/* Function: p7_ParameterEstimation()
* Incept: SRE, Sat Mar 24 10:15:37 2007 [Janelia]
*
* Purpose: Given an <hmm> containing collected, weighted counts;
* and given a mixture Dirichlet prior <pri>;
* calculate mean posterior parameter estimates for
* all model parameters, converting the
* HMM to a parameterized probabilistic model.
*
* Returns: <eslOK> on success.
*/
int
p7_ParameterEstimation(P7_HMM *hmm, const P7_PRIOR *pri)
{
int k;
double c[p7_MAXABET];
double p[p7_MAXABET];
double mix[p7_MAXDCHLET];
/* Match transitions 0,1..M: 0 is the B state
* TMD at node M is 0.
*/
for (k = 0; k <= hmm->M; k++) {
esl_vec_F2D(hmm->t[k], 3, c);
esl_mixdchlet_MPParameters(c, 3, pri->tm, mix, p);
esl_vec_D2F(p, 3, hmm->t[k]);
}
hmm->t[hmm->M][p7H_MD] = 0.0;
esl_vec_FNorm(hmm->t[hmm->M], 3);
/* Insert transitions, 0..M
*/
for (k = 0; k <= hmm->M; k++) {
esl_vec_F2D(hmm->t[k]+3, 2, c);
esl_mixdchlet_MPParameters(c, 2, pri->ti, mix, p);
esl_vec_D2F(p, 2, hmm->t[k]+3);
}
/* Delete transitions, 1..M-1
* For k=0, which is unused; convention sets TMM=1.0, TMD=0.0
* For k=M, TMM = 1.0 (to the E state) and TMD=0.0 (no next D; must go to E).
*/
for (k = 1; k < hmm->M; k++) {
esl_vec_F2D(hmm->t[k]+5, 2, c);
esl_mixdchlet_MPParameters(c, 2, pri->td, mix, p);
esl_vec_D2F(p, 2, hmm->t[k]+5);
}
hmm->t[0][p7H_DM] = hmm->t[hmm->M][p7H_DM] = 1.0;
hmm->t[0][p7H_DD] = hmm->t[hmm->M][p7H_DD] = 0.0;
/* Match emissions, 1..M
* Convention sets mat[0] to a valid pvector: first elem 1, the rest 0.
*/
for (k = 1; k <= hmm->M; k++) {
esl_vec_F2D(hmm->mat[k], hmm->abc->K, c);
esl_mixdchlet_MPParameters(c, hmm->abc->K, pri->em, mix, p);
esl_vec_D2F(p, hmm->abc->K, hmm->mat[k]);
}
esl_vec_FSet(hmm->mat[0], hmm->abc->K, 0.);
hmm->mat[0][0] = 1.0;
/* Insert emissions 0..M
*/
for (k = 0; k <= hmm->M; k++) {
esl_vec_F2D(hmm->ins[k], hmