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;
}
/* Function: p7_ViterbiMu()
* Synopsis: Determines the local Viterbi Gumbel mu parameter for a model.
* Incept: SRE, Tue May 19 10:26:19 2009 [Janelia]
*
* Purpose: Identical to p7_MSVMu(), above, except that it fits
* Viterbi scores instead of MSV scores.
*
* The difference between the two mus is small, but can be
* up to ~1 bit or so for large, low-info models [J4/126] so
* decided to calibrate the two mus separately [J5/8].
*
* Args: r : source of random numbers
* om : score profile (length config is changed upon return!)
* bg : null model (length config is changed upon return!)
* L : length of sequences to simulate
* N : number of sequences to simulate
* lambda : known Gumbel lambda parameter
* ret_vmu : RETURN: ML estimate of location param mu
*
* Returns: <eslOK> on success, and <ret_mu> contains the ML estimate
* of $\mu$.
*
* Throws: (no abnormal error conditions)
*/
int
p7_ViterbiMu(ESL_RANDOMNESS *r, P7_OPROFILE *om, P7_BG *bg, int L, int N, double lambda, double *ret_vmu)
{
P7_OMX *ox = p7_omx_Create(om->M, 0, 0); /* DP matrix: 1 row version */
ESL_DSQ *dsq = NULL;
double *xv = NULL;
int i;
float sc, nullsc;
#ifndef p7_IMPL_DUMMY
float maxsc = (32767.0 - om->base_w) / om->scale_w; /* if score overflows, use this [J4/139] */
#endif
int status;
if (ox == NULL) { status = eslEMEM; goto ERROR; }
ESL_ALLOC(xv, sizeof(double) * N);
ESL_ALLOC(dsq, sizeof(ESL_DSQ) * (L+2));
p7_oprofile_ReconfigLength(om, L);
p7_bg_SetLength(bg, L);
for (i = 0; i < N; i++)
{
if ((status = esl_rsq_xfIID(r, bg->f, om->abc->K, L, dsq)) != eslOK) goto ERROR;
if ((status = p7_bg_NullOne(bg, dsq, L, &nullsc)) != eslOK) goto ERROR;
status = p7_ViterbiFilter(dsq, L, om, ox, &sc);
#ifndef p7_IMPL_DUMMY
if (status == eslERANGE) { sc = maxsc; status = eslOK; }
#endif
if (status != eslOK) goto ERROR;
xv[i] = (sc - nullsc) / eslCONST_LOG2;
}
if ((status = esl_gumbel_FitCompleteLoc(xv, N, lambda, ret_vmu)) != eslOK) goto ERROR;
p7_omx_Destroy(ox);
free(xv);
free(dsq);
return eslOK;
ERROR:
*ret_vmu = 0.0;
if (ox != NULL) p7_omx_Destroy(ox);
if (xv != NULL) free(xv);
if (dsq != NULL) free(dsq);
return status;
}
/* Function: p7_Tau()
* Synopsis: Determine Forward tau by brief simulation.
* Incept: SRE, Thu Aug 9 15:08:39 2007 [Janelia]
*
* Purpose: Determine the <tau> parameter for an exponential tail fit
* to the Forward score distribution for model <om>, on
* random sequences with the composition of the background
* model <bg>. This <tau> parameter is for an exponential
* distribution anchored from $P=1.0$, so it's not really a
* tail per se; but it's only an accurate fit in the tail
* of the Forward score distribution, from about $P=0.001$
* or so.
*
* The determination of <tau> is done by a brief simulation
* in which we fit a Gumbel distribution to a small number
* of Forward scores of random sequences, and use that to
* predict the location of the tail at probability <tailp>.
*
* The Gumbel is of course inaccurate, but we can use it
* here solely as an empirical distribution to determine
* the location of a reasonable <tau> more accurately on a
* smaller number of samples than we could do with raw
* order statistics.
*
* Typical choices are L=100, N=200, tailp=0.04, which
* typically yield estimates $\hat{\mu}$ with a precision
* (standard deviation) of $\pm$ 0.2 bits, corresponding to
* a $\pm$ 15\% error in E-values. See [J1/135].
*
* The use of Gumbel fitting to a small number of $N$
* samples and the extrapolation of $\hat{\mu}$ from the
* estimated location of the 0.04 tail mass are both
* empirical and carefully optimized against several
* tradeoffs. Most importantly, around this choice of tail
* probability, a systematic error introduced by the use of
* the Gumbel fit is being cancelled by systematic error
* introduced by the use of a higher tail probability than
* the regime in which the exponential tail is a valid
* approximation. See [J1/135] for discussion.
*
* This function changes the length configuration of both
* <om> and <bg>. The caller must remember to reconfigure
* both of their length models appropriately for any
* subsequent alignments.
*
* Args: r : source of randomness
* om : configured profile to sample sequences from
* bg : null model (for background residue frequencies)
* L : mean length model for seq emission from profile
* N : number of sequences to generate
* lambda : expected slope of the exponential tail (from p7_Lambda())
* tailp : tail mass from which we will extrapolate mu
* ret_mu : RETURN: estimate for the Forward mu (base of exponential tail)
*
* Returns: <eslOK> on success, and <*ret_fv> is the score difference
* in bits.
*
* Throws: <eslEMEM> on allocation error, and <*ret_fv> is 0.
*/
int
p7_Tau(ESL_RANDOMNESS *r, P7_OPROFILE *om, P7_BG *bg, int L, int N, double lambda, double tailp, double *ret_tau)
{
P7_OMX *ox = p7_omx_Create(om->M, 0, L); /* DP matrix: for ForwardParser, L rows */
ESL_DSQ *dsq = NULL;
double *xv = NULL;
float fsc, nullsc;
double gmu, glam;
int status;
int i;
ESL_ALLOC(xv, sizeof(double) * N);
ESL_ALLOC(dsq, sizeof(ESL_DSQ) * (L+2));
if (ox == NULL) { status = eslEMEM; goto ERROR; }
p7_oprofile_ReconfigLength(om, L);
p7_bg_SetLength(bg, L);
for (i = 0; i < N; i++)
{
if ((status = esl_rsq_xfIID(r, bg->f, om->abc->K, L, dsq)) != eslOK) goto ERROR;
if ((status = p7_ForwardParser(dsq, L, om, ox, &fsc)) != eslOK) goto ERROR;
if ((status = p7_bg_NullOne(bg, dsq, L, &nullsc)) != eslOK) goto ERROR;
xv[i] = (fsc - nullsc) / eslCONST_LOG2;
}
if ((status = esl_gumbel_FitComplete(xv, N, &gmu, &glam)) != eslOK) goto ERROR;
/* Explanation of the eqn below: first find the x at which the Gumbel tail
* mass is predicted to be equal to tailp. Then back up from that x
* by log(tailp)/lambda to set the origin of the exponential tail to 1.0
* instead of tailp.
*/
*ret_tau = esl_gumbel_invcdf(1.0-tailp, gmu, glam) + (log(tailp) / lambda);
free(xv);
free(dsq);
p7_omx_Destroy(ox);
return eslOK;
ERROR:
*ret_tau = 0.;
if (xv != NULL) free(xv);
if (dsq != NULL) free(dsq);
if (ox != NULL) p7_omx_Destroy(ox);
return status;
}
/* Forward is hard to validate.
* We do know that the Forward score is >= Viterbi.
* We also know that the expected score on random seqs is <= 0 (not
* exactly - we'd have to sample the random length from the background
* model too, not just use a fixed L - but it's close enough to
* being true to be a useful test.)
