static void
utest_Compare(void)
{
ESL_RANDOMNESS *r = esl_randomness_CreateFast(42);
ESL_ALPHABET *abc = esl_alphabet_Create(eslAMINO);
P7_HMM *hmm = NULL;
P7_BG *bg = NULL;
P7_PROFILE *gm = NULL;
P7_PROFILE *gm2 = NULL;
int M = 200;
int L = 400;
p7_modelsample(r, M, abc, &hmm); /* master and worker's sampled profiles are identical */
bg = p7_bg_Create(abc);
gm = p7_profile_Create(hmm->M, abc);
gm2 = p7_profile_Create(hmm->M, abc);
p7_profile_Config(gm, hmm, bg);
p7_profile_Config(gm2, hmm, bg);
p7_profile_SetLength(gm, L);
p7_profile_SetLength(gm2, L);
if (p7_profile_Compare(gm, gm2, 0.001) != eslOK) p7_Die("identical profile comparison failed");
p7_profile_Destroy(gm);
p7_profile_Destroy(gm2);
p7_bg_Destroy(bg);
p7_hmm_Destroy(hmm);
esl_alphabet_Destroy(abc);
esl_randomness_Destroy(r);
return;
}
/* The MSV score can be validated against Viterbi (provided we trust
* Viterbi), by creating a multihit local profile in which:
* 1. All t_MM scores = 0
* 2. All other core transitions = -inf
* 3. All t_BMk entries uniformly log 2/(M(M+1))
*/
static void
utest_msv(ESL_GETOPTS *go, ESL_RANDOMNESS *r, ESL_ALPHABET *abc, P7_BG *bg, P7_PROFILE *gm, int nseq, int L)
{
P7_PROFILE *g2 = NULL;
ESL_DSQ *dsq = NULL;
P7_GMX *gx = NULL;
float sc1, sc2;
int k, idx;
if ((dsq = malloc(sizeof(ESL_DSQ) *(L+2))) == NULL) esl_fatal("malloc failed");
if ((gx = p7_gmx_Create(gm->M, L)) == NULL) esl_fatal("matrix creation failed");
if ((g2 = p7_profile_Clone(gm)) == NULL) esl_fatal("profile clone failed");
/* Make g2's scores appropriate for simulating the MSV algorithm in Viterbi */
esl_vec_FSet(g2->tsc, p7P_NTRANS * g2->M, -eslINFINITY);
for (k = 1; k < g2->M; k++) p7P_TSC(g2, k, p7P_MM) = 0.0f;
for (k = 0; k < g2->M; k++) p7P_TSC(g2, k, p7P_BM) = log(2.0f / ((float) g2->M * (float) (g2->M+1)));
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_GMSV (dsq, L, gm, gx, 2.0, &sc1) != eslOK) esl_fatal("MSV failed");
if (p7_GViterbi(dsq, L, g2, gx, &sc2) != eslOK) esl_fatal("viterbi failed");
if (fabs(sc1-sc2) > 0.0001) esl_fatal("MSV score not equal to Viterbi score");
}
p7_gmx_Destroy(gx);
p7_profile_Destroy(g2);
free(dsq);
return;
}
/* ViterbiScore() unit test
*
* We can compare these scores to GViterbi() almost exactly; the only
* differences should be negligible roundoff errors. Must convert
* the optimized profile to lspace, though, rather than pspace.
*/
static void
utest_viterbi_score(ESL_RANDOMNESS *r, ESL_ALPHABET *abc, P7_BG *bg, int M, int L, int N)
{
P7_HMM *hmm = NULL;
P7_PROFILE *gm = NULL;
P7_OPROFILE *om = NULL;
ESL_DSQ *dsq = malloc(sizeof(ESL_DSQ) * (L+2));
P7_OMX *ox = p7_omx_Create(M, 0, 0);
P7_GMX *gx = p7_gmx_Create(M, L);
float sc1, sc2;
p7_oprofile_Sample(r, abc, bg, M, L, &hmm, &gm, &om);
p7_oprofile_Logify(om);
while (N--)
{
esl_rsq_xfIID(r, bg->f, abc->K, L, dsq);
p7_ViterbiScore(dsq, L, om, ox, &sc1);
p7_GViterbi (dsq, L, gm, gx, &sc2);
if (fabs(sc1-sc2) > 0.001) esl_fatal("viterbi score unit test failed: scores differ");
}
free(dsq);
p7_hmm_Destroy(hmm);
p7_omx_Destroy(ox);
p7_gmx_Destroy(gx);
p7_profile_Destroy(gm);
p7_oprofile_Destroy(om);
}
int
main(int argc, char **argv)
{
char *msg = "p7_gmx unit test driver failed";
ESL_GETOPTS *go = p7_CreateDefaultApp(options, 0, argc, argv, banner, usage);
ESL_RANDOMNESS *r = esl_randomness_CreateFast(esl_opt_GetInteger(go, "-s"));
ESL_ALPHABET *abc = esl_alphabet_Create(eslAMINO);
P7_BG *bg = p7_bg_Create(abc);
P7_HMM *hmm = NULL;
P7_PROFILE *gm = NULL;
int M = esl_opt_GetInteger(go, "-M");
int L = esl_opt_GetInteger(go, "-L");
float tol = esl_opt_GetReal (go, "-t");
p7_FLogsumInit();
if (p7_hmm_Sample(r, M, abc, &hmm) != eslOK) esl_fatal(msg);
if ((gm = p7_profile_Create(hmm->M, abc)) == NULL) esl_fatal(msg);
if (p7_bg_SetLength(bg, L) != eslOK) esl_fatal(msg);
if (p7_ProfileConfig(hmm, bg, gm, L, p7_UNILOCAL) != eslOK) esl_fatal(msg);
utest_GrowTo();
utest_Compare(r, gm, bg, L, tol);
esl_getopts_Destroy(go);
esl_randomness_Destroy(r);
esl_alphabet_Destroy(abc);
p7_bg_Destroy(bg);
p7_hmm_Destroy(hmm);
p7_profile_Destroy(gm);
return eslOK;
}
/* Function: p7_oprofile_Sample()
* Synopsis: Sample a random profile.
