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;
}
/* 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;
}
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;
}
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;
}
/* The "enumeration" test samples a random enumerable HMM (transitions to insert are 0,
* so the generated seq space only includes seqs of L<=M).
*
* The test scores all seqs of length <=M by both Viterbi and Forward, verifies that
* the two scores are identical, and verifies that the sum of all the probabilities is
* 1.0. It also verifies that the score of a sequence of length M+1 is indeed -infinity.
*
* Because this function is going to work in unscaled probabilities, adding them up,
* all P(seq) terms must be >> DBL_EPSILON. That means M must be small; on the order
* of <= 10.
*/
static void
utest_enumeration(ESL_GETOPTS *go, ESL_RANDOMNESS *r, ESL_ALPHABET *abc, int M)
{
char errbuf[eslERRBUFSIZE];
P7_HMM *hmm = NULL;
P7_PROFILE *gm = NULL;
P7_BG *bg = NULL;
ESL_DSQ *dsq = NULL;
P7_GMX *gx = NULL;
float vsc, fsc;
float bg_ll; /* log P(seq | bg) */
double vp, fp; /* P(seq,\pi | model) and P(seq | model) */
int L;
int i;
double total_p;
char *seq;
/* Sample an enumerable HMM & profile of length M.
*/
if (p7_hmm_SampleEnumerable(r, M, abc, &hmm) != eslOK) esl_fatal("failed to sample an enumerable 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, 0, p7_UNILOCAL) != 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);
if ( (dsq = malloc(sizeof(ESL_DSQ) * (M+3))) == NULL) esl_fatal("allocation failed");
if ( (seq = malloc(sizeof(char) * (M+2))) == NULL) esl_fatal("allocation failed");
if ((gx = p7_gmx_Create(hmm->M, M+3)) == NULL) esl_fatal("matrix creation failed");
/* Enumerate all sequences of length L <= M
*/
total_p = 0;
for (L = 0; L <= M; L++)
{
/* Initialize dsq of length L at 0000... */
dsq[0] = dsq[L+1] = eslDSQ_SENTINEL;
for (i = 1; i <= L; i++) dsq[i] = 0;
while (1) /* enumeration of seqs of length L*/
{
if (p7_GViterbi(dsq, L, gm, gx, &vsc) != eslOK) esl_fatal("viterbi failed");
if (p7_GForward(dsq, L, gm, gx, &fsc) != eslOK) esl_fatal("forward failed");
/* calculate bg log likelihood component of the scores */
for (bg_ll = 0., i = 1; i <= L; i++) bg_ll += log(bg->f[dsq[i]]);
/* convert to probabilities, adding the bg LL back to the LLR */
vp = exp(vsc + bg_ll);
fp = exp(fsc + bg_ll);
if (esl_opt_GetBoolean(go, "--vv")) {
esl_abc_Textize(abc, dsq, L, seq);
printf("probability of sequence: %10s %16g (lod v=%8.4f f=%8.4f)\n", seq, fp, vsc, fsc);
}
total_p += fp;
/* Increment dsq like a reversed odometer */
for (i = 1; i <= L; i++)
if (dsq[i] < abc->K-1) { dsq[i]++; break; } else { dsq[i] = 0; }
if (i > L) break; /* we're done enumerating sequences */
}
}
/* That sum is subject to significant numerical error because of
* discretization error in FLogsum(); don't expect it to be too close.
*/
if (total_p < 0.999 || total_p > 1.001) esl_fatal("Enumeration unit test failed: total Forward p isn't near 1.0 (%g)", total_p);
if (esl_opt_GetBoolean(go, "-v")) {
printf("enumeration test: total p is %g\n", total_p);
}
p7_gmx_Destroy(gx);
p7_bg_Destroy(bg);
p7_profile_Destroy(gm);
p7_hmm_Destroy(hmm);
free(dsq);
free(seq);
}