*/
static void
utest_forward(ESL_GETOPTS *go, ESL_RANDOMNESS *r, ESL_ALPHABET *abc, P7_BG *bg, P7_PROFILE *gm, int nseq, int L)
{
float avg_sc;
ESL_DSQ *dsq = NULL;
P7_GMX *fwd = NULL;
P7_GMX *bck = NULL;
int idx;
float fsc, bsc;
float vsc, nullsc;
if ((dsq = malloc(sizeof(ESL_DSQ) *(L+2))) == NULL) esl_fatal("malloc failed");
if ((fwd = p7_gmx_Create(gm->M, L)) == NULL) esl_fatal("matrix creation failed");
if ((bck = p7_gmx_Create(gm->M, L)) == NULL) esl_fatal("matrix creation failed");
avg_sc = 0.;
for (idx = 0; idx < nseq; idx++)
{
if (esl_rsq_xfIID(r, bg->f, abc->K, L, dsq) != eslOK) esl_fatal("seq generation failed");
if (p7_GViterbi(dsq, L, gm, fwd, &vsc) != eslOK) esl_fatal("viterbi failed");
if (p7_GForward(dsq, L, gm, fwd, &fsc) != eslOK) esl_fatal("forward failed");
if (p7_GBackward(dsq, L, gm, bck, &bsc) != eslOK) esl_fatal("backward failed");
if (fsc < vsc) esl_fatal("Foward score can't be less than Viterbi score");
if (fabs(fsc-bsc) > 0.001) esl_fatal("Forward/Backward failed: %f %f\n", fsc, bsc);
if (p7_bg_NullOne(bg, dsq, L, &nullsc) != eslOK) esl_fatal("null score failed");
avg_sc += fsc - nullsc;
if (esl_opt_GetBoolean(go, "--vv"))
printf("utest_forward: Forward score: %.4f (total so far: %.4f)\n", fsc, avg_sc);
}
avg_sc /= (float) nseq;
if (avg_sc > 0.) esl_fatal("Forward scores have positive expectation (%f nats)", avg_sc);
p7_gmx_Destroy(fwd);
p7_gmx_Destroy(bck);
free(dsq);
return;
}
/* Viterbi validation is done by comparing the returned score
* to the score of the optimal trace. Not foolproof, but catches
* many kinds of errors.
*
* Another check is that the average score should be <= 0,
* since the random sequences are drawn from the null model.
*/
static void
utest_viterbi(ESL_GETOPTS *go, ESL_RANDOMNESS *r, ESL_ALPHABET *abc, P7_BG *bg, P7_PROFILE *gm, int nseq, int L)
{
float avg_sc = 0.;
char errbuf[eslERRBUFSIZE];
ESL_DSQ *dsq = NULL;
P7_GMX *gx = NULL;
P7_TRACE *tr = NULL;
int idx;
float sc1, sc2;
if ((dsq = malloc(sizeof(ESL_DSQ) *(L+2))) == NULL) esl_fatal("malloc failed");
if ((tr = p7_trace_Create()) == NULL) esl_fatal("trace creation failed");
if ((gx = p7_gmx_Create(gm->M, L)) == NULL) esl_fatal("matrix creation failed");
for (idx = 0; idx < nseq; idx++)
{
if (esl_rsq_xfIID(r, bg->f, abc->K, L, dsq) != eslOK) esl_fatal("seq generation failed");
if (p7_GViterbi(dsq, L, gm, gx, &sc1) != eslOK) esl_fatal("viterbi failed");
if (p7_GTrace (dsq, L, gm, gx, tr) != eslOK) esl_fatal("trace failed");
if (p7_trace_Validate(tr, abc, dsq, errbuf) != eslOK) esl_fatal("trace invalid:\n%s", errbuf);
if (p7_trace_Score(tr, dsq, gm, &sc2) != eslOK) esl_fatal("trace score failed");
if (esl_FCompare(sc1, sc2, 1e-6) != eslOK) esl_fatal("Trace score != Viterbi score");
if (p7_bg_NullOne(bg, dsq, L, &sc2) != eslOK) esl_fatal("null score failed");
avg_sc += (sc1 - sc2);
if (esl_opt_GetBoolean(go, "--vv"))
printf("utest_viterbi: Viterbi score: %.4f (null %.4f) (total so far: %.4f)\n", sc1, sc2, avg_sc);
p7_trace_Reuse(tr);
}
avg_sc /= (float) nseq;
if (avg_sc > 0.) esl_fatal("Viterbi scores have positive expectation (%f nats)", avg_sc);
p7_gmx_Destroy(gx);
p7_trace_Destroy(tr);
free(dsq);
return;
}
/* The "generation" test scores sequences generated by the same profile.
* Each Viterbi and Forward score should be >= the trace score of the emitted seq.
* The expectation of Forward scores should be positive.
*/
static void
utest_generation(ESL_GETOPTS *go, ESL_RANDOMNESS *r, ESL_ALPHABET *abc,
P7_PROFILE *gm, P7_HMM *hmm, P7_BG *bg, int nseq)
{
ESL_SQ *sq = esl_sq_CreateDigital(abc);
P7_GMX *gx = p7_gmx_Create(gm->M, 100);
P7_TRACE *tr = p7_trace_Create();
float vsc, fsc, nullsc, tracesc;
float avg_fsc;
int idx;
avg_fsc = 0.0;
for (idx = 0; idx < nseq; idx++)
{
if (p7_ProfileEmit(r, hmm, gm, bg, sq, tr) != eslOK) esl_fatal("profile emission failed");
if (p7_gmx_GrowTo(gx, gm->M, sq->n) != eslOK) esl_fatal("failed to reallocate gmx");
if (p7_GViterbi(sq->dsq, sq->n, gm, gx, &vsc) != eslOK) esl_fatal("viterbi failed");
if (p7_GForward(sq->dsq, sq->n, gm, gx, &fsc) != eslOK) esl_fatal("forward failed");
if (p7_trace_Score(tr, sq->dsq, gm, &tracesc) != eslOK) esl_fatal("trace score failed");
if (p7_bg_NullOne(bg, sq->dsq, sq->n, &nullsc) != eslOK) esl_fatal("null score failed");
if (vsc < tracesc) esl_fatal("viterbi score is less than trace");
if (fsc < tracesc) esl_fatal("forward score is less than trace");
if (vsc > fsc) esl_fatal("viterbi score is greater than forward");
if (esl_opt_GetBoolean(go, "--vv"))
printf("generated: len=%d v=%8.4f f=%8.4f t=%8.4f\n", (int) sq->n, vsc, fsc, tracesc);
avg_fsc += (fsc - nullsc);
}
avg_fsc /= (float) nseq;
if (avg_fsc < 0.) esl_fatal("generation: Forward scores have negative expectation (%f nats)", avg_fsc);
p7_gmx_Destroy(gx);
p7_trace_Destroy(tr);
esl_sq_Destroy(sq);
}
int
main(int argc, char **argv)
{
ESL_GETOPTS *go = esl_getopts_CreateDefaultApp(options, 2, argc, argv, banner, usage);
char *hmmfile = esl_opt_GetArg(go, 1);
char *seqfile = esl_opt_GetArg(go, 2);
ESL_ALPHABET *abc = NULL;
P7_HMMFILE *hfp = NULL;
P7_HMM *hmm = NULL;
P7_BG *bg = NULL;
P7_PROFILE *gm = NULL;
P7_OPROFILE *om = NULL;
P7_OMX *ox = NULL;
P7_GMX *gx = NULL;
ESL_SQ *sq = NULL;
ESL_SQFILE *sqfp = NULL;
int format = eslSQFILE_UNKNOWN;