* Incept: MSF Tue Nov 3, 2009 [Janelia]
*
* Purpose: Sample a random profile of <M> nodes for alphabet <abc>,
* using <r> as the source of random numbers. Parameterize
* it for generation of target sequences of mean length
* <L>. Calculate its log-odds scores using background
* model <bg>.
*
* Args: r - random number generator
* abc - emission alphabet
* bg - background frequency model
* M - size of sampled profile, in nodes
* L - configured target seq mean length
* opt_hmm - optRETURN: sampled HMM
* opt_gm - optRETURN: sampled normal profile
* opt_om - RETURN: optimized profile
*
* Returns: <eslOK> on success.
*
* Throws: (no abnormal error conditions)
*/
int
p7_oprofile_Sample(ESL_RANDOMNESS *r, const ESL_ALPHABET *abc, const P7_BG *bg, int M, int L,
P7_HMM **opt_hmm, P7_PROFILE **opt_gm, P7_OPROFILE **ret_om)
{
P7_HMM *hmm = NULL;
P7_PROFILE *gm = NULL;
P7_OPROFILE *om = NULL;
int status;
if ((gm = p7_profile_Create (M, abc)) == NULL) { status = eslEMEM; goto ERROR; }
if ((om = p7_oprofile_Create(M, abc)) == NULL) { status = eslEMEM; goto ERROR; }
if ((status = p7_hmm_Sample(r, M, abc, &hmm)) != eslOK) goto ERROR;
if ((status = p7_ProfileConfig(hmm, bg, gm, L, p7_LOCAL)) != eslOK) goto ERROR;
if ((status = p7_oprofile_Convert(gm, om)) != eslOK) goto ERROR;
if ((status = p7_oprofile_ReconfigLength(om, L)) != eslOK) goto ERROR;
if (opt_hmm != NULL) *opt_hmm = hmm; else p7_hmm_Destroy(hmm);
if (opt_gm != NULL) *opt_gm = gm; else p7_profile_Destroy(gm);
*ret_om = om;
return eslOK;
ERROR:
if (opt_hmm != NULL) *opt_hmm = NULL;
if (opt_gm != NULL) *opt_gm = NULL;
*ret_om = NULL;
return status;
}
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_GMX *gx1 = NULL;
P7_GMX *gx2 = 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;
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);
gx1 = p7_gmx_Create(gm->M, L);
gx2 = p7_gmx_Create(gm->M, L);
esl_rsq_xfIID(r, bg->f, abc->K, L, dsq);
p7_GForward (dsq, L, gm, gx1, &fsc);
p7_GBackward(dsq, L, gm, gx2, &bsc);
p7_GDecoding(gm, gx1, gx2, gx2);
esl_stopwatch_Start(w);
for (i = 0; i < N; i++)
p7_GNull2_ByExpectation(gm, gx2, null2);
esl_stopwatch_Stop(w);
Mcs = (double) N * (double) L * (double) gm->M * 1e-6 / w->user;
esl_stopwatch_Display(stdout, w, "# CPU time: ");
printf("# M = %d\n", gm->M);
printf("# %.1f Mc/s\n", Mcs);
free(dsq);
p7_gmx_Destroy(gx1);
p7_gmx_Destroy(gx2);
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;
}
int
main(int argc, char **argv)
{
ESL_GETOPTS *go = esl_getopts_CreateDefaultApp(options, 0, argc, argv, banner, usage);
ESL_RANDOMNESS *r = esl_randomness_CreateFast(esl_opt_GetInteger(go, "-s"));
ESL_ALPHABET *abc = NULL;
P7_HMM *hmm = NULL;
P7_PROFILE *gm = NULL;
P7_BG *bg = NULL;
int M = 100;
int L = 200;
int nseq = 20;
char errbuf[eslERRBUFSIZE];
if ((abc = esl_alphabet_Create(eslAMINO)) == NULL) esl_fatal("failed to create alphabet");
if (p7_hmm_Sample(r, M, abc, &hmm) != eslOK) esl_fatal("failed to sample an HMM");
if ((bg = p7_bg_Create(abc)) == NULL) esl_fatal("failed to create null model");
if ((gm = p7_profile_Create(hmm->M, abc)) == NULL) esl_fatal("failed to create profile");
if (p7_ProfileConfig(hmm, bg, gm, L, p7_LOCAL) != eslOK) esl_fatal("failed to config profile");
if (p7_hmm_Validate (hmm, errbuf, 0.0001) != eslOK) esl_fatal("whoops, HMM is bad!: %s", errbuf);
if (p7_profile_Validate(gm, errbuf, 0.0001) != eslOK) esl_fatal("whoops, profile is bad!: %s", errbuf);
utest_basic (go);
utest_viterbi(go, r, abc, bg, gm, nseq, L);
p7_profile_Destroy(gm);
p7_bg_Destroy(bg);
p7_hmm_Destroy(hmm);
esl_alphabet_Destroy(abc);
esl_randomness_Destroy(r);
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)
{
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;
}
/* Function: p7_Builder()
* Synopsis: Build a new HMM from an MSA.
