hmmfile_Close(hfp);
esl_alphabet_Destroy(abc);
esl_stopwatch_Destroy(w);
esl_randomness_Destroy(r);
esl_getopts_Destroy(go);
return 0;
}
/* Function: hmmpgmd2msa()
* Synopsis: Convert an HMMPGMD-derived data stream to an MSA, based
* on the corresponding hmm
*
* Purpose: Given a data stream from HMMPGMD of the form shown
* here, produce an MSA:
* HMMD_SEARCH_STATS
* P7_HITS array of size (nhits) from above?
* then repeats of P7_DOMAIN and P7_ALIDISPLAY data
* for the hits, where each hit with d domains
* produces
* d P7_DOMAINs
* then
* d P7_ALIDISPLAYs
* ... optionally adding a sequence with length matching
* that of the hmm, which will be included in the alignment.
*
* A further extension has been the ability to include or exclude
* sequences form the list of hits.
*
* This function's expected use is as a helper function for
* the hmmer website, which gets the above data stream from
* hmmpgmd.
*
* Args : data: a pointer to binary data in the format given above
* hmm: the HMM against which the alidisplay traces and
* additional sequences/traces are threaded to reach
* the returned msa.
* qsq : optional sequence to be included in the output msa;
* must have the same number of residues as the hmm
* has states, as each residue i will be aligned to
* state i.
* incl: optional array of sequence names, in the case of
* hmmpgmd a list of ints, which are are excluded due
* to the sequence threshold, but have been selected
* to be included in the alignment. This ties in
* with the way jackhmmer is implemented on the
* HMMER website.
* incl_size: required size of the incl array. zero if incl is null.
* excl: optional array of sequence names, in the case of
* hmmpgmd a list of ints, which are are included as they
* score above threshold, but have been selected
* to be excluded from the alignment.
* excl_size: required size of the excl array. zero if excl is null.
*
*
* Returns: Pointer to completed MSA object. NULL on error
*
*/
int
hmmpgmd2msa(void *data, P7_HMM *hmm, ESL_SQ *qsq, int *incl, int incl_size, int *excl, int excl_size, ESL_MSA **ret_msa) {
int i, j;
int c;
int status;
int set_included;
/* trace of the query sequence with N residues onto model with N match states */
P7_TRACE *qtr = NULL;
int extra_sqcnt = 0;
/* vars used to read from the binary data */
HMMD_SEARCH_STATS *stats = NULL; /* pointer to a single stats object, at the beginning of data */
P7_HIT *hits = NULL; /* an array of hits, at the appropriate offset in data */
/* vars used in msa construction */
P7_TOPHITS th;
P7_ALIDISPLAY *ad, *ad2;
ESL_MSA *msa = NULL;
P7_DOMAIN *dom = NULL;
char *p = (char*)data; /*pointer used to walk along data, must be char* to allow pointer arithmetic */
th.N = 0;
th.unsrt = NULL;
th.hit = NULL;
/* optionally build a faux trace for the query sequence: relative to core model (B->M_1..M_L->E) */
if (qsq != NULL) {
if (qsq->n != hmm->M) {
status = eslFAIL;
goto ERROR;
}
if ((qtr = p7_trace_Create()) == NULL) {status = eslFAIL; goto ERROR; }
if ((status = p7_trace_Append(qtr, p7T_B, 0, 0)) != eslOK) goto ERROR;
for (i = 1; i <= qsq->n; i++)
if ((status = p7_trace_Append(qtr, p7T_M, i, i)) != eslOK) goto ERROR;
if ((status = p7_trace_Append(qtr, p7T_E, 0, 0)) != eslOK) goto ERROR;
qtr->M = qsq->n;
qtr->L = qsq->n;
extra_sqcnt = 1;
}
/* get search stats + hit info */
stats = (HMMD_SEARCH_STATS*)p;
/* sanity check */
if ( ( stats->Z_setby != p7_ZSETBY_NTARGETS && stats->Z_setby != p7_ZSETBY_OPTION && stats->Z_setby != p7_ZSETBY_FILEINFO )
|| ( stats->domZ_setby != p7_ZSETBY_NTARGETS && stats->domZ_setby != p7_ZSETBY_OPTION && stats->domZ_setby != p7_ZSETBY_FILEINFO )
|| stats->nhits > 10000000
|| stats->elapsed > 1000000
) {
status = eslFAIL;
goto ERROR;
}
/* ok, it looks legitimate */
p += sizeof(HMMD_SEARCH_STATS);
hits = (P7_HIT*)p;
p += sizeof(P7_HIT) * stats->nhits;
/* create a tophits object, to be passed to p7_tophits_Alignment() */