float vfraw, nullsc, vfscore;
float graw, gscore;
double P, gP;
int status;
/* Read in one HMM */
if (p7_hmmfile_Open(hmmfile, NULL, &hfp) != eslOK) p7_Fail("Failed to open HMM file %s", hmmfile);
if (p7_hmmfile_Read(hfp, &abc, &hmm) != eslOK) p7_Fail("Failed to read HMM");
/* Read in one sequence */
sq = esl_sq_CreateDigital(abc);
status = esl_sqfile_Open(seqfile, format, NULL, &sqfp);
if (status == eslENOTFOUND) p7_Fail("No such file.");
else if (status == eslEFORMAT) p7_Fail("Format unrecognized.");
else if (status == eslEINVAL) p7_Fail("Can't autodetect stdin or .gz.");
else if (status != eslOK) p7_Fail("Open failed, code %d.", status);
/* create default null model, then create and optimize profile */
bg = p7_bg_Create(abc);
p7_bg_SetLength(bg, sq->n);
gm = p7_profile_Create(hmm->M, abc);
p7_ProfileConfig(hmm, bg, gm, sq->n, p7_LOCAL);
om = p7_oprofile_Create(gm->M, abc);
p7_oprofile_Convert(gm, om);
/* allocate DP matrices, both a generic and an optimized one */
ox = p7_omx_Create(gm->M, 0, sq->n);
gx = p7_gmx_Create(gm->M, sq->n);
/* Useful to place and compile in for debugging:
p7_oprofile_Dump(stdout, om); dumps the optimized profile
p7_omx_SetDumpMode(ox, TRUE); makes the fast DP algorithms dump their matrices
p7_gmx_Dump(stdout, gx); dumps a generic DP matrix
*/
while ((status = esl_sqio_Read(sqfp, sq)) == eslOK)
{
p7_oprofile_ReconfigLength(om, sq->n);
p7_ReconfigLength(gm, sq->n);
p7_bg_SetLength(bg, sq->n);
p7_omx_GrowTo(ox, om->M, 0, sq->n);
p7_gmx_GrowTo(gx, gm->M, sq->n);
p7_ViterbiFilter (sq->dsq, sq->n, om, ox, &vfraw);
p7_bg_NullOne (bg, sq->dsq, sq->n, &nullsc);
vfscore = (vfraw - nullsc) / eslCONST_LOG2;
P = esl_gumbel_surv(vfscore, om->evparam[p7_VMU], om->evparam[p7_VLAMBDA]);
p7_GViterbi (sq->dsq, sq->n, gm, gx, &graw);
gscore = (graw - nullsc) / eslCONST_LOG2;
gP = esl_gumbel_surv(gscore, gm->evparam[p7_VMU], gm->evparam[p7_VLAMBDA]);
if (esl_opt_GetBoolean(go, "-1"))
{
printf("%-30s\t%-20s\t%9.2g\t%7.2f\t%9.2g\t%7.2f\n", sq->name, hmm->name, P, vfscore, gP, gscore);
}
else if (esl_opt_GetBoolean(go, "-P"))
{ /* output suitable for direct use in profmark benchmark postprocessors: */
printf("%g\t%.2f\t%s\t%s\n", P, vfscore, sq->name, hmm->name);
}
else
{
printf("target sequence: %s\n", sq->name);
printf("vit filter raw score: %.2f nats\n", vfraw);
printf("null score: %.2f nats\n", nullsc);
printf("per-seq score: %.2f bits\n", vfscore);
printf("P-value: %g\n", P);
printf("GViterbi raw score: %.2f nats\n", graw);
printf("GViterbi seq score: %.2f bits\n", gscore);
printf("GViterbi P-value: %g\n", gP);
}
esl_sq_Reuse(sq);
}
/* cleanup */
esl_sq_Destroy(sq);
esl_sqfile_Close(sqfp);
p7_omx_Destroy(ox);
p7_gmx_Destroy(gx);
p7_oprofile_Destroy(om);
p7_profile_Destroy(gm);
p7_bg_Destroy(bg);
p7_hmm_Destroy(hmm);
p7_hmmfile_Close(hfp);
esl_alphabet_Destroy(abc);
esl_getopts_Destroy(go);
return 0;
}
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_OPROFILE *om = NULL;
P7_GMX *gx = NULL;
P7_OMX *fwd = NULL;
P7_OMX *bck = NULL;
ESL_SQ *sq = NULL;
ESL_SQFILE *sqfp = NULL;
int format = eslSQFILE_UNKNOWN;
float fraw, braw, nullsc, fsc;
float gfraw, gbraw, gfsc;
double P, gP;
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");
/* Open sequence file for reading */
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);
/* create default null model, then create and optimize profile */
bg = p7_bg_Create(abc);
p7_bg_SetLength(bg, sq->n);
gm = p7_profile_Create(hmm->M, abc);
p7_ProfileConfig(hmm, bg, gm, sq->n, p7_UNILOCAL);
om = p7_oprofile_Create(gm->M, abc);
p7_oprofile_Convert(gm, om);
/* p7_oprofile_Dump(stdout, om); */
/* allocate DP matrices for O(M+L) parsers */
fwd = p7_omx_Create(gm->M, 0, sq->n);
bck = p7_omx_Create(gm->M, 0, sq->n);
gx = p7_gmx_Create(gm->M, sq->n);
/* allocate DP matrices for O(ML) fills */
/* fwd = p7_omx_Create(gm->M, sq->n, sq->n); */
/* bck = p7_omx_Create(gm->M, sq->n, sq->n); */
/* p7_omx_SetDumpMode(stdout, fwd, TRUE); */ /* makes the fast DP algorithms dump their matrices */
/* p7_omx_SetDumpMode(stdout, bck, TRUE); */
while ((status = esl_sqio_Read(sqfp, sq)) == eslOK)
{
p7_oprofile_ReconfigLength(om, sq->n);
p7_ReconfigLength(gm, sq->n);
p7_bg_SetLength(bg, sq->n);
p7_omx_GrowTo(fwd, om->M, 0, sq->n);
p7_omx_GrowTo(bck, om->M, 0, sq->n);
p7_gmx_GrowTo(gx, gm->M, sq->n);
p7_bg_NullOne (bg, sq->dsq, sq->n, &nullsc);
p7_ForwardParser (sq->dsq, sq->n, om, fwd, &fraw);
p7_BackwardParser(sq->dsq, sq->n, om, fwd, bck, &braw);
/* p7_Forward (sq->dsq, sq->n, om, fwd, &fsc); printf("forward: %.2f nats\n", fsc); */
/* p7_Backward(sq->dsq, sq->n, om, fwd, bck, &bsc); printf("backward: %.2f nats\n", bsc); */
/* Comparison to other F/B implementations */
p7_GForward (sq->dsq, sq->n, gm, gx, &gfraw);
p7_GBackward (sq->dsq, sq->n, gm, gx, &gbraw);
/* p7_gmx_Dump(stdout, gx); */
fsc = (fraw-nullsc) / eslCONST_LOG2;
gfsc = (gfraw-nullsc) / eslCONST_LOG2;
P = esl_exp_surv(fsc, om->evparam[p7_FTAU], om->evparam[p7_FLAMBDA]);
gP = esl_exp_surv(gfsc, gm->evparam[p7_FTAU], gm->evparam[p7_FLAMBDA]);
if (esl_opt_GetBoolean(go, "-1"))
{
printf("%-30s\t%-20s\t%9.2g\t%6.1f\t%9.2g\t%6.1f\n", sq->name, hmm->name, P, fsc, gP, gfsc);
}
else if (esl_opt_GetBoolean(go, "-P"))