*
* Purpose: Take the multiple sequence alignment <msa> and a build configuration <bld>,
* and build a new HMM.
*
* Effective sequence number determination and calibration steps require
* additionally providing a null model <bg>.
*
* Args: bld - build configuration
* msa - multiple sequence alignment
* bg - null model
* opt_hmm - optRETURN: new HMM
* opt_trarr - optRETURN: array of faux tracebacks, <0..nseq-1>
* opt_gm - optRETURN: profile corresponding to <hmm>
* opt_om - optRETURN: optimized profile corresponding to <gm>
* opt_postmsa - optRETURN: RF-annotated, possibly modified MSA
*
* Returns: <eslOK> on success. The new HMM is optionally returned in
* <*opt_hmm>, along with optional returns of an array of faux tracebacks
* for each sequence in <*opt_trarr>, the annotated MSA used to construct
* the model in <*opt_postmsa>, a configured search profile in
* <*opt_gm>, and an optimized search profile in <*opt_om>. These are
* all optional returns because the caller may, for example, be interested
* only in an optimized profile, or may only be interested in the HMM.
*
* Returns <eslENORESULT> if no consensus columns were annotated.
* Returns <eslEFORMAT> on MSA format problems, such as a missing RF annotation
* line in hand architecture construction. On any returned error,
* <bld->errbuf> contains an informative error message.
*
* Throws: <eslEMEM> on allocation error.
* <eslEINVAL> if relative weights couldn't be calculated from <msa>.
*
* Xref: J4/30.
*/
int
p7_Builder(P7_BUILDER *bld, ESL_MSA *msa, P7_BG *bg,
P7_HMM **opt_hmm, P7_TRACE ***opt_trarr, P7_PROFILE **opt_gm, P7_OPROFILE **opt_om,
ESL_MSA **opt_postmsa)
{
int i,j;
uint32_t checksum = 0; /* checksum calculated for the input MSA. hmmalign --mapali verifies against this. */
P7_HMM *hmm = NULL;
P7_TRACE **tr = NULL;
P7_TRACE ***tr_ptr = (opt_trarr != NULL || opt_postmsa != NULL) ? &tr : NULL;
int status;
if ((status = validate_msa (bld, msa)) != eslOK) goto ERROR;
if ((status = esl_msa_Checksum (msa, &checksum)) != eslOK) ESL_XFAIL(status, bld->errbuf, "Failed to calculate checksum");
if ((status = relative_weights (bld, msa)) != eslOK) goto ERROR;
if ((status = esl_msa_MarkFragments(msa, bld->fragthresh)) != eslOK) goto ERROR;
if ((status = build_model (bld, msa, &hmm, tr_ptr)) != eslOK) goto ERROR;
//Ensures that the weighted-average I->I count <= bld->max_insert_len
//(MI currently contains the number of observed insert-starts)
if (bld->max_insert_len>0)
for (i=1; i<hmm->M; i++ )
hmm->t[i][p7H_II] = ESL_MIN(hmm->t[i][p7H_II], bld->max_insert_len*hmm->t[i][p7H_MI]);
if ((status = effective_seqnumber (bld, msa, hmm, bg)) != eslOK) goto ERROR;
if ((status = parameterize (bld, hmm)) != eslOK) goto ERROR;
if ((status = annotate (bld, msa, hmm)) != eslOK) goto ERROR;
if ((status = calibrate (bld, hmm, bg, opt_gm, opt_om)) != eslOK) goto ERROR;
if ((status = make_post_msa (bld, msa, hmm, tr, opt_postmsa)) != eslOK) goto ERROR;
//force masked positions to background (it'll be close already, so no relevant impact on weighting)
if (hmm->mm != NULL)
for (i=1; i<hmm->M; i++ )
if (hmm->mm[i] == 'm')
for (j=0; j<hmm->abc->K; j++)
hmm->mat[i][j] = bg->f[j];
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;
}
hmm->checksum = checksum;
hmm->flags |= p7H_CHKSUM;
if (opt_hmm != NULL) *opt_hmm = hmm; else p7_hmm_Destroy(hmm);
if (opt_trarr != NULL) *opt_trarr = tr; else p7_trace_DestroyArray(tr, msa->nseq);
