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;
}
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;
}
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;
}
/* 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);
}
/* 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;
}
/*
* 1. Compare accscore to GOptimalAccuracy().
* 2. Compare trace to GOATrace().
*
* Note: This test is subject to some expected noise and can fail
* for entirely innocent reasons. Generic Forward/Backward calculations with
* p7_GForward(), p7_GBackward() use coarse-grain table lookups to sum
* log probabilities, and sufficient roundoff error can accumulate to
* change the optimal accuracy traceback, causing this test to fail.
* So, if optacc_utest fails, before you go looking for bugs, first
* go to ../logsum.c, change the #ifdef to activate the slow/accurate
* version, recompile and rerun optacc_utest. If the failure goes away,
* you can ignore it. - SRE, Wed Dec 17 09:45:31 2008
*/
static void
utest_optacc(ESL_GETOPTS *go, ESL_RANDOMNESS *r, ESL_ALPHABET *abc, P7_BG *bg, int M, int L, int N)
{
char *msg = "optimal accuracy unit test failed";
P7_HMM *hmm = NULL;
P7_PROFILE *gm = NULL;
P7_OPROFILE *om = NULL;
ESL_SQ *sq = esl_sq_CreateDigital(abc);
P7_OMX *ox1 = p7_omx_Create(M, L, L);
P7_OMX *ox2 = p7_omx_Create(M, L, L);
P7_GMX *gx1 = p7_gmx_Create(M, L);
P7_GMX *gx2 = p7_gmx_Create(M, L);
P7_TRACE *tr = p7_trace_CreateWithPP();
P7_TRACE *trg = p7_trace_CreateWithPP();
P7_TRACE *tro = p7_trace_CreateWithPP();
float accscore_o;
float fsc, bsc, accscore;
float fsc_g, bsc_g, accscore_g, accscore_g2;
float pptol = 0.01;
float sctol = 0.001;
float gtol;
p7_FLogsumInit();
gtol = ( (p7_FLogsumError(-0.4, -0.5) > 0.0001) ? 0.1 : 0.001);
if (p7_oprofile_Sample(r, abc, bg, M, L, &hmm, &gm, &om)!= eslOK) esl_fatal(msg);
while (N--)
{
if (p7_ProfileEmit(r, hmm, gm, bg, sq, tro) != eslOK) esl_fatal(msg);
if (p7_omx_GrowTo(ox1, M, sq->n, sq->n) != eslOK) esl_fatal(msg);
if (p7_omx_GrowTo(ox2, M, sq->n, sq->n) != eslOK) esl_fatal(msg);
if (p7_gmx_GrowTo(gx1, M, sq->n) != eslOK) esl_fatal(msg);
if (p7_gmx_GrowTo(gx2, M, sq->n) != eslOK) esl_fatal(msg);
if (p7_Forward (sq->dsq, sq->n, om, ox1, &fsc) != eslOK) esl_fatal(msg);
if (p7_Backward(sq->dsq, sq->n, om, ox1, ox2, &bsc) != eslOK) esl_fatal(msg);
if (p7_Decoding(om, ox1, ox2, ox2) != eslOK) esl_fatal(msg);
if (p7_OptimalAccuracy(om, ox2, ox1, &accscore) != eslOK) esl_fatal(msg);
#if 0
p7_omx_FDeconvert(ox1, gx1);
p7_gmx_Dump(stdout, gx1, p7_DEFAULT);
p7_omx_FDeconvert(ox2, gx1);
p7_gmx_Dump(stdout, gx1, p7_DEFAULT);
#endif
if (p7_OATrace(om, ox2, ox1, tr) != eslOK) esl_fatal(msg);
if (p7_GForward (sq->dsq, sq->n, gm, gx1, &fsc_g) != eslOK) esl_fatal(msg);
if (p7_GBackward(sq->dsq, sq->n, gm, gx2, &bsc_g) != eslOK) esl_fatal(msg);
#if 0
p7_gmx_Dump(stdout, gx1, p7_DEFAULT); /* fwd */
p7_gmx_Dump(stdout, gx2, p7_DEFAULT); /* bck */
#endif
if (p7_GDecoding(gm, gx1, gx2, gx2) != eslOK) esl_fatal(msg);
if (p7_GOptimalAccuracy(gm, gx2, gx1, &accscore_g) != eslOK) esl_fatal(msg);
#if 0
p7_gmx_Dump(stdout, gx1, p7_DEFAULT); /* oa */