{ /* output suitable for direct use in profmark benchmark postprocessors: */
printf("%g\t%.2f\t%s\t%s\n", P, fsc, sq->name, hmm->name);
}
else
{
printf("target sequence: %s\n", sq->name);
printf("fwd filter raw score: %.2f nats\n", fraw);
printf("bck filter raw score: %.2f nats\n", braw);
printf("null score: %.2f nats\n", nullsc);
printf("per-seq score: %.2f bits\n", fsc);
printf("P-value: %g\n", P);
printf("GForward raw score: %.2f nats\n", gfraw);
printf("GBackward raw score: %.2f nats\n", gbraw);
printf("GForward seq score: %.2f bits\n", gfsc);
printf("GForward P-value: %g\n", gP);
}
esl_sq_Reuse(sq);
}
/* cleanup */
esl_sq_Destroy(sq);
esl_sqfile_Close(sqfp);
p7_omx_Destroy(bck);
p7_omx_Destroy(fwd);
p7_gmx_Destroy(gx);
p7_oprofile_Destroy(om);
p7_profile_Destroy(gm);
p7_bg_Destroy(bg);
p7_hmm_Destroy(hmm);
p7_hmmfile_Close(hfp);
esl_alphabet_Destroy(abc);
esl_getopts_Destroy(go);
return 0;
}
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;
ESL_SQFILE *sqfp = NULL;
int format = eslSQFILE_UNKNOWN;
ESL_SQ *sq = NULL;
float nullsc, filtersc, H;
int status;
/* Read one HMM from <hmmfile> */
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);
/* Open <seqfile> for reading */
status = esl_sqfile_OpenDigital(abc, seqfile, format, NULL, &sqfp);
if (status == eslENOTFOUND) esl_fatal("No such file.");
else if (status == eslEFORMAT) esl_fatal("Format unrecognized.");
else if (status != eslOK) esl_fatal("Open failed, code %d.", status);
sq = esl_sq_CreateDigital(abc);
bg = p7_bg_Create(abc);
p7_bg_SetFilterByHMM(bg, hmm);
H = esl_vec_FEntropy(bg->f, bg->abc->K);
printf("bg iid H = %.4f\n", H);
H = esl_vec_FEntropy(bg->mcomp, bg->abc->K);
printf("modelcomp H = %.4f\n", H);
while ((status = esl_sqio_Read(sqfp, sq)) == eslOK)
{
p7_bg_SetLength(bg, sq->n);
p7_bg_NullOne (bg, sq->dsq, sq->n, &nullsc);
p7_bg_FilterScore(bg, sq->dsq, sq->n, &filtersc);
printf("%-20s %5d %8.5f %8.5f %8.5f\n", sq->name, (int) sq->n, nullsc, filtersc, filtersc-nullsc);
esl_sq_Reuse(sq);
}
if (status == eslEFORMAT) esl_fatal("Parse failed (sequence file %s line %" PRId64 "):\n%s\n",
sqfp->filename, sqfp->linenumber, sqfp->errbuf);
else if (status != eslEOF) esl_fatal("Unexpected error %d reading sequence file %s",
status, sqfp->filename);
esl_sqfile_Close(sqfp);
esl_sq_Destroy(sq);
p7_bg_Destroy(bg);
p7_hmm_Destroy(hmm);
esl_alphabet_Destroy(abc);
esl_getopts_Destroy(go);
return 0;
}
/* process_workunit()
*
* This is the routine that actually does the work.
*
* A work unit consists of one HMM, <hmm>.
* The result is the <scores> array, which contains an array of N scores;
* caller provides this memory.
* How those scores are generated is controlled by the application configuration in <cfg>.
*/
static int
process_workunit(ESL_GETOPTS *go, struct cfg_s *cfg, char *errbuf, P7_HMM *hmm, double *scores, int *alilens)
{
int L = esl_opt_GetInteger(go, "-L");
P7_PROFILE *gm = NULL;
P7_OPROFILE *om = NULL;
P7_REFMX *rmx = NULL;
P7_CHECKPTMX *cx = NULL;
P7_FILTERMX *fx = NULL;
P7_TRACE *tr = NULL;
ESL_DSQ *dsq = NULL;
int i;
int scounts[p7T_NSTATETYPES]; /* state usage counts from a trace */
float sc;
float nullsc;
int status;
P7_HARDWARE *hw;
if ((hw = p7_hardware_Create ()) == NULL) p7_Fail("Couldn't get HW information data structure");
/* Optionally set a custom background, determined by model composition;
* an experimental hack.
*/
if (esl_opt_GetBoolean(go, "--bgcomp"))
{
float *p = NULL;
float KL;
p7_hmm_CompositionKLDist(hmm, cfg->bg, &KL, &p);
esl_vec_FCopy(p, cfg->abc->K, cfg->bg->f);
}
/* Create and configure our generic profile, as requested */
gm = p7_profile_Create(hmm->M, cfg->abc);
if (esl_opt_GetBoolean(go, "--multi"))
{
if (esl_opt_GetBoolean(go, "--dual")) { p7_profile_Config (gm, hmm, cfg->bg); }
else if (esl_opt_GetBoolean(go, "--local")) { p7_profile_ConfigLocal (gm, hmm, cfg->bg, L); }
else if (esl_opt_GetBoolean(go, "--glocal")) { p7_profile_ConfigGlocal(gm, hmm, cfg->bg, L); }
}
else if (esl_opt_GetBoolean(go, "--uni"))
{
if (esl_opt_GetBoolean(go, "--dual")) { p7_profile_ConfigCustom (gm, hmm, cfg->bg, L, 0.0, 0.5); }
else if (esl_opt_GetBoolean(go, "--local")) { p7_profile_ConfigUnilocal (gm, hmm, cfg->bg, L); }
else if (esl_opt_GetBoolean(go, "--glocal")) { p7_profile_ConfigUniglocal(gm, hmm, cfg->bg, L); }
}
p7_profile_SetLength(gm, L);
p7_bg_SetLength(cfg->bg, L);
if (esl_opt_GetBoolean(go, "--x-no-lengthmodel")) elide_length_model(gm, cfg->bg);
/* Allocate DP matrix for <gm>.
*/
rmx = p7_refmx_Create(gm->M, L);
/* Create and configure the vectorized profile, if needed;
* and allocate its DP matrix
*/
if (esl_opt_GetBoolean(go, "--vector"))
{
om = p7_oprofile_Create(gm->M, cfg->abc, om->simd);
p7_oprofile_Convert(gm, om);
cx = p7_checkptmx_Create(gm->M, L, ESL_MBYTES(32), om->simd);
fx = p7_filtermx_Create(gm->M, om->simd);
}
/* Remaining allocation */
ESL_ALLOC(dsq, sizeof(ESL_DSQ) * (L+2));
tr = p7_trace_Create();
/* Collect scores from N random sequences of length L */
for (i = 0; i < cfg->N; i++)
{
esl_rsq_xfIID(cfg->r, cfg->bg->f, cfg->abc->K, L, dsq);
sc = eslINFINITY;
/* Vectorized implementations of Viterbi, MSV may overflow.
* In this case, they'll leave sc=eslINFINITY.
* Then we fail over to the nonvector "generic" implementation.
* That's why this next block isn't an if/else.