return eslOK;
ERROR:
p7_hmm_Destroy(hmm);
p7_trace_DestroyArray(tr, msa->nseq);
if (opt_gm != NULL) p7_profile_Destroy(*opt_gm);
if (opt_om != NULL) p7_oprofile_Destroy(*opt_om);
return status;
}
static void
utest_oprofileSendRecv(int my_rank, int nproc)
{
ESL_RANDOMNESS *r = esl_randomness_CreateFast(42);
ESL_ALPHABET *abc = esl_alphabet_Create(eslAMINO);
P7_HMM *hmm = NULL;
P7_BG *bg = NULL;
P7_PROFILE *gm = NULL;
P7_OPROFILE *om = NULL;
P7_OPROFILE *om2 = NULL;
int M = 200;
int L = 400;
char *wbuf = NULL;
int wn = 0;
int i;
char errbuf[eslERRBUFSIZE];
p7_hmm_Sample(r, M, abc, &hmm); /* master and worker's sampled profiles are identical */
bg = p7_bg_Create(abc);
gm = p7_profile_Create(hmm->M, abc);
om = p7_oprofile_Create(hmm->M, abc);
p7_ProfileConfig(hmm, bg, gm, L, p7_LOCAL);
p7_oprofile_Convert(gm, om);
p7_bg_SetLength (bg, L);
if (my_rank == 0)
{
for (i = 1; i < nproc; i++)
{
ESL_DPRINTF1(("Master: receiving test profile\n"));
p7_oprofile_MPIRecv(MPI_ANY_SOURCE, 0, MPI_COMM_WORLD, &wbuf, &wn, &abc, &om2);
ESL_DPRINTF1(("Master: test profile received\n"));
if (p7_oprofile_Compare(om, om2, 0.001, errbuf) != eslOK)
p7_Die("Received profile not identical to what was sent\n%s", errbuf);
p7_oprofile_Destroy(om2);
}
}
else
{
ESL_DPRINTF1(("Worker %d: sending test profile\n", my_rank));
p7_oprofile_MPISend(om, 0, 0, MPI_COMM_WORLD, &wbuf, &wn);
ESL_DPRINTF1(("Worker %d: test profile sent\n", my_rank));
}
free(wbuf);
p7_profile_Destroy(gm);
p7_oprofile_Destroy(om);
p7_bg_Destroy(bg);
p7_hmm_Destroy(hmm);
esl_alphabet_Destroy(abc);
esl_randomness_Destroy(r);
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;
}
/*
* compare to GForward() scores.
*/
static void
utest_fwdback(ESL_RANDOMNESS *r, ESL_ALPHABET *abc, P7_BG *bg, int M, int L, int N)
{
char *msg = "forward/backward unit test failed";
P7_HMM *hmm = NULL;
P7_PROFILE *gm = NULL;
P7_OPROFILE *om = NULL;
ESL_DSQ *dsq = malloc(sizeof(ESL_DSQ) * (L+2));
P7_OMX *fwd = p7_omx_Create(M, 0, L);
P7_OMX *bck = p7_omx_Create(M, 0, L);
P7_OMX *oxf = p7_omx_Create(M, L, L);
P7_OMX *oxb = p7_omx_Create(M, L, L);
P7_GMX *gx = p7_gmx_Create(M, L);
float tolerance;
float fsc1, fsc2;
float bsc1, bsc2;
float generic_sc;
p7_FLogsumInit();
if (p7_FLogsumError(-0.4, -0.5) > 0.0001) tolerance = 1.0; /* weaker test against GForward() */
else tolerance = 0.0001; /* stronger test: FLogsum() is in slow exact mode. */
p7_oprofile_Sample(r, abc, bg, M, L, &hmm, &gm, &om);
while (N--)
{
esl_rsq_xfIID(r, bg->f, abc->K, L, dsq);
p7_Forward (dsq, L, om, oxf, &fsc1);
p7_Backward (dsq, L, om, oxf, oxb, &bsc1);
p7_ForwardParser (dsq, L, om, fwd, &fsc2);
p7_BackwardParser(dsq, L, om, fwd, bck, &bsc2);
p7_GForward (dsq, L, gm, gx, &generic_sc);
/* Forward and Backward scores should agree with high tolerance */
if (fabs(fsc1-bsc1) > 0.0001) esl_fatal(msg);
if (fabs(fsc2-bsc2) > 0.0001) esl_fatal(msg);
if (fabs(fsc1-fsc2) > 0.0001) esl_fatal(msg);
/* GForward scores should approximate Forward scores,
* with tolerance that depends on how logsum.c was compiled
*/
if (fabs(fsc1-generic_sc) > tolerance) esl_fatal(msg);
}
free(dsq);
p7_hmm_Destroy(hmm);
p7_omx_Destroy(oxb);
p7_omx_Destroy(oxf);
p7_omx_Destroy(bck);
p7_omx_Destroy(fwd);
p7_gmx_Destroy(gx);
p7_profile_Destroy(gm);
p7_oprofile_Destroy(om);
}
/* The Config test simply makes sure a random profile passes
* a Validate() check.