ESL_ALLOC( th.unsrt, sizeof(P7_HIT) * stats->nhits);
memcpy( th.unsrt, hits, sizeof(P7_HIT) * stats->nhits);
ESL_ALLOC( th.hit, sizeof(P7_HIT*) * stats->nhits);
for (i=0; i<stats->nhits; i++) {
th.hit[i] = &(th.unsrt[i]);
if ( th.hit[i]->ndom > 10000
|| th.hit[i]->flags > p7_IS_INCLUDED + p7_IS_REPORTED + p7_IS_NEW + p7_IS_DROPPED + p7_IS_DUPLICATE
) {
status = eslFAIL;
goto ERROR;
}
}
// th.unsrt = NULL;
th.N = stats->nhits;
th.nreported = 0;
th.nincluded = 0;
th.is_sorted_by_sortkey = 0;
th.is_sorted_by_seqidx = 0;
for (i = 0; i < th.N; i++) {
ESL_ALLOC( th.hit[i]->dcl, sizeof(P7_DOMAIN) * th.hit[i]->ndom);
/* Go through the hits and set to be excluded or included as necessary */
set_included = 0;
if(th.hit[i]->flags & p7_IS_INCLUDED){
if(excl_size > 0){
for( c = 0; c < excl_size; c++){
if(excl[c] == (long)(th.hit[i]->name) ){
th.hit[i]->flags = p7_IS_DROPPED;
th.hit[i]->nincluded = 0;
break;
}
}
}
}else{
if(incl_size > 0){
for( c = 0; c < incl_size; c++){
if(incl[c] == (long)th.hit[i]->name ){
th.hit[i]->flags = p7_IS_INCLUDED;
set_included = 1;
}
}
}
}
/* first grab all the P7_DOMAINs for the hit */
for (j=0; j < th.hit[i]->ndom; j++) {
dom = th.hit[i]->dcl + j;
memcpy(dom , (P7_DOMAIN*)p, sizeof(P7_DOMAIN));
/* Possibly set domains to be include if being
* externally set via incl list*/
if(set_included) th.hit[i]->dcl[j].is_included = 1;
p += sizeof(P7_DOMAIN);
}
/* then grab the P7_ALIDISPLAYs for the hit */
for (j=0; j < th.hit[i]->ndom; j++) {
ad = (P7_ALIDISPLAY*)p;
ESL_ALLOC(th.hit[i]->dcl[j].ad, sizeof(P7_ALIDISPLAY));
ad2 = th.hit[i]->dcl[j].ad;
ad2->memsize = ad->memsize;
ad2->rfline = ad->rfline;
ad2->mmline = ad->mmline;
ad2->csline = ad->csline ;
ad2->model = ad->model ;
ad2->mline = ad->mline ;
ad2->aseq = ad->aseq ;
ad2->ppline = ad->ppline;
ad2->N = ad->N;
ad2->hmmname = ad->hmmname;
ad2->hmmacc = ad->hmmacc ;
ad2->hmmdesc = ad->hmmdesc;
ad2->hmmfrom = ad->hmmfrom;
ad2->hmmto = ad->hmmto;
ad2->M = ad->M;
ad2->sqname = ad->sqname;
ad2->sqacc = ad->sqacc ;
ad2->sqdesc = ad->sqdesc;
ad2->sqfrom = ad->sqfrom;
ad2->sqto = ad->sqto;
ad2->L = ad->L;
p += sizeof(P7_ALIDISPLAY);
ESL_ALLOC(ad2->mem, ad2->memsize);
memcpy(ad2->mem, p, ad->memsize);
p += ad2->memsize;
p7_alidisplay_Deserialize(ad2);
}
}
/* use the tophits and trace info above to produce an alignment */
if ( (status = p7_tophits_Alignment(&th, hmm->abc, &qsq, &qtr, extra_sqcnt, p7_ALL_CONSENSUS_COLS, &msa)) != eslOK) goto ERROR;
/* free memory */
if (qtr != NULL) free(qtr);
for (i = 0; i < th.N; i++) {
for (j=0; j < th.hit[i]->ndom; j++)
p7_alidisplay_Destroy(th.hit[i]->dcl[j].ad);
if (th.hit[i]->dcl != NULL) free (th.hit[i]->dcl);
}
if (th.unsrt != NULL) free (th.unsrt);
if (th.hit != NULL) free (th.hit);
*ret_msa = msa;
return eslOK;
ERROR:
/* free memory */
if (qtr != NULL) free(qtr);
for (i = 0; i < th.N; i++) {
for (j=0; j < th.hit[i]->ndom; j++)
p7_alidisplay_Destroy(th.hit[i]->dcl[j].ad);
if (th.hit[i]->dcl != NULL) free (th.hit[i]->dcl);
}
if (th.unsrt != NULL) free (th.unsrt);
if (th.hit != NULL) free (th.hit);
return status;
}
int
main(int argc, char **argv)
{
ESL_GETOPTS *go = p7_CreateDefaultApp(options, 2, argc, argv, banner, usage);
ESL_RANDOMNESS *rng = esl_randomness_CreateFast(esl_opt_GetInteger(go, "-s"));
ESL_ALPHABET *abc = NULL;
char *ghmmfile = esl_opt_GetArg(go, 1); /* HMMs parameterized for sequence generation */
char *ahmmfile = esl_opt_GetArg(go, 2); /* HMMs parameterized for alignment */
int N = esl_opt_GetInteger(go, "-N");
P7_HMMFILE *ghfp = NULL;
P7_HMMFILE *ahfp = NULL;
P7_HMM *ghmm = NULL;
P7_HMM *ahmm = NULL;
P7_PROFILE *ggm = NULL;
P7_PROFILE *agm = NULL;
P7_OPROFILE *aom = NULL;
P7_BG *bg = NULL;
ESL_SQ *sq = NULL;
P7_TRACE *reftr = p7_trace_Create();
P7_TRACE *testtr = p7_trace_Create();
P7_TRACE_METRICS *tmetrics = p7_trace_metrics_Create();
P7_REFMX *rmx = p7_refmx_Create(100,100);
// P7_FILTERMX *ox = NULL;
P7_HARDWARE *hw;