*/
if (esl_opt_GetBoolean(go, "--vector"))
{
if (esl_opt_GetBoolean(go, "--vit")) p7_ViterbiFilter(dsq, L, om, fx, &sc);
else if (esl_opt_GetBoolean(go, "--fwd")) p7_ForwardFilter(dsq, L, om, cx, &sc);
else if (esl_opt_GetBoolean(go, "--msv")) p7_MSVFilter (dsq, L, om, fx, &sc);
}
/* If we tried a vector calculation above but it overflowed,
* or if we're to do --generic DP calculations, sc==eslINFINITY now;
* hence the if condition here:
*/
if (sc == eslINFINITY)
{
if (esl_opt_GetBoolean(go, "--fwd")) p7_ReferenceForward(dsq, L, gm, rmx, &sc); /* any mode: dual,local,glocal; gm's config takes care of this */
else if (esl_opt_GetBoolean(go, "--vit")) p7_ReferenceViterbi(dsq, L, gm, rmx, tr, &sc); /* local-only mode. cmdline opts processing has already assured that --local set */
else if (esl_opt_GetBoolean(go, "--msv")) p7_Die("We used to be able to do a generic MSV algorithm - but no longer");
}
/* Optional: get Viterbi alignment length too. */
if (esl_opt_GetBoolean(go, "-a")) /* -a only works with Viterbi; getopts has checked this already; <tr> must be valid */
{
p7_trace_GetStateUseCounts(tr, scounts);
/* there's various ways we could counts "alignment length".
* Here we'll use the total length of model used, in nodes: M+D states.
* score vs al would gives us relative entropy / model position.
*/
/* alilens[i] = scounts[p7T_D] + scounts[p7T_I]; SRE: temporarily testing this instead */
alilens[i] = scounts[p7T_ML] + scounts[p7T_DL] + scounts[p7T_IL] +
scounts[p7T_MG] + scounts[p7T_DG] + scounts[p7T_IG];
p7_trace_Reuse(tr);
}
p7_bg_NullOne(cfg->bg, dsq, L, &nullsc);
scores[i] = (sc - nullsc) / eslCONST_LOG2;
if (cx) p7_checkptmx_Reuse(cx);
if (fx) p7_filtermx_Reuse(fx);
p7_refmx_Reuse(rmx);
}
status = eslOK;
/* deliberate flowthru */
ERROR:
if (dsq != NULL) free(dsq);
p7_checkptmx_Destroy(cx);
p7_filtermx_Destroy(fx);
p7_oprofile_Destroy(om);
p7_profile_Destroy(gm);
p7_refmx_Destroy(rmx);
p7_trace_Destroy(tr);
if (status == eslEMEM) sprintf(errbuf, "allocation failure");
return status;
}
void run_hmmer_pipeline(const char* seq) {
int index, i, status;
ESL_SQ* sq = esl_sq_CreateFrom(NULL, seq, NULL, NULL, NULL);
P7_OPROFILE *om = NULL;
P7_PROFILE *gm = NULL;
float usc, vfsc, fwdsc; /* filter scores */
float filtersc; /* HMM null filter score */
float nullsc; /* null model score */
float seqbias;
float seq_score; /* the corrected per-seq bit score */
double P;
WRAPPER_RESULT* result;
num_results = 0;
if(sq->n == 0) {
esl_sq_Destroy(sq);
return;
}
esl_sq_Digitize(abc, sq);
int n = 0;
float oasc;
for(index = 0;index < num_models;index++) {
om = models[index];
p7_omx_Reuse(oxf);
p7_omx_Reuse(oxb);
p7_omx_GrowTo(oxf, om->M, sq->n, sq->n);
p7_omx_GrowTo(oxb, om->M, sq->n, sq->n);
p7_oprofile_ReconfigLength(om, sq->n);
p7_bg_SetFilter(bg, om->M, om->compo);
p7_bg_SetLength(bg, sq->n);
//Calibrate null model
p7_bg_NullOne(bg, sq->dsq, sq->n, &nullsc);
//MSV Filter
p7_MSVFilter(sq->dsq, sq->n, om, oxf, &usc);
seq_score = (usc - nullsc) / eslCONST_LOG2;
P = esl_gumbel_surv(seq_score, om->evparam[p7_MMU], om->evparam[p7_MLAMBDA]);
if (P > f1) continue;
//Bias filter (model compo)
p7_bg_FilterScore(bg, sq->dsq, sq->n, &filtersc);
seq_score = (usc - filtersc) / eslCONST_LOG2;
P = esl_gumbel_surv(seq_score, om->evparam[p7_MMU], om->evparam[p7_MLAMBDA]);
if (P > f1) continue;
//Viterbi filter (Only do if P value from Bias is high)
if(P > f2) {
p7_ViterbiFilter(sq->dsq, sq->n, om, oxf, &vfsc);
seq_score = (vfsc - filtersc) / eslCONST_LOG2;
P = esl_gumbel_surv(seq_score, om->evparam[p7_VMU], om->evparam[p7_VLAMBDA]);
if (P > f2) continue;
}
//Get the real probability (forward)
p7_Forward(sq->dsq, sq->n, om, oxf, &fwdsc);
seq_score = (fwdsc - filtersc) / eslCONST_LOG2;
P = esl_exp_surv(seq_score, om->evparam[p7_FTAU], om->evparam[p7_FLAMBDA]);
if(hmmer_error) {
fprintf(stderr, "HMM: %s, seq: %s", om->name, seq);
hmmer_error = 0;
continue;
}
if (P > f3) continue;
//Real hit, go in to posterior decoding and alignment
p7_omx_Reuse(oxb);
p7_trace_Reuse(tr);
p7_Backward(sq->dsq, sq->n, om, oxf, oxb, NULL);
status = p7_Decoding(om, oxf, oxb, oxb);
if(status == eslOK) {
//And then trace the result
p7_OptimalAccuracy(om, oxb, oxf, &oasc);
p7_OATrace(om, oxb, oxf, tr);
} else if(status == eslERANGE) {
fprintf(stderr, "Decoding overflow on model %s\n", om->name);
gm = gmodels[index];
if(gxf == NULL) {
gxf = p7_gmx_Create(gm->M, sq->n);
gxb = p7_gmx_Create(gm->M, sq->n);
} else {
p7_gmx_GrowTo(gxf, gm->M, sq->n);
p7_gmx_GrowTo(gxb, gm->M, sq->n);
}
p7_ReconfigLength(gm, sq->n);
p7_GForward (sq->dsq, sq->n, gm, gxf, &fwdsc);
p7_GBackward(sq->dsq, sq->n, gm, gxb, NULL);
p7_GDecoding(gm, gxf, gxb, gxb);
p7_GOptimalAccuracy(gm, gxb, gxf, &oasc);
p7_GOATrace (gm, gxb, gxf, tr);
p7_gmx_Reuse(gxf);
p7_gmx_Reuse(gxb);
}
if(hmmer_error) {
fprintf(stderr, "HMM: %s, seq: %s", om->name, seq);
hmmer_error = 0;
continue;
}
result = wrapper_results[num_results];
reuse_result(result, tr->N + om->M, om->name); //We're way overallocating here, but it's hard to know at this point how much space we'll need for the alignment (plus leading and trailing gaps)
trace_into(tr, result, sq, abc, om->M);
result->bits = seq_score;
num_results++;
}
esl_sq_Destroy(sq);
}
/* Function: p7_SSVFilter_longtarget()
* Synopsis: Finds windows with SSV scores above some threshold (vewy vewy fast, in limited precision)
*
* Purpose: Calculates an approximation of the SSV (single ungapped diagonal)
* score for regions of sequence <dsq> of length <L> residues, using
* optimized profile <om>, and a preallocated one-row DP matrix <ox>,
* and captures the positions at which such regions exceed the score
* required to be significant in the eyes of the calling function,
* which depends on the <bg> and <p> (usually p=0.02 for nhmmer).
* Note that this variant performs only SSV computations, never
* passing through the J state - the score required to pass SSV at
* the default threshold (or less restrictive) is sufficient to
* pass MSV in essentially all DNA models we've tested.
*
* Above-threshold diagonals are captured into a preallocated list
* <windowlist>. Rather than simply capturing positions at which a
* score threshold is reached, this function establishes windows
* around those high-scoring positions, using scores in <msvdata>.
* These windows can be merged by the calling function.