*/
static void
utest_Config(P7_HMM *hmm, P7_BG *bg)
{
char *msg = "modelconfig.c::p7_ProfileConfig() unit test failed";
P7_PROFILE *gm = NULL;
if ((gm = p7_profile_Create(hmm->M, hmm->abc)) == NULL) esl_fatal(msg);
if (p7_ProfileConfig(hmm, bg, gm, 350, p7_LOCAL) != eslOK) esl_fatal(msg);
if (p7_profile_Validate(gm, NULL, 0.0001) != eslOK) esl_fatal(msg);
p7_profile_Destroy(gm);
return;
}
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;
}
/* compare results to GDecoding(). */
static void
utest_null2_expectation(ESL_RANDOMNESS *r, ESL_ALPHABET *abc, P7_BG *bg, int M, int L, int N, float tolerance)
{
char *msg = "decoding unit test failed";
P7_HMM *hmm = NULL;
P7_PROFILE *gm = NULL;
P7_OPROFILE *om = NULL;
ESL_DSQ *dsq = malloc(sizeof(ESL_DSQ) * (L+2));
P7_OMX *fwd = p7_omx_Create(M, L, L);
P7_OMX *bck = p7_omx_Create(M, L, L);
P7_OMX *pp = p7_omx_Create(M, L, L);
P7_GMX *gxf = p7_gmx_Create(M, L);
P7_GMX *gxb = p7_gmx_Create(M, L);
P7_GMX *gpp = p7_gmx_Create(M, L);
float *on2 = malloc(sizeof(float) * abc->Kp);
float *gn2 = malloc(sizeof(float) * abc->Kp);
float fsc1, fsc2;
float bsc1, bsc2;
if (!gn2 || !on2) esl_fatal(msg);
if (p7_oprofile_Sample(r, abc, bg, M, L, &hmm, &gm, &om) != eslOK) esl_fatal(msg);
while (N--)
{
if (esl_rsq_xfIID(r, bg->f, abc->K, L, dsq) != eslOK) esl_fatal(msg);
if (p7_Forward (dsq, L, om, fwd, &fsc1) != eslOK) esl_fatal(msg);
if (p7_Backward (dsq, L, om, fwd, bck, &bsc1) != eslOK) esl_fatal(msg);
if (p7_Decoding(om, fwd, bck, pp) != eslOK) esl_fatal(msg);
if (p7_Null2_ByExpectation(om, pp, on2) != eslOK) esl_fatal(msg);
if (p7_GForward (dsq, L, gm, gxf, &fsc2) != eslOK) esl_fatal(msg);
if (p7_GBackward(dsq, L, gm, gxb, &bsc2) != eslOK) esl_fatal(msg);
if (p7_GDecoding(gm, gxf, gxb, gpp) != eslOK) esl_fatal(msg);
if (p7_GNull2_ByExpectation(gm, gpp, gn2) != eslOK) esl_fatal(msg);
if (esl_vec_FCompare(gn2, on2, abc->Kp, tolerance) != eslOK) esl_fatal(msg);
}
p7_gmx_Destroy(gpp);
p7_gmx_Destroy(gxf);
p7_gmx_Destroy(gxb);
p7_omx_Destroy(pp);
p7_omx_Destroy(fwd);
p7_omx_Destroy(bck);
free(on2);
free(gn2);
free(dsq);
p7_oprofile_Destroy(om);
p7_profile_Destroy(gm);
p7_hmm_Destroy(hmm);
}
/* Function: p7_profile_Clone()
* Synopsis: Duplicates a profile.
*
* Purpose: Duplicate profile <gm>; return a pointer
* to the newly allocated copy.
*
* Returns: ptr to new clone.
*
* Throws: <NULL> on allocation failure.
*/
P7_PROFILE *
p7_profile_Clone(const P7_PROFILE *gm)
{
P7_PROFILE *g2 = NULL;
int status;
if ((g2 = p7_profile_Create(gm->allocM, gm->abc)) == NULL) return NULL;
if ((status = p7_profile_Copy(gm, g2)) != eslOK) goto ERROR;
return g2;
ERROR:
p7_profile_Destroy(g2);
return NULL;
}
/* Function: p7_oprofile_Copy()
* Synopsis: Allocate an optimized profile structure.
* Incept: MSF Tue Nov 3, 2009 [Janelia]
*
* Purpose: Allocate for profiles of up to <allocM> nodes for digital alphabet <abc>.
*
* Throws: <NULL> on allocation error.
*/
P7_OPROFILE *
p7_oprofile_Copy(P7_OPROFILE *om1)
{
P7_PROFILE *dst = NULL;
if ((dst = p7_profile_Create(om1->M, om1->abc)) == NULL) goto ERROR;
if ((p7_profile_Copy(om1, dst)) != eslOK) goto ERROR;
return dst;
ERROR:
p7_profile_Destroy(dst);
return NULL;
}
/* compare results to GDecoding(). */
static void
utest_decoding(ESL_RANDOMNESS *r, ESL_ALPHABET *abc, P7_BG *bg, int M, int L, int N, float tolerance)
{
char *msg = "decoding unit test failed";
P7_HMM *hmm = NULL;
P7_PROFILE *gm = NULL;
P7_OPROFILE *om = NULL;
ESL_DSQ *dsq = malloc(sizeof(ESL_DSQ) * (L+2));
P7_OMX *fwd = p7_omx_Create(M, L, L);
P7_OMX *bck = p7_omx_Create(M, L, L);
P7_OMX *pp = p7_omx_Create(M, L, L);
P7_GMX *gxf = p7_gmx_Create(M, L);
P7_GMX *gxb = p7_gmx_Create(M, L);
P7_GMX *gxp1 = p7_gmx_Create(M, L);
P7_GMX *gxp2 = p7_gmx_Create(M, L);
float fsc1, fsc2;
float bsc1, bsc2;