if ((hw = p7_hardware_Create ()) == NULL) p7_Fail("Couldn't get HW information data structure");
P7_SPARSEMASK *sm = p7_sparsemask_Create(100, 100, hw->simd);
P7_SPARSEMX *sxv = p7_sparsemx_Create(NULL);
int idx;
char errbuf[eslERRBUFSIZE];
int status;
p7_Init();
/* open HMM file containing models parameterized for generation (sampling) of seqs */
status = p7_hmmfile_OpenE(ghmmfile, NULL, &ghfp, errbuf);
if (status == eslENOTFOUND) p7_Fail("File existence/permissions problem in trying to open HMM file %s.\n%s\n", ghmmfile, errbuf);
else if (status == eslEFORMAT) p7_Fail("File format problem in trying to open HMM file %s.\n%s\n", ghmmfile, errbuf);
else if (status != eslOK) p7_Fail("Unexpected error %d in opening HMM file %s.\n%s\n", status, ghmmfile, errbuf);
/* open HMM file containing models parameterized for alignment (may be the same as ghmmfile) */
status = p7_hmmfile_OpenE(ahmmfile, NULL, &ahfp, errbuf);
if (status == eslENOTFOUND) p7_Fail("File existence/permissions problem in trying to open HMM file %s.\n%s\n", ahmmfile, errbuf);
else if (status == eslEFORMAT) p7_Fail("File format problem in trying to open HMM file %s.\n%s\n", ahmmfile, errbuf);
else if (status != eslOK) p7_Fail("Unexpected error %d in opening HMM file %s.\n%s\n", status, ahmmfile, errbuf);
while ( (status = p7_hmmfile_Read(ghfp, &abc, &ghmm)) == eslOK) /* <abc> gets set on first read */
{
/* read the counterpart HMM from <ahfp> */
status = p7_hmmfile_Read(ahfp, &abc, &ahmm);
if (status == eslEFORMAT) p7_Fail("Bad file format in HMM file %s:\n%s\n", ahfp->fname, ahfp->errbuf);
else if (status == eslEINCOMPAT) p7_Fail("HMM in %s is not in the expected %s alphabet\n", ahfp->fname, esl_abc_DecodeType(abc->type));
else if (status == eslEOF) p7_Fail("Empty HMM file %s? No HMM data found.\n", ahfp->fname);
else if (status != eslOK) p7_Fail("Unexpected error in reading HMMs from %s\n", ahfp->fname);
/* try to validate that they're the "same" */
if (ahmm->M != ghmm->M || strcmp(ahmm->name, ghmm->name) != 0) p7_Fail("<gen-hmmfile>, <ali-hmmfile> contain different set or order of models");
/* deferred one-time creation of structures that need to know the alphabet */
if (!bg) bg = p7_bg_Create(abc);
if (!sq) sq = esl_sq_CreateDigital(abc);
ggm = p7_profile_Create(ghmm->M, abc);
agm = p7_profile_Create(ahmm->M, abc);
aom = p7_oprofile_Create(ahmm->M, abc, hw->simd);
p7_profile_ConfigCustom(ggm, ghmm, bg, esl_opt_GetInteger(go, "--gL"), esl_opt_GetReal(go, "--gnj"), esl_opt_GetReal(go, "--gpglocal"));
p7_profile_ConfigCustom(agm, ahmm, bg, 100, esl_opt_GetReal(go, "--anj"), esl_opt_GetReal(go, "--apglocal"));
p7_oprofile_Convert(agm, aom);
for (idx = 1; idx <= N; idx++)
{
p7_ProfileEmit(rng, ghmm, ggm, bg, sq, reftr);
if (esl_opt_GetBoolean(go, "--dumpseqs")) {
esl_sq_FormatName(sq, "seq%d", idx);
esl_sqio_Write(stdout, sq, eslSQFILE_FASTA, FALSE);
}
p7_bg_SetLength(bg, sq->n);
p7_profile_SetLength(agm, sq->n);
p7_sparsemask_Reinit(sm, agm->M, sq->n);
p7_sparsemask_AddAll(sm);
if (esl_opt_GetBoolean(go, "--vit")) p7_ReferenceViterbi(sq->dsq, sq->n, agm, rmx, testtr, /*opt_vsc=*/NULL);
else p7_SparseViterbi (sq->dsq, sq->n, agm, sm, sxv, testtr, /*opt_vsc=*/NULL);
p7_trace_metrics(reftr, testtr, tmetrics);
p7_sparsemask_Reuse(sm);
p7_sparsemx_Reuse(sxv);
//p7_filtermx_Reuse(ox);
p7_refmx_Reuse(rmx);
esl_sq_Reuse(sq);
p7_trace_Reuse(reftr);
p7_trace_Reuse(testtr);
}
p7_oprofile_Destroy(aom);
p7_profile_Destroy(ggm);
p7_profile_Destroy(agm);
p7_hmm_Destroy(ghmm);
p7_hmm_Destroy(ahmm);
}
/* we leave the loop with <status> set by a p7_hmmfile_Read() on ghfp; if all is well, status=eslEOF */
if (status == eslEFORMAT) p7_Fail("Bad file format in HMM file %s:\n%s\n", ghfp->fname, ghfp->errbuf);
else if (status == eslEINCOMPAT) p7_Fail("HMM in %s is not in the expected %s alphabet\n", ghfp->fname, esl_abc_DecodeType(abc->type));
else if (status != eslEOF) p7_Fail("Unexpected error in reading HMMs from %s\n", ghfp->fname);