p7_gmx_Dump(stdout, gx2, p7_DEFAULT); /* pp */
#endif
if (p7_GOATrace(gm, gx2, gx1, trg) != eslOK) esl_fatal(msg);
if (p7_trace_SetPP(tro, gx2) != eslOK) esl_fatal(msg);
if (esl_opt_GetBoolean(go, "--traces"))
{
p7_trace_Dump(stdout, tro, gm, sq->dsq);
p7_trace_Dump(stdout, tr, gm, sq->dsq);
p7_trace_Dump(stdout, trg, gm, sq->dsq);
}
if (p7_trace_Validate(tr, abc, sq->dsq, NULL) != eslOK) esl_fatal(msg);
if (p7_trace_Validate(trg, abc, sq->dsq, NULL) != eslOK) esl_fatal(msg);
if (p7_trace_Compare(tr, trg, pptol) != eslOK) esl_fatal(msg);
accscore_o = p7_trace_GetExpectedAccuracy(tro); /* according to gx2; see p7_trace_SetPP() call above */
accscore_g2 = p7_trace_GetExpectedAccuracy(trg);
#if 0
printf("%f %f %f %f\n", accscore, accscore_g, accscore_g2, accscore_o);
#endif
if (esl_FCompare(fsc, bsc, sctol) != eslOK) esl_fatal(msg);
if (esl_FCompare(fsc_g, bsc_g, gtol) != eslOK) esl_fatal(msg);
if (esl_FCompare(fsc, fsc_g, gtol) != eslOK) esl_fatal(msg);
if (esl_FCompare(accscore, accscore_g, gtol) != eslOK) esl_fatal(msg);
if (esl_FCompare(accscore_g, accscore_g2, gtol) != eslOK) esl_fatal(msg);
if (accscore_g2 < accscore_o) esl_fatal(msg);
/* the above deserves explanation:
* - accscore_o is the accuracy of the originally emitted trace, according
* to the generic posterior decoding matrix <gx2>. This is a lower bound
* on the expected # of accurately aligned residues found by a DP
* optimization.
* - accscore is the accuracy found by the fast (vector) code DP implementation.
* - accscore_g is the accuracy found by the generic DP implementation.
* accscore and accscore_g should be nearly identical,
* within tolerance of roundoff error accumulation and
* the imprecision of Logsum() tables.
* - accscore_g2 is the accuracy of the traceback identified by the generic
* DP implementation. It should be identical (within order-of-evaluation
* roundoff error) to accscore_g.
*
* the "accscore_g2 < accscore_o" test is carefully contrived.
* accscore_o is a theoretical lower bound but because of fp error,
* accscore and (much more rarely) even accscore_g can exceed accscore_o.
* accscore_g2, however, is calculated with identical order of evaluation
* as accscore_o if the optimal trace does turn out to be identical to
* the originally emitted trace. It should be extremely unlikely (though
* not impossible) for accscore_o to exceed accscore_g2. (The DP algorithm
* would have to identify a trace that was different than the original trace,
* which the DP algorithm, by order-of-evaluation, assigned higher accuracy,
* but order-of-evaluation in traceback dependent code assigned lower accuracy.
* [xref J5/29]
*/
esl_sq_Reuse(sq);
p7_trace_Reuse(tr);
p7_trace_Reuse(trg);
p7_trace_Reuse(tro);
}
p7_trace_Destroy(tro);
p7_trace_Destroy(trg);
p7_trace_Destroy(tr);
p7_gmx_Destroy(gx2);
p7_gmx_Destroy(gx1);
p7_omx_Destroy(ox2);
p7_omx_Destroy(ox1);
esl_sq_Destroy(sq);
p7_oprofile_Destroy(om);
p7_profile_Destroy(gm);
p7_hmm_Destroy(hmm);
}
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;
}
/* "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);
}