if (p7_oprofile_Sample(r, abc, bg, M, L, &hmm, &gm, &om) != eslOK) esl_fatal(msg);
while (N--)
{
if (esl_rsq_xfIID(r, bg->f, abc->K, L, dsq) != eslOK) esl_fatal(msg);
if (p7_Forward (dsq, L, om, fwd, &fsc1) != eslOK) esl_fatal(msg);
if (p7_Backward (dsq, L, om, fwd, bck, &bsc1) != eslOK) esl_fatal(msg);
if (p7_Decoding(om, fwd, bck, pp) != eslOK) esl_fatal(msg);
if (p7_omx_FDeconvert(pp, gxp1) != eslOK) esl_fatal(msg);
if (p7_GForward (dsq, L, gm, gxf, &fsc2) != eslOK) esl_fatal(msg);
if (p7_GBackward(dsq, L, gm, gxb, &bsc2) != eslOK) esl_fatal(msg);
if (p7_GDecoding(gm, gxf, gxb, gxp2) != eslOK) esl_fatal(msg);
// p7_gmx_Dump(stdout, gxp1, p7_DEFAULT);
// p7_gmx_Dump(stdout, gxp2, p7_DEFAULT);
if (p7_gmx_Compare(gxp1, gxp2, tolerance) != eslOK) esl_fatal(msg);
}
p7_gmx_Destroy(gxp1);
p7_gmx_Destroy(gxp2);
p7_gmx_Destroy(gxf);
p7_gmx_Destroy(gxb);
p7_omx_Destroy(fwd);
p7_omx_Destroy(bck);
p7_omx_Destroy(pp);
free(dsq);
p7_oprofile_Destroy(om);
p7_profile_Destroy(gm);
p7_hmm_Destroy(hmm);
}
/* 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;
}
void destroy_hmmer_wrapper() {
int index;
if(models != NULL) {
for(index = 0;index < num_models;index++) {
p7_oprofile_Destroy(models[index]);
p7_profile_Destroy(gmodels[index]);
}
free(models);
free(gmodels);
}
if(wrapper_results != NULL) {
for(index = 0;index < num_models;index++) {
destroy_result(wrapper_results[index]);
}
free(wrapper_results);
}
if(bg != NULL) {
p7_bg_Destroy(bg);
}
if(hmm_fp != NULL) {
p7_hmmfile_Close(hmm_fp);
}
if(oxf) {
p7_omx_Destroy(oxf);
}
if(oxb) {
p7_omx_Destroy(oxb);
}
if(gxf) {
p7_gmx_Destroy(gxf);
}
if(gxb) {
p7_gmx_Destroy(gxb);
}
if(abc) {
esl_alphabet_Destroy(abc);
}
if(tr) {
p7_trace_Destroy(tr);
}
}
int
main(int argc, char **argv)
{
ESL_GETOPTS *go = esl_getopts_CreateDefaultApp(options, 0, argc, argv, banner, usage);
ESL_RANDOMNESS *r = esl_randomness_CreateFast(esl_opt_GetInteger(go, "-s"));
ESL_ALPHABET *abc = NULL;
P7_HMM *hmm = NULL;
P7_PROFILE *gm = NULL;
P7_OPROFILE *om = NULL;
P7_BG *bg = NULL;
ESL_DSQ *dsq = NULL;
ESL_SQ *sq = NULL;
int M = 6;
int L = 10;
int ntrace = 1000;
if ((abc = esl_alphabet_Create(eslAMINO)) == NULL) esl_fatal("failed to create alphabet");
if (p7_hmm_Sample(r, M, abc, &hmm) != eslOK) esl_fatal("failed to sample an HMM");
if ((bg = p7_bg_Create(abc)) == NULL) esl_fatal("failed to create null model");
if ((gm = p7_profile_Create(hmm->M, abc)) == NULL) esl_fatal("failed to create profile");
if (p7_ProfileConfig(hmm, bg, gm, L, p7_LOCAL) != eslOK) esl_fatal("failed to config profile");
if ((om = p7_oprofile_Create(gm->M, abc)) == NULL) esl_fatal("failed to create optimized profile");
if (p7_oprofile_Convert(gm, om) != eslOK) esl_fatal("failed to convert profile");
/* Test with randomly generated (iid) sequence */
if ((dsq = malloc(sizeof(ESL_DSQ) *(L+2))) == NULL) esl_fatal("malloc failed");
if (esl_rsq_xfIID(r, bg->f, abc->K, L, dsq) != eslOK) esl_fatal("seq generation failed");
utest_stotrace(go, r, abc, gm, om, dsq, L, ntrace);
/* Test with seq sampled from profile */
if ((sq = esl_sq_CreateDigital(abc)) == NULL) esl_fatal("sequence allocation failed");
if (p7_ProfileEmit(r, hmm, gm, bg, sq, NULL) != eslOK) esl_fatal("profile emission failed");
utest_stotrace(go, r, abc, gm, om, sq->dsq, sq->n, ntrace);
esl_sq_Destroy(sq);
free(dsq);
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_getopts_Destroy(go);
return 0;
}
int
main(int argc, char **argv)
{
char *hmmfile = argv[1];
ESL_ALPHABET *abc = NULL;
P7_HMMFILE *hfp = NULL;
P7_HMM *hmm = NULL;
P7_BG *bg = NULL;
P7_PROFILE *gm = NULL;
P7_OPROFILE *om1 = NULL;
P7_OPROFILE *om2 = NULL;
int status;
char errmsg[512];
status = p7_hmmfile_Open(hmmfile, NULL, &hfp);
if (status == eslENOTFOUND) esl_fatal("Failed to open HMM file %s for reading.\n", hmmfile);
else if (status == eslEFORMAT) esl_fatal("File %s does not appear to be in a recognized HMM format.\n", hmmfile);