p7_trace_metrics_Dump(stdout, tmetrics);
p7_hmmfile_Close(ghfp);
p7_hmmfile_Close(ahfp);
// p7_filtermx_Destroy(ox);
p7_sparsemask_Destroy(sm);
p7_sparsemx_Destroy(sxv);
p7_refmx_Destroy(rmx);
p7_trace_metrics_Destroy(tmetrics);
p7_trace_Destroy(testtr);
p7_trace_Destroy(reftr);
p7_bg_Destroy(bg);
esl_alphabet_Destroy(abc);
esl_randomness_Destroy(rng);
esl_getopts_Destroy(go);
}
int
main(int argc, char **argv)
{
ESL_GETOPTS *go = esl_getopts_CreateDefaultApp(options, 1, argc, argv, banner, usage);
char *hmmfile = esl_opt_GetArg(go, 1);
int N = esl_opt_GetInteger(go, "-N");
ESL_STOPWATCH *w = esl_stopwatch_Create();
ESL_RANDOMNESS *r = esl_randomness_CreateFast(esl_opt_GetInteger(go, "-s"));
ESL_ALPHABET *abc = NULL;
P7_HMMFILE *hfp = NULL;
P7_HMM *hmm = NULL;
P7_BG *bg = NULL;
P7_PROFILE *gm = NULL;
P7_OPROFILE *om = NULL;
P7_TRACE *tr = NULL;
ESL_SQ *sq = NULL;
P7_ALIDISPLAY *ad = NULL;
int i,z;
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);
bg = p7_bg_Create(abc);
p7_bg_SetLength(bg, 0);
gm = p7_profile_Create(hmm->M, abc);
p7_ProfileConfig(hmm, bg, gm, 0, p7_UNIGLOCAL); /* that sets N,C,J to generate nothing */
om = p7_oprofile_Create(gm->M, abc);
p7_oprofile_Convert(gm, om);
if (esl_opt_GetBoolean(go, "-p")) tr = p7_trace_CreateWithPP();
else tr = p7_trace_Create();
sq = esl_sq_CreateDigital(abc);
esl_stopwatch_Start(w);
for (i = 0; i < N; i++)
{
p7_ProfileEmit(r, hmm, gm, bg, sq, tr);
esl_sq_SetName(sq, "random");
if (! esl_opt_GetBoolean(go, "-b"))
{
if (esl_opt_GetBoolean(go, "-p"))
for (z = 0; z < tr->N; z++)
if (tr->i[z] > 0) tr->pp[z] = esl_random(r);
ad = p7_alidisplay_Create(tr, 0, om, sq);
p7_alidisplay_Print(stdout, ad, 40, 80, FALSE);
p7_alidisplay_Destroy(ad);
}
p7_trace_Reuse(tr);
esl_sq_Reuse(sq);
}
esl_stopwatch_Stop(w);
esl_stopwatch_Display(stdout, w, "# CPU time: ");
esl_sq_Destroy(sq);
p7_trace_Destroy(tr);
p7_oprofile_Destroy(om);
p7_profile_Destroy(gm);
p7_bg_Destroy(bg);
p7_hmm_Destroy(hmm);
esl_alphabet_Destroy(abc);
esl_randomness_Destroy(r);
esl_stopwatch_Destroy(w);
esl_getopts_Destroy(go);
return 0;
}
/* 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;
}
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;
ESL_RANDOMNESS *rng = esl_randomness_CreateFast(0);
P7_HMMFILE *hfp = NULL;
P7_HMM *hmm = NULL;
P7_BG *bg = NULL;
P7_PROFILE *gm = NULL;
P7_OPROFILE *om = NULL;
P7_GMX *gx = NULL;
P7_OMX *fwd = NULL;
P7_TRACE *tr = NULL;
ESL_SQ *sq = NULL;
ESL_SQFILE *sqfp = NULL;
int format = eslSQFILE_UNKNOWN;
int N = esl_opt_GetInteger(go, "-N");
int i;
float vsc, fsc, tsc;
char errbuf[eslERRBUFSIZE];
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");
/* 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");
/* create default null model, then create and optimize profile */
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);
om = p7_oprofile_Create(gm->M, abc); p7_oprofile_Convert(gm, om);
if (esl_opt_GetBoolean(go, "-p")) p7_oprofile_Dump(stdout, om);
fwd = p7_omx_Create(gm->M, sq->n, sq->n);
gx = p7_gmx_Create(gm->M, sq->n);
tr = p7_trace_Create();
if (esl_opt_GetBoolean(go, "-m") == TRUE) p7_omx_SetDumpMode(stdout, fwd, TRUE);
p7_GViterbi(sq->dsq, sq->n, gm, gx, &vsc);
p7_Forward (sq->dsq, sq->n, om, fwd, &fsc);
for (i = 0; i < N; i++)
{
p7_StochasticTrace(rng, sq->dsq, sq->n, om, fwd, tr);
p7_trace_Score(tr, sq->dsq, gm, &tsc);
if (esl_opt_GetBoolean(go, "-t") == TRUE) p7_trace_Dump(stdout, tr, gm, sq->dsq);
if (p7_trace_Validate(tr, abc, sq->dsq, errbuf) != eslOK) p7_Die("trace %d fails validation:\n%s\n", i, errbuf);
printf("Sampled trace: %.4f nats\n", tsc);
p7_trace_Reuse(tr);
}
printf("Forward score: %.4f nats\n", fsc);
printf("Viterbi score: %.4f nats\n", vsc);
/* cleanup */
esl_sq_Destroy(sq);
esl_sqfile_Close(sqfp);
p7_trace_Destroy(tr);
p7_omx_Destroy(fwd);
p7_gmx_Destroy(gx);
p7_oprofile_Destroy(om);
p7_profile_Destroy(gm);
p7_bg_Destroy(bg);
p7_hmm_Destroy(hmm);
p7_hmmfile_Close(hfp);
esl_randomness_Destroy(rng);