*
*
* Args: dsq - digital target sequence, 1..L
* L - length of dsq in residues
* om - optimized profile
* ox - DP matrix
* msvdata - compact representation of substitution scores, for backtracking diagonals
* bg - the background model, required for translating a P-value threshold into a score threshold
* P - p-value below which a region is captured as being above threshold
* windowlist - preallocated container for all hits (resized if necessary)
*
*
* Note: We misuse the matrix <ox> here, using only a third of the
* first dp row, accessing it as <dp[0..Q-1]> rather than
* in triplets via <{MDI}MX(q)> macros, since we only need
* to store M state values. We know that if <ox> was big
* enough for normal DP calculations, it must be big enough
* to hold the MSVFilter calculation.
*
* Returns: <eslOK> on success.
*
* Throws: <eslEINVAL> if <ox> allocation is too small.
*/
int
p7_SSVFilter_longtarget(const ESL_DSQ *dsq, int L, P7_OPROFILE *om, P7_OMX *ox, const P7_SCOREDATA *msvdata,
P7_BG *bg, double P, P7_HMM_WINDOWLIST *windowlist)
{
register __m128i mpv; /* previous row values */
register __m128i xEv; /* E state: keeps max for Mk->E for a single iteration */
register __m128i xBv; /* B state: splatted vector of B[i-1] for B->Mk calculations */
register __m128i sv; /* temp storage of 1 curr row value in progress */
register __m128i biasv; /* emission bias in a vector */
uint8_t xJ; /* special states' scores */
int i; /* counter over sequence positions 1..L */
int q; /* counter over vectors 0..nq-1 */
int Q = p7O_NQB(om->M); /* segment length: # of vectors */
__m128i *dp = ox->dpb[0]; /* we're going to use dp[0][0..q..Q-1], not {MDI}MX(q) macros*/
__m128i *rsc; /* will point at om->rbv[x] for residue x[i] */
__m128i tecv; /* vector for E->C cost */
__m128i tjbmv; /* vector for J->B move cost + B->M move costs */
__m128i basev; /* offset for scores */
__m128i ceilingv; /* saturated simd value used to test for overflow */
__m128i tempv; /* work vector */
int cmp;
int k;
int n;
int end;
int rem_sc;
int start;
int target_end;
int target_start;
int max_end;
int max_sc;
int sc;
int pos_since_max;
float ret_sc;
union { __m128i v; uint8_t b[16]; } u;
/*
* Computing the score required to let P meet the F1 prob threshold
* In original code, converting from a scaled int MSV
* score S (the score getting to state E) to a probability goes like this:
* usc = S - om->tec_b - om->tjb_b - om->base_b;
* usc /= om->scale_b;
* usc -= 3.0;
* P = f ( (usc - nullsc) / eslCONST_LOG2 , mu, lambda)
* and we're computing the threshold usc, so reverse it:
* (usc - nullsc) / eslCONST_LOG2 = inv_f( P, mu, lambda)
* usc = nullsc + eslCONST_LOG2 * inv_f( P, mu, lambda)
* usc += 3
* usc *= om->scale_b
* S = usc + om->tec_b + om->tjb_b + om->base_b
*
* Here, I compute threshold with length model based on max_length. Doesn't
* matter much - in any case, both the bg and om models will change with roughly
* 1 bit for each doubling of the length model, so they offset.
*/
float nullsc;
__m128i sc_threshv;
uint8_t sc_thresh;
float invP = esl_gumbel_invsurv(P, om->evparam[p7_MMU], om->evparam[p7_MLAMBDA]);
/* Check that the DP matrix is ok for us. */
if (Q > ox->allocQ16) ESL_EXCEPTION(eslEINVAL, "DP matrix allocated too small");
ox->M = om->M;
p7_bg_SetLength(bg, om->max_length);
p7_oprofile_ReconfigMSVLength(om, om->max_length);
p7_bg_NullOne (bg, dsq, om->max_length, &nullsc);
sc_thresh = (int) ceil( ( ( nullsc + (invP * eslCONST_LOG2) + 3.0 ) * om->scale_b ) + om->base_b + om->tec_b + om->tjb_b );
sc_threshv = _mm_set1_epi8((int8_t) 255 - sc_thresh);
/* Initialization. In offset unsigned arithmetic, -infinity is 0, and 0 is om->base.
*/
biasv = _mm_set1_epi8((int8_t) om->bias_b); /* yes, you can set1() an unsigned char vector this way */
ceilingv = _mm_cmpeq_epi8(biasv, biasv);
for (q = 0; q < Q; q++) dp[q] = _mm_setzero_si128();
xJ = 0;
basev = _mm_set1_epi8((int8_t) om->base_b);
tecv = _mm_set1_epi8((int8_t) om->tec_b);
tjbmv = _mm_set1_epi8((int8_t) om->tjb_b + (int8_t) om->tbm_b);
xBv = _mm_subs_epu8(basev, tjbmv);
for (i = 1; i <= L; i++) {
rsc = om->rbv[dsq[i]];
xEv = _mm_setzero_si128();
/* Right shifts by 1 byte. 4,8,12,x becomes x,4,8,12.
* Because ia32 is littlendian, this means a left bit shift.
* Zeros shift on automatically, which is our -infinity.
*/
mpv = _mm_slli_si128(dp[Q-1], 1);
for (q = 0; q < Q; q++) {
/* Calculate new MMXo(i,q); don't store it yet, hold it in sv. */
sv = _mm_max_epu8(mpv, xBv);
sv = _mm_adds_epu8(sv, biasv);
sv = _mm_subs_epu8(sv, *rsc); rsc++;
xEv = _mm_max_epu8(xEv, sv);
mpv = dp[q]; /* Load {MDI}(i-1,q) into mpv */
dp[q] = sv; /* Do delayed store of M(i,q) now that memory is usable */
}
/* test if the pthresh significance threshold has been reached;
* note: don't use _mm_cmpgt_epi8, because it's a signed comparison, which won't work on uint8s */
tempv = _mm_adds_epu8(xEv, sc_threshv);
tempv = _mm_cmpeq_epi8(tempv, ceilingv);
cmp = _mm_movemask_epi8(tempv);
if (cmp != 0) { //hit pthresh, so add position to list and reset values
//figure out which model state hit threshold
end = -1;
rem_sc = -1;
for (q = 0; q < Q; q++) { /// Unpack and unstripe, so we can find the state that exceeded pthresh
u.v = dp[q];
for (k = 0; k < 16; k++) { // unstripe
//(q+Q*k+1) is the model position k at which the xE score is found
if (u.b[k] >= sc_thresh && u.b[k] > rem_sc && (q+Q*k+1) <= om->M) {
end = (q+Q*k+1);
rem_sc = u.b[k];
}
}
dp[q] = _mm_set1_epi8(0); // while we're here ... this will cause values to get reset to xB in next dp iteration
}
//recover the diagonal that hit threshold
start = end;
target_end = target_start = i;
sc = rem_sc;
while (rem_sc > om->base_b - om->tjb_b - om->tbm_b) {
rem_sc -= om->bias_b - msvdata->msv_scores[start*om->abc->Kp + dsq[target_start]];
--start;
--target_start;
}
start++;
target_start++;
//extend diagonal further with single diagonal extension
k = end+1;
n = target_end+1;
max_end = target_end;
max_sc = sc;
pos_since_max = 0;
while (k<om->M && n<=L) {
sc += om->bias_b - msvdata->msv_scores[k*om->abc->Kp + dsq[n]];
if (sc >= max_sc) {
max_sc = sc;
max_end = n;
pos_since_max=0;
} else {
pos_since_max++;
if (pos_since_max == 5)
break;
}
k++;
n++;
}
end += (max_end - target_end);
k += (max_end - target_end);
target_end = max_end;
ret_sc = ((float) (max_sc - om->tjb_b) - (float) om->base_b);
ret_sc /= om->scale_b;
ret_sc -= 3.0; // that's ~ L \log \frac{L}{L+3}, for our NN,CC,JJ
p7_hmmwindow_new(windowlist, 0, target_start, k, end, end-start+1 , ret_sc, p7_NOCOMPLEMENT );