else if (status != eslOK) esl_fatal("Unexpected error %d in opening HMM file %s.\n", status, hmmfile);
status = p7_hmmfile_Read(hfp, &abc, &hmm);
if (status == eslEFORMAT) esl_fatal("Bad file format in HMM file %s:\n%s\n", hfp->fname, hfp->errbuf);
else if (status == eslEINCOMPAT) esl_fatal("HMM in %s is not in the expected %s alphabet\n", hfp->fname, esl_abc_DecodeType(abc->type));
else if (status == eslEOF) esl_fatal("Empty HMM file %s? No HMM data found.\n", hfp->fname);
else if (status != eslOK) esl_fatal("Unexpected error in reading HMMs from %s\n", hfp->fname);
bg = p7_bg_Create(abc);
gm = p7_profile_Create(hmm->M, abc);
om1 = p7_oprofile_Create(hmm->M, abc);
p7_ProfileConfig(hmm, bg, gm, 400, p7_LOCAL);
p7_oprofile_Convert(gm, om1);
om2 = p7_oprofile_Copy(om1);
if (p7_oprofile_Compare(om1, om2, 0.001f, errmsg) != eslOK) esl_fatal("Compare failed %s\n", errmsg);
p7_oprofile_Destroy(om1);
p7_profile_Destroy(gm);
p7_bg_Destroy(bg);
p7_hmm_Destroy(hmm);
p7_hmmfile_Close(hfp);
esl_alphabet_Destroy(abc);
return eslOK;
}
/* Function: p7_hmmcache_Close()
* Synopsis: Free a profile cache.
*/
void
p7_hmmcache_Close(P7_HMMCACHE *cache)
{
int i;
if (cache)
{
if (cache->name) free(cache->name);
if (cache->abc) esl_alphabet_Destroy(cache->abc);
for (i = 0; i < cache->n; i++)
{
if (cache->gmlist) p7_profile_Destroy (cache->gmlist[i]);
if (cache->omlist) p7_oprofile_Destroy(cache->omlist[i]);
}
if (cache->gmlist) free(cache->gmlist);
if (cache->omlist) free(cache->omlist);
free(cache);
}
}
/* ViterbiFilter() unit test
*
* We can check that scores are identical (within machine error) to
* scores of generic DP with scores rounded the same way. Do this for
* a random model of length <M>, for <N> test sequences of length <L>.
*
* We assume that we don't accidentally generate a high-scoring random
* sequence that overflows ViterbiFilter()'s limited range.
*
*/
static void
utest_viterbi_filter(ESL_RANDOMNESS *r, ESL_ALPHABET *abc, P7_BG *bg, int M, int L, int N)
{
P7_HMM *hmm = NULL;
P7_PROFILE *gm = NULL;
P7_OPROFILE *om = NULL;
ESL_DSQ *dsq = malloc(sizeof(ESL_DSQ) * (L+2));
P7_OMX *ox = p7_omx_Create(M, 0, 0);
P7_GMX *gx = p7_gmx_Create(M, L);
float sc1, sc2;
p7_oprofile_Sample(r, abc, bg, M, L, &hmm, &gm, &om);
p7_profile_SameAsVF(om, gm); /* round and scale the scores in <gm> the same as in <om> */
#if 0
p7_oprofile_Dump(stdout, om); // dumps the optimized profile
p7_omx_SetDumpMode(stdout, ox, TRUE); // makes the fast DP algorithms dump their matrices
#endif
while (N--)
{
esl_rsq_xfIID(r, bg->f, abc->K, L, dsq);
p7_ViterbiFilter(dsq, L, om, ox, &sc1);
p7_GViterbi (dsq, L, gm, gx, &sc2);
#if 0
p7_gmx_Dump(stdout, gx); // dumps a generic DP matrix
#endif
sc2 /= om->scale_w;
sc2 -= 3.0;
if (fabs(sc1-sc2) > 0.001) esl_fatal("viterbi filter unit test failed: scores differ (%.2f, %.2f)", sc1, sc2);
}
free(dsq);
p7_hmm_Destroy(hmm);
p7_omx_Destroy(ox);
p7_gmx_Destroy(gx);
p7_profile_Destroy(gm);
p7_oprofile_Destroy(om);
}
int
main(int argc, char **argv)
{
ESL_GETOPTS *go = p7_CreateDefaultApp(options, 0, argc, argv, banner, usage);
ESL_RANDOMNESS *r = esl_randomness_CreateFast(esl_opt_GetInteger(go, "-s"));
ESL_ALPHABET *abc = esl_alphabet_Create(eslAMINO);
P7_HMM *hmm = NULL;
P7_BG *bg = NULL;
P7_PROFILE *gm = NULL;
P7_GMX *fwd = NULL;
P7_GMX *bck = NULL;
ESL_DSQ *dsq = NULL;
int M = esl_opt_GetInteger(go, "-M");
int L = esl_opt_GetInteger(go, "-L");
/* Sample a random HMM */
p7_hmm_Sample(r, M, abc, &hmm);
/* Configure a profile from the sampled 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);
/* Other initial allocations */
dsq = malloc(sizeof(ESL_DSQ) * (L+2));
fwd = p7_gmx_Create(gm->M, L);
bck = p7_gmx_Create(gm->M, L);
p7_FLogsumInit();
utest_correct_normalization(r, gm, bg, dsq, L, fwd, bck);