esl_alphabet_Destroy(abc);
esl_getopts_Destroy(go);
return 0;
}
int
main(int argc, char **argv)
{
ESL_GETOPTS *go = esl_getopts_CreateDefaultApp(options, 1, argc, argv, banner, usage);
char *hmmfile = esl_opt_GetArg(go, 1);
ESL_STOPWATCH *w = esl_stopwatch_Create();
ESL_RANDOMNESS *r = esl_randomness_CreateFast(esl_opt_GetInteger(go, "-s"));
ESL_ALPHABET *abc = NULL;
P7_HMMFILE *hfp = NULL;
P7_HMM *hmm = NULL;
P7_BG *bg = NULL;
P7_PROFILE *gm = NULL;
P7_OPROFILE *om = NULL;
P7_GMX *gx = NULL;
P7_OMX *fwd = NULL;
P7_TRACE *tr = NULL;
int L = esl_opt_GetInteger(go, "-L");
int N = esl_opt_GetInteger(go, "-N");
ESL_DSQ *dsq = malloc(sizeof(ESL_DSQ) * (L+2));
int i;
float sc, fsc, vsc;
float bestsc = -eslINFINITY;
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");
bg = p7_bg_Create(abc); p7_bg_SetLength(bg, L);
gm = p7_profile_Create(hmm->M, abc); p7_ProfileConfig(hmm, bg, gm, L, p7_UNILOCAL);
om = p7_oprofile_Create(gm->M, abc); p7_oprofile_Convert(gm, om);
fwd = p7_omx_Create(gm->M, L, L);
gx = p7_gmx_Create(gm->M, L);
tr = p7_trace_Create();
esl_rsq_xfIID(r, bg->f, abc->K, L, dsq);
p7_GViterbi(dsq, L, gm, gx, &vsc);
p7_Forward (dsq, L, om, fwd, &fsc);
esl_stopwatch_Start(w);
for (i = 0; i < N; i++)
{
p7_StochasticTrace(r, dsq, L, om, fwd, tr);
p7_trace_Score(tr, dsq, gm, &sc);
bestsc = ESL_MAX(bestsc, sc);
p7_trace_Reuse(tr);
}
esl_stopwatch_Stop(w);
esl_stopwatch_Display(stdout, w, "# CPU time: ");
printf("forward sc = %.4f nats\n", fsc);
printf("viterbi sc = %.4f nats\n", vsc);
printf("max trace sc = %.4f nats\n", bestsc);
free(dsq);
p7_trace_Destroy(tr);
p7_gmx_Destroy(gx);
p7_omx_Destroy(fwd);
p7_oprofile_Destroy(om);
p7_profile_Destroy(gm);
p7_bg_Destroy(bg);
p7_hmm_Destroy(hmm);
p7_hmmfile_Close(hfp);
esl_alphabet_Destroy(abc);
esl_stopwatch_Destroy(w);
esl_randomness_Destroy(r);
esl_getopts_Destroy(go);
return 0;
}
/* process_workunit()
*
* This is the routine that actually does the work.
*
* A work unit consists of one HMM, <hmm>.
* The result is the <scores> array, which contains an array of N scores;
* caller provides this memory.
* How those scores are generated is controlled by the application configuration in <cfg>.
*/
static int
process_workunit(ESL_GETOPTS *go, struct cfg_s *cfg, char *errbuf, P7_HMM *hmm, double *scores, int *alilens)
{
int L = esl_opt_GetInteger(go, "-L");
P7_PROFILE *gm = NULL;
P7_OPROFILE *om = NULL;
P7_REFMX *rmx = NULL;
P7_CHECKPTMX *cx = NULL;
P7_FILTERMX *fx = NULL;
P7_TRACE *tr = NULL;
ESL_DSQ *dsq = NULL;
int i;
int scounts[p7T_NSTATETYPES]; /* state usage counts from a trace */
float sc;
float nullsc;
int status;
P7_HARDWARE *hw;
if ((hw = p7_hardware_Create ()) == NULL) p7_Fail("Couldn't get HW information data structure");
/* Optionally set a custom background, determined by model composition;
* an experimental hack.
*/
if (esl_opt_GetBoolean(go, "--bgcomp"))
{
float *p = NULL;
float KL;
p7_hmm_CompositionKLDist(hmm, cfg->bg, &KL, &p);
esl_vec_FCopy(p, cfg->abc->K, cfg->bg->f);
}
/* Create and configure our generic profile, as requested */
gm = p7_profile_Create(hmm->M, cfg->abc);
if (esl_opt_GetBoolean(go, "--multi"))
{
if (esl_opt_GetBoolean(go, "--dual")) { p7_profile_Config (gm, hmm, cfg->bg); }
else if (esl_opt_GetBoolean(go, "--local")) { p7_profile_ConfigLocal (gm, hmm, cfg->bg, L); }
else if (esl_opt_GetBoolean(go, "--glocal")) { p7_profile_ConfigGlocal(gm, hmm, cfg->bg, L); }
}
else if (esl_opt_GetBoolean(go, "--uni"))
{
if (esl_opt_GetBoolean(go, "--dual")) { p7_profile_ConfigCustom (gm, hmm, cfg->bg, L, 0.0, 0.5); }
else if (esl_opt_GetBoolean(go, "--local")) { p7_profile_ConfigUnilocal (gm, hmm, cfg->bg, L); }
else if (esl_opt_GetBoolean(go, "--glocal")) { p7_profile_ConfigUniglocal(gm, hmm, cfg->bg, L); }
}
p7_profile_SetLength(gm, L);
p7_bg_SetLength(cfg->bg, L);
if (esl_opt_GetBoolean(go, "--x-no-lengthmodel")) elide_length_model(gm, cfg->bg);
/* Allocate DP matrix for <gm>.