i = target_end; // skip forward
}
} /* end loop over sequence residues 1..L */
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);
float nu = esl_opt_GetReal(go, "--nu");
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;
float sc, nullsc, seqscore, P;
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);
/* Open sequence file */
sq = esl_sq_CreateDigital(abc);
status = esl_sqfile_Open(seqfile, format, NULL, &sqfp);
if (status == eslENOTFOUND) p7_Fail("No such file.");
else if (status == eslEFORMAT) p7_Fail("Format unrecognized.");
else if (status == eslEINVAL) p7_Fail("Can't autodetect stdin or .gz.");
else if (status != eslOK) p7_Fail("Open failed, code %d.", status);
/* Configure a profile from the HMM */
bg = p7_bg_Create(abc);
gm = p7_profile_Create(hmm->M, abc);
p7_ProfileConfig(hmm, bg, gm, sq->n, p7_LOCAL);
/* Allocate matrix */
fwd = p7_gmx_Create(gm->M, sq->n);
while ((status = esl_sqio_Read(sqfp, sq)) == eslOK)
{
p7_ReconfigLength(gm, sq->n);
p7_bg_SetLength(bg, sq->n);
p7_gmx_GrowTo(fwd, gm->M, sq->n);
/* Run MSV */
p7_GMSV(sq->dsq, sq->n, gm, fwd, nu, &sc);
/* Calculate bit score and P-value using standard null1 model*/
p7_bg_NullOne (bg, sq->dsq, sq->n, &nullsc);
seqscore = (sc - nullsc) / eslCONST_LOG2;
P = esl_gumbel_surv(seqscore, gm->evparam[p7_MMU], gm->evparam[p7_MLAMBDA]);
/* output suitable for direct use in profmark benchmark postprocessors:
* <Pvalue> <bitscore> <target name> <query name>
*/
printf("%g\t%.2f\t%s\t%s\n", P, seqscore, sq->name, hmm->name);
esl_sq_Reuse(sq);
}
if (status == eslEFORMAT) esl_fatal("Parse failed (sequence file %s):\n%s\n", sqfp->filename, esl_sqfile_GetErrorBuf(sqfp));
else if (status != eslEOF) esl_fatal("Unexpected error %d reading sequence file %s", status, sqfp->filename);
/* Cleanup */
esl_sqfile_Close(sqfp);
esl_sq_Destroy(sq);
p7_trace_Destroy(tr);
p7_gmx_Destroy(fwd);
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_GETOPTS *go = p7_CreateDefaultApp(options, 2, argc, argv, banner, usage);
ESL_ALPHABET *abc = esl_alphabet_Create(eslAMINO);
char *qfile = esl_opt_GetArg(go, 1);
char *tfile = esl_opt_GetArg(go, 2);
ESL_SQFILE *qfp = NULL;
ESL_SQFILE *tfp = NULL;
ESL_SQ *qsq = esl_sq_CreateDigital(abc);
ESL_SQ *tsq = esl_sq_CreateDigital(abc);
ESL_SCOREMATRIX *S = esl_scorematrix_Create(abc);
ESL_DMATRIX *Q = NULL;
P7_BG *bg = p7_bg_Create(abc);
P7_HMM *hmm = NULL;
P7_PROFILE *gm = NULL;
P7_REFMX *vit = p7_refmx_Create(200, 400); /* will grow as needed */
double *fa = malloc(sizeof(double) * abc->K);
double popen = 0.02;
double pextend = 0.4;
double lambda;
float vsc;
float nullsc;
int status;
esl_composition_BL62(fa);
esl_vec_D2F(fa, abc->K, bg->f);
esl_scorematrix_Set("BLOSUM62", S);
esl_scorematrix_ProbifyGivenBG(S, fa, fa, &lambda, &Q);
esl_scorematrix_JointToConditionalOnQuery(abc, Q);
if (esl_sqfile_OpenDigital(abc, qfile, eslSQFILE_UNKNOWN, NULL, &qfp) != eslOK) esl_fatal("failed to open %s", qfile);
if (esl_sqio_Read(qfp, qsq) != eslOK) esl_fatal("failed to read query seq");
p7_Seqmodel(abc, qsq->dsq, qsq->n, qsq->name, Q, bg->f, popen, pextend, &hmm);
p7_hmm_SetComposition(hmm);
p7_hmm_SetConsensus(hmm, qsq);
gm = p7_profile_Create(hmm->M, abc);
p7_profile_ConfigUnilocal(gm, hmm, bg, 400);
if (esl_sqfile_OpenDigital(abc, tfile, eslSQFILE_UNKNOWN, NULL, &tfp) != eslOK) esl_fatal("failed to open %s", tfile);
while ((status = esl_sqio_Read(tfp, tsq)) == eslOK)
{
p7_bg_SetLength (bg, tsq->n);
p7_profile_SetLength(gm, tsq->n);
p7_ReferenceViterbi(tsq->dsq, tsq->n, gm, vit, NULL, &vsc);
p7_bg_NullOne(bg, tsq->dsq, tsq->n, &nullsc);
printf("%.4f %-25s %-25s\n", (vsc - nullsc) / eslCONST_LOG2, tsq->name, gm->name);
esl_sq_Reuse(tsq);
p7_refmx_Reuse(vit);
}
p7_refmx_Destroy(vit);
p7_profile_Destroy(gm);
p7_hmm_Destroy(hmm);
p7_bg_Destroy(bg);
esl_dmatrix_Destroy(Q);
esl_scorematrix_Destroy(S);
free(fa);
esl_sq_Destroy(qsq);
esl_sq_Destroy(tsq);
esl_sqfile_Close(qfp);
esl_sqfile_Close(tfp);
esl_alphabet_Destroy(abc);
esl_getopts_Destroy(go);
return 0;
}
int
main(int argc, char **argv)
{
ESL_GETOPTS *go = p7_CreateDefaultApp(options, 2, argc, argv, banner, usage);
char *hmmfile = esl_opt_GetArg(go, 1);
char *seqfile = esl_opt_GetArg(go, 2);
ESL_ALPHABET *abc = NULL;
P7_HMMFILE *hfp = NULL;
P7_HMM *hmm = NULL;
P7_BG *bg = NULL;
P7_PROFILE *gm = NULL;
P7_GMX *fwd = NULL;
P7_GMX *bck = NULL;
ESL_SQ *sq = NULL;
ESL_SQFILE *sqfp = NULL;
int format = eslSQFILE_UNKNOWN;
float fsc, bsc;
float nullsc;
int status;
/* Initialize log-sum calculator */
p7_FLogsumInit();
/* Read in one HMM */
if (p7_hmmfile_OpenE(hmmfile, NULL, &hfp, NULL) != eslOK) p7_Fail("Failed to open HMM file %s", hmmfile);
if (p7_hmmfile_Read(hfp, &abc, &hmm) != eslOK) p7_Fail("Failed to read HMM");
p7_hmmfile_Close(hfp);
/* Read in one sequence */
sq = esl_sq_CreateDigital(abc);
status = esl_sqfile_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);
/* Configure a profile from the HMM */
bg = p7_bg_Create(abc);
gm = p7_profile_Create(hmm->M, abc);
/* Now reconfig the models however we were asked to */
if (esl_opt_GetBoolean(go, "--fs")) p7_ProfileConfig(hmm, bg, gm, sq->n, p7_LOCAL);
else if (esl_opt_GetBoolean(go, "--sw")) p7_ProfileConfig(hmm, bg, gm, sq->n, p7_UNILOCAL);
else if (esl_opt_GetBoolean(go, "--ls")) p7_ProfileConfig(hmm, bg, gm, sq->n, p7_GLOCAL);
else if (esl_opt_GetBoolean(go, "--s")) p7_ProfileConfig(hmm, bg, gm, sq->n, p7_UNIGLOCAL);
/* Allocate matrices */
fwd = p7_gmx_Create(gm->M, sq->n);
bck = p7_gmx_Create(gm->M, sq->n);
printf("%-30s %-10s %-10s %-10s %-10s\n", "# seq name", "fwd (raw)", "bck (raw) ", "fwd (bits)", "bck (bits)");
printf("%-30s %10s %10s %10s %10s\n", "#--------------", "----------", "----------", "----------", "----------");
while ( (status = esl_sqio_Read(sqfp, sq)) != eslEOF)
{
if (status == eslEFORMAT) p7_Fail("Parse failed (sequence file %s)\n%s\n", sqfp->filename, sqfp->get_error(sqfp));
else if (status != eslOK) p7_Fail("Unexpected error %d reading sequence file %s", status, sqfp->filename);
/* Resize the DP matrices if necessary */
p7_gmx_GrowTo(fwd, gm->M, sq->n);
p7_gmx_GrowTo(bck, gm->M, sq->n);
/* Set the profile and null model's target length models */
p7_bg_SetLength(bg, sq->n);
p7_ReconfigLength(gm, sq->n);
/* Run Forward, Backward */
p7_GForward (sq->dsq, sq->n, gm, fwd, &fsc);
p7_GBackward(sq->dsq, sq->n, gm, bck, &bsc);
p7_gmx_Dump(stdout, fwd, p7_DEFAULT);
/* Those scores are partial log-odds likelihoods in nats.