free(dsq);
p7_gmx_Destroy(fwd);
p7_gmx_Destroy(bck);
p7_profile_Destroy(gm);
p7_bg_Destroy(bg);
p7_hmm_Destroy(hmm);
esl_alphabet_Destroy(abc);
esl_randomness_Destroy(r);
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_ALPHABET *abc = NULL;
P7_HMMFILE *hfp = NULL;
P7_HMM *hmm = NULL;
P7_BG *bg = NULL;
P7_PROFILE *gm = NULL;
P7_OPROFILE *om = NULL;
int L = esl_opt_GetInteger(go, "-L");
int N = esl_opt_GetInteger(go, "-N");
int i;
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_LOCAL);
om = p7_oprofile_Create(gm->M, abc);
esl_stopwatch_Start(w);
for (i = 0; i < N; i++)
p7_oprofile_Convert(gm, om);
esl_stopwatch_Stop(w);
esl_stopwatch_Display(stdout, w, "# CPU time: ");
printf("# M = %d\n", gm->M);
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_getopts_Destroy(go);
return 0;
}
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_ALPHABET *abc = NULL;
P7_HMMFILE *hfp = NULL;
P7_HMM *hmm = NULL;
P7_BG *bg = NULL;
P7_PROFILE *gm = NULL;
float ftol = 1e-4; /* floating-point tolerance for checking parameters against expected probs or summing to 1 */
char errbuf[eslERRBUFSIZE];
/* Read in one HMM; sets alphabet to the HMM's alphabet */
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);
/* Set up a null model */
bg = p7_bg_Create(abc);
/* Allocate and configure a profile from HMM and null model */
gm = p7_profile_Create(hmm->M, abc);
p7_profile_Config(gm, hmm, bg);
p7_profile_SetLength(gm, 400); /* 400 is arbitrary here; this is whatever your target seq length L is */
printf("profile memory consumed: %" PRId64 " bytes\n", (int64_t) p7_profile_Sizeof(gm));
/* Debugging tools allow dumping, validating the object */
if (p7_profile_Validate(gm, errbuf, ftol) != eslOK) p7_Fail("profile validation failed\n %s\n", errbuf);
if (esl_opt_GetBoolean(go, "--vv"))
p7_profile_Dump(stdout, gm);
p7_profile_Destroy(gm);
p7_bg_Destroy(bg);
p7_hmm_Destroy(hmm);
esl_alphabet_Destroy(abc);
esl_getopts_Destroy(go);
return 0;
}
/* recalibrate_model()
*
* Optionally, user can choose (with --recal) to replace the
* statistical parameters of the HMM. The calibrated params are used
* to generate "expected" distributions in output plots. see
* p7_Calibrate(), which this is a partial copy of. (p7_Calibrate()
* requires a P7_BUILDER object that we don't have.)
*
* On success, returns <eslOK> and the statistical parameters of the
* <hmm> have been recalibrated and replaced.
*/
static int
recalibrate_model(ESL_GETOPTS *go, struct cfg_s *cfg, char *errbuf, P7_HMM *hmm)
{
P7_PROFILE *gm = NULL;
P7_OPROFILE *om = NULL;
int EmL = esl_opt_GetInteger(go, "--EmL");
int EmN = esl_opt_GetInteger(go, "--EmN");
int EvL = esl_opt_GetInteger(go, "--EvL");
int EvN = esl_opt_GetInteger(go, "--EvN");
int EfL = esl_opt_GetInteger(go, "--EfL");
int EfN = esl_opt_GetInteger(go, "--EfN");
double Eft = esl_opt_GetReal (go, "--Eft");
double lambda, mmu, vmu, ftau;
gm = p7_profile_Create(hmm->M, cfg->abc);
p7_profile_Config(gm, hmm, cfg->bg); /* dual-mode multihit; L=0 (no length model needed right now) */
P7_HARDWARE *hw;
if ((hw = p7_hardware_Create ()) == NULL) p7_Fail("Couldn't get HW information data structure");
om = p7_oprofile_Create(gm->M, cfg->abc, hw->simd);
p7_oprofile_Convert(gm, om); /* om is now *local* multihit */
p7_Lambda(hmm, cfg->bg, &lambda);
p7_MSVMu (cfg->r, om, cfg->bg, EmL, EmN, lambda, &mmu);
p7_ViterbiMu(cfg->r, om, cfg->bg, EvL, EvN, lambda, &vmu);
p7_Tau (cfg->r, om, cfg->bg, EfL, EfN, lambda, Eft, &ftau);
hmm->evparam[p7_MLAMBDA] = lambda;
hmm->evparam[p7_VLAMBDA] = lambda;
hmm->evparam[p7_FLAMBDA] = lambda;
hmm->evparam[p7_MMU] = mmu;
hmm->evparam[p7_VMU] = vmu;
hmm->evparam[p7_FTAU] = ftau;
hmm->flags |= p7H_STATS;
p7_profile_Destroy(gm);
p7_oprofile_Destroy(om);
return eslOK;
}
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;
}