*/
rmx = p7_refmx_Create(gm->M, L);
/* Create and configure the vectorized profile, if needed;
* and allocate its DP matrix
*/
if (esl_opt_GetBoolean(go, "--vector"))
{
om = p7_oprofile_Create(gm->M, cfg->abc, om->simd);
p7_oprofile_Convert(gm, om);
cx = p7_checkptmx_Create(gm->M, L, ESL_MBYTES(32), om->simd);
fx = p7_filtermx_Create(gm->M, om->simd);
}
/* Remaining allocation */
ESL_ALLOC(dsq, sizeof(ESL_DSQ) * (L+2));
tr = p7_trace_Create();
/* Collect scores from N random sequences of length L */
for (i = 0; i < cfg->N; i++)
{
esl_rsq_xfIID(cfg->r, cfg->bg->f, cfg->abc->K, L, dsq);
sc = eslINFINITY;
/* Vectorized implementations of Viterbi, MSV may overflow.
* In this case, they'll leave sc=eslINFINITY.
* Then we fail over to the nonvector "generic" implementation.
* That's why this next block isn't an if/else.
*/
if (esl_opt_GetBoolean(go, "--vector"))
{
if (esl_opt_GetBoolean(go, "--vit")) p7_ViterbiFilter(dsq, L, om, fx, &sc);
else if (esl_opt_GetBoolean(go, "--fwd")) p7_ForwardFilter(dsq, L, om, cx, &sc);
else if (esl_opt_GetBoolean(go, "--msv")) p7_MSVFilter (dsq, L, om, fx, &sc);
}
/* If we tried a vector calculation above but it overflowed,
* or if we're to do --generic DP calculations, sc==eslINFINITY now;
* hence the if condition here:
*/
if (sc == eslINFINITY)
{
if (esl_opt_GetBoolean(go, "--fwd")) p7_ReferenceForward(dsq, L, gm, rmx, &sc); /* any mode: dual,local,glocal; gm's config takes care of this */
else if (esl_opt_GetBoolean(go, "--vit")) p7_ReferenceViterbi(dsq, L, gm, rmx, tr, &sc); /* local-only mode. cmdline opts processing has already assured that --local set */
else if (esl_opt_GetBoolean(go, "--msv")) p7_Die("We used to be able to do a generic MSV algorithm - but no longer");
}
/* Optional: get Viterbi alignment length too. */
if (esl_opt_GetBoolean(go, "-a")) /* -a only works with Viterbi; getopts has checked this already; <tr> must be valid */
{
p7_trace_GetStateUseCounts(tr, scounts);
/* there's various ways we could counts "alignment length".
* Here we'll use the total length of model used, in nodes: M+D states.
* score vs al would gives us relative entropy / model position.
*/
/* alilens[i] = scounts[p7T_D] + scounts[p7T_I]; SRE: temporarily testing this instead */
alilens[i] = scounts[p7T_ML] + scounts[p7T_DL] + scounts[p7T_IL] +
scounts[p7T_MG] + scounts[p7T_DG] + scounts[p7T_IG];
p7_trace_Reuse(tr);
}
p7_bg_NullOne(cfg->bg, dsq, L, &nullsc);
scores[i] = (sc - nullsc) / eslCONST_LOG2;
if (cx) p7_checkptmx_Reuse(cx);
if (fx) p7_filtermx_Reuse(fx);
p7_refmx_Reuse(rmx);
}
status = eslOK;
/* deliberate flowthru */
ERROR:
if (dsq != NULL) free(dsq);
p7_checkptmx_Destroy(cx);
p7_filtermx_Destroy(fx);
p7_oprofile_Destroy(om);
p7_profile_Destroy(gm);
p7_refmx_Destroy(rmx);
p7_trace_Destroy(tr);
if (status == eslEMEM) sprintf(errbuf, "allocation failure");
return status;
}
/* "generation" test
* Compare a randomly sampled profile to sequences sampled
* from that profile.
*
* This test is not very stringent, because we don't know the "true"
* envelopes. Rather, this is more of a test that nothing obviously
* bad happens, like a crash, or obviously incorrect data.
*
* We test:
* 1. Seq coordinates of each envelope are coherent:
* 1 <= oa <= ia <= i0 <= ib <= ob <= L
*
* 2. Envelopes do not overlap (assuming default threshold of
* 0.5 when defining them):
* ia(d) > ib(d-1) for d = 2..D
* (Outer envelopes, in contrast, can overlap.)
*
* 3. envsc(d) <= asc_sc <= fwdsc.
*
* 4. If D=1 (single domain) in both the generated trace
* and the inferred envelopes, and the domain coords in
* the trace are encompassed by the outer envelope,
* then envsc(d) >= generated trace score.