* Subtract off the rest of the null model, convert to bits.
*/
p7_bg_NullOne(bg, sq->dsq, sq->n, &nullsc);
printf("%-30s %10.4f %10.4f %10.4f %10.4f\n",
sq->name,
fsc, bsc,
(fsc - nullsc) / eslCONST_LOG2, (bsc - nullsc) / eslCONST_LOG2);
p7_gmx_Reuse(fwd);
p7_gmx_Reuse(bck);
esl_sq_Reuse(sq);
}
/* Cleanup */
esl_sqfile_Close(sqfp);
esl_sq_Destroy(sq);
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_getopts_Destroy(go);
return 0;
}
/* Function: p7_SSVFM_longlarget()
* Synopsis: Finds windows with SSV scores above given threshold, using FM-index
*
* Details: Uses FM-index to find high-scoring diagonals (seeds), then extends those
* seeds to maximal scoring diagonals (no gaps). Windows meeting the SSV
* scoring threshold (usually score s.t. p=0.02) are captured, and passed
* on to the Viterbi and Forward stages of the pipeline.
*
* Args: om - optimized profile
* nu - configuration: expected number of hits (use 2.0 as a default)
* bg - the background model, required for translating a P-value threshold into a score threshold
* F1 - p-value below which a window is captured as being above threshold
* fmf - data for forward traversal of the FM-index
* fmb - data for backward traversal of the FM-index
* fm_cfg - FM-index meta data
* ssvdata - compact data required for computing SSV scores
* windowlist - RETURN: collection of SSV-passing windows, with meta data required for downstream stages.
*
* Returns: <eslOK> on success.
*
* Throws: <eslEMEM> if trouble allocating memory for seeds
*/
int
p7_SSVFM_longlarget( P7_OPROFILE *om, float nu, P7_BG *bg, double F1,
const FM_DATA *fmf, const FM_DATA *fmb, FM_CFG *fm_cfg, const P7_SCOREDATA *ssvdata,
P7_HMM_WINDOWLIST *windowlist)
{
float sc_thresh, sc_threshFM;
float invP, invP_FM;
float nullsc;
int i;
float tloop = logf((float) om->max_length / (float) (om->max_length+3));
float tloop_total = tloop * om->max_length;
float tmove = logf( 3.0f / (float) (om->max_length+3));
float tbmk = logf( 2.0f / ((float) om->M * (float) (om->M+1)));
float tec = logf(1.0f / nu);
FM_DIAG *diag;
ESL_SQ *tmp_sq;
FM_DIAGLIST seeds;
int status;
status = fm_initSeeds(&seeds);
if (status != eslOK)
ESL_EXCEPTION(eslEMEM, "Error allocating memory for seed list\n");
/* Set false target length. This is a conservative estimate of the length of window that'll
* soon be passed on to later phases of the pipeline; used to recover some bits of the score
* that we would miss if we left length parameters set to the full target length */
p7_oprofile_ReconfigMSVLength(om, om->max_length);
p7_bg_SetLength(bg, om->max_length);
p7_bg_NullOne (bg, NULL, om->max_length, &nullsc);
tmp_sq = esl_sq_CreateDigital(om->abc);
/*
* Computing the score required to let P meet the F1 prob threshold
* In original code, converting from an SSV score S (the score getting
* to state C) to a probability goes like this:
* S = XMX(L,p7G_C)
* usc = S + tmove + tloop_total
* P = f ( (usc - nullsc) / eslCONST_LOG2 , mu, lambda)
* and XMX(C) was the diagonal score + tmove + tbmk + tec
* and we're computing the threshold score S, so reverse it:
* (usc - nullsc) / eslCONST_LOG2 = inv_f( P, mu, lambda)
* usc = nullsc + eslCONST_LOG2 * inv_f( P, mu, lambda)
* S = usc - tmove - tloop_total - tmove - tbmk - tec
*
*
* Here, I compute threshold with length model based on max_length. Usually, the
* length of a window returned by this scan will be 2*max_length-1 or longer. Doesn't
* really matter - in any case, both the bg and om models will change with roughly
* 1 bit for each doubling of the length model, so they offset.
*/
invP = esl_gumbel_invsurv(F1, om->evparam[p7_MMU], om->evparam[p7_MLAMBDA]);
sc_thresh = (invP * eslCONST_LOG2) + nullsc - (tmove + tloop_total + tmove + tbmk + tec);
invP_FM = esl_gumbel_invsurv(0.5, om->evparam[p7_MMU], om->evparam[p7_MLAMBDA]);
sc_threshFM = ESL_MAX(fm_cfg->scthreshFM, (invP_FM * eslCONST_LOG2) + nullsc - (tmove + tloop_total + tmove + tbmk + tec) ) ;
sc_threshFM *= fm_cfg->info_deficit_ratio;
sc_threshFM = ESL_MAX(7.0, sc_threshFM);
//get diagonals that score above sc_threshFM
status = FM_getSeeds(fmf, fmb, fm_cfg, ssvdata, om->abc->Kp, sc_threshFM, &seeds );
if (status != eslOK)
ESL_EXCEPTION(eslEMEM, "Error allocating memory for seed computation\n");
//now extend those diagonals to find ones scoring above sc_thresh
for(i=0; i<seeds.count; i++) {
FM_extendSeed( seeds.diags+i, fmf, ssvdata, fm_cfg, tmp_sq);
}
for(i=0; i<seeds.count; i++) {
diag = seeds.diags+i;
if (diag->score >= sc_thresh)
FM_window_from_diag(diag, fmf, fm_cfg->meta, windowlist );
}
esl_sq_Destroy(tmp_sq);
free(seeds.diags);
return eslEOF;
//ERROR:
// ESL_EXCEPTION(eslEMEM, "Error allocating memory for hit list\n");
}