*/
static void
utest_generation(ESL_RANDOMNESS *rng, int M, const ESL_ALPHABET *abc, int N)
{
char msg[] = "reference_envelopes:: generation unit test failed";
ESL_SQ *sq = esl_sq_CreateDigital(abc);
P7_BG *bg = p7_bg_Create(abc);
P7_HMM *hmm = NULL;
P7_PROFILE *gm = p7_profile_Create(M, abc);
P7_TRACE *gtr = p7_trace_Create(); // generated trace
P7_TRACE *vtr = p7_trace_Create(); // Viterbi trace
P7_REFMX *rxf = p7_refmx_Create(M, 20); // Fwd, Vit ~~> ASC Decode UP
P7_REFMX *rxd = p7_refmx_Create(M, 20); // Bck, Decode ~~> ASC Decode DOWN
P7_REFMX *afu = p7_refmx_Create(M, 20); // ASC Fwd UP
P7_REFMX *afd = p7_refmx_Create(M, 20); // ASC Fwd DOWN
P7_REFMX *apu = rxf; // for 'clarity' we use two names for this mx
P7_REFMX *apd = rxd; // ... and this one too.
float *wrk = NULL;
P7_ANCHORS *anch = p7_anchors_Create();
P7_ANCHORHASH *ah = p7_anchorhash_Create();
P7_ENVELOPES *env = p7_envelopes_Create();
float tol = 0.001;
float gsc, fsc, asc;
int idx;
int d;
if ( p7_modelsample(rng, M, abc, &hmm) != eslOK) esl_fatal(msg);
if ( p7_profile_Config(gm, hmm, bg) != eslOK) esl_fatal(msg);
for (idx = 0; idx < N; idx++)
{
/* Emit sequence from model, using an arbitrary length model of <M>;
* restrict the emitted sequence length to 6M, arbitrarily, to
* keep it down to something reasonable.
*/
if ( p7_profile_SetLength(gm, M) != eslOK) esl_fatal(msg);
do {
esl_sq_Reuse(sq);
if (p7_ProfileEmit(rng, hmm, gm, bg, sq, gtr) != eslOK) esl_fatal(msg);
} while (sq->n > M * 6);
if (p7_trace_Index (gtr) != eslOK) esl_fatal(msg);
if (p7_trace_Score (gtr, sq->dsq, gm, &gsc) != eslOK) esl_fatal(msg);
/* Reset the length model to the actual length sq->n, then
* put it through the domain postprocessing analysis pipeline
*/
if ( p7_profile_SetLength(gm, sq->n) != eslOK) esl_fatal(msg);
/* First pass analysis */
if ( p7_ReferenceViterbi (sq->dsq, sq->n, gm, rxf, vtr, NULL) != eslOK) esl_fatal(msg);
if ( p7_ReferenceForward (sq->dsq, sq->n, gm, rxf, &fsc) != eslOK) esl_fatal(msg);
if ( p7_ReferenceBackward(sq->dsq, sq->n, gm, rxd, NULL) != eslOK) esl_fatal(msg);
if ( p7_ReferenceDecoding(sq->dsq, sq->n, gm, rxf, rxd, rxd) != eslOK) esl_fatal(msg);
/* Anchor determination (MPAS algorithm) */
if ( p7_reference_Anchors(rng, sq->dsq, sq->n, gm, rxf, rxd, vtr, &wrk, ah,
afu, afd, anch, &asc, NULL, NULL) != eslOK) esl_fatal(msg);
/* Reuse rxf,rxd as apu, apd; finish ASC analysis with Backward, Decoding */
p7_refmx_Reuse(apu); p7_refmx_Reuse(apd);
if ( p7_ReferenceASCBackward(sq->dsq, sq->n, gm, anch->a, anch->D, apu, apd, NULL) != eslOK) esl_fatal(msg);
if ( p7_ReferenceASCDecoding(sq->dsq, sq->n, gm, anch->a, anch->D, afu, afd, apu, apd, apu, apd) != eslOK) esl_fatal(msg);
/* Envelope calculation */
if ( p7_reference_Envelopes(sq->dsq, sq->n, gm, anch->a, anch->D, apu, apd, afu, afd, env) != eslOK) esl_fatal(msg);
/* Test 1. Coords of each domain are coherent */
if (anch->D != env->D) esl_fatal(msg);
for (d = 1; d <= anch->D; d++)
if (! (1 <= env->arr[d].oa &&
env->arr[d].oa <= env->arr[d].ia &&
env->arr[d].ia <= env->arr[d].i0 &&
env->arr[d].i0 <= env->arr[d].ib &&
env->arr[d].ib <= env->arr[d].ob &&
env->arr[d].ob <= sq->n)) esl_fatal(msg);
/* Test 2. Envelopes do not overlap. */
for (d = 1; d <= anch->D; d++)
if (! (env->arr[d].ia > env->arr[d-1].ib)) esl_fatal(msg);
/* Test 3. envsc(d) <= asc_sc <= fwdsc */
for (d = 1; d <= anch->D; d++)
if (! (env->arr[d].env_sc <= asc+tol && asc <= fsc+tol)) esl_fatal(msg);
/* Test 4, only on D=1 case with generated trace's domain
* encompassed by the outer envelope
*/
if (gtr->ndom == 1 && anch->D == 1 &&
gtr->sqfrom[0] >= env->arr[1].oa && // in <gtr>, domains are 0..D-1; in <env>, 1..D
gtr->sqto[0] <= env->arr[1].ob)
if (! ( env->arr[1].env_sc >= gsc)) esl_fatal(msg);
p7_envelopes_Reuse(env);
p7_anchors_Reuse(anch);
p7_anchorhash_Reuse(ah);
p7_refmx_Reuse(rxf); p7_refmx_Reuse(rxd);
p7_refmx_Reuse(afu); p7_refmx_Reuse(afd);
p7_trace_Reuse(gtr); p7_trace_Reuse(vtr);
esl_sq_Reuse(sq);
}
if (wrk) free(wrk);
p7_envelopes_Destroy(env);
p7_anchors_Destroy(anch);
p7_anchorhash_Destroy(ah);
p7_refmx_Destroy(afu); p7_refmx_Destroy(afd);
p7_refmx_Destroy(rxf); p7_refmx_Destroy(rxd);
p7_trace_Destroy(vtr); p7_trace_Destroy(gtr);
p7_profile_Destroy(gm);
p7_hmm_Destroy(hmm);
p7_bg_Destroy(bg);
esl_sq_Destroy(sq);
}
int
main(int argc, char **argv)
{
ESL_GETOPTS *go = p7_CreateDefaultApp(options, 2, argc, argv, banner, usage);
ESL_RANDOMNESS *rng = esl_randomness_Create(esl_opt_GetInteger(go, "-s"));
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;
ESL_SQ *sq = NULL;
ESL_SQFILE *sqfp = NULL;
int format = eslSQFILE_UNKNOWN;
P7_ANCHORS *anch = p7_anchors_Create();
P7_ANCHORHASH *ah = p7_anchorhash_Create();
P7_ENVELOPES *env = p7_envelopes_Create();
P7_REFMX *rxf = NULL;
P7_REFMX *rxd = NULL;
P7_REFMX *afu = NULL;
P7_REFMX *afd = NULL;
P7_REFMX *apu = NULL;
P7_REFMX *apd = NULL;
P7_TRACE *tr = NULL;
float *wrk = NULL;
P7_MPAS_PARAMS prm;
P7_MPAS_STATS stats;
float fsc, vsc, asc, asc_b;
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);
/* Open sequence file */
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);
/* Read one sequence */
status = esl_sqio_Read(sqfp, sq);
if (status == eslEFORMAT) p7_Fail("Parse failed (sequence file %s)\n%s\n", sqfp->filename, sqfp->get_error(sqfp));
else if (status != eslOK) p7_Fail("Unexpected error %d reading sequence file %s", status, sqfp->filename);
esl_sqfile_Close(sqfp);
/* Configure a profile from the HMM */
bg = p7_bg_Create(abc);
gm = p7_profile_Create(hmm->M, abc);
p7_profile_Config(gm, hmm, bg);
/* Set the profile and null model's target length models */
p7_bg_SetLength (bg, sq->n);
p7_profile_SetLength(gm, sq->n);
/* Allocate DP matrices and tracebacks */
rxf = p7_refmx_Create(gm->M, sq->n);
rxd = p7_refmx_Create(gm->M, sq->n);
tr = p7_trace_Create();
afu = p7_refmx_Create(gm->M, sq->n);
afd = p7_refmx_Create(gm->M, sq->n);
/* First pass analysis */
p7_ReferenceViterbi (sq->dsq, sq->n, gm, rxf, tr, &vsc);
p7_ReferenceForward (sq->dsq, sq->n, gm, rxf, &fsc);
p7_ReferenceBackward(sq->dsq, sq->n, gm, rxd, NULL);
p7_ReferenceDecoding(sq->dsq, sq->n, gm, rxf, rxd, rxd);
/* Customize MPAS parameters if you want; these are the defaults. */
prm.max_iterations = 1000;
prm.loss_threshold = 0.001;
prm.nmax_sampling = FALSE;
prm.be_verbose = FALSE;
/* MPAS algorithm gets us an anchor set */
p7_reference_Anchors(rng, sq->dsq, sq->n, gm, rxf, rxd, tr, &wrk, ah,
afu, afd, anch, &asc, &prm, &stats);
//printf("# ASC Forward UP:\n"); p7_refmx_Dump(stdout, afu);
//printf("# ASC Forward DOWN:\n"); p7_refmx_Dump(stdout, afd);
/* We no longer need rxf and rxd.
* Use their space for apu/apd pair, which will briefly
* hold ASC Backward matrices, then get used for ASC Decoding.
*/
apu = rxf; p7_refmx_Reuse(apu);
apd = rxd; p7_refmx_Reuse(apd);
p7_ReferenceASCBackward(sq->dsq, sq->n, gm, anch->a, anch->D, apu, apd, &asc_b);
//printf("# Backward score (raw, nats): %.2f\n", asc_b);
//printf("# ASC Backward UP:\n"); p7_refmx_Dump(stdout, apu);
//printf("# ASC Backward DOWN:\n"); p7_refmx_Dump(stdout, apd);
/* ASC Decoding takes afu/afd and abu/abd as input;
* overwrites abu/abd with decoding matrices
*/
p7_ReferenceASCDecoding(sq->dsq, sq->n, gm, anch->a, anch->D, afu, afd, apu, apd, apu, apd);
//printf("# ASC Decoding UP matrix:\n"); p7_refmx_Dump(stdout, apu);
//printf("# ASC Decoding DOWN:\n"); p7_refmx_Dump(stdout, apu);
/* Envelope calculation needs to get four matrices:
* ASC Decoding pair, apu/apd, and it will leave these constant;
* ASC Forward pair, afu/afd, and it will overwrite these.
*/
p7_reference_Envelopes(sq->dsq, sq->n, gm, anch->a, anch->D, apu, apd, afu, afd, env);
p7_envelopes_Dump(stdout, env);
p7_envelopes_Destroy(env);
p7_anchorhash_Destroy(ah);
p7_anchors_Destroy(anch);
if (wrk) free(wrk);
p7_trace_Destroy(tr);
p7_refmx_Destroy(afd);
p7_refmx_Destroy(afu);
p7_refmx_Destroy(rxd);
p7_refmx_Destroy(rxf);
esl_sq_Destroy(sq);
p7_profile_Destroy(gm);
p7_bg_Destroy(bg);
p7_hmm_Destroy(hmm);
esl_alphabet_Destroy(abc);
esl_randomness_Destroy(rng);
esl_getopts_Destroy(go);
return 0;
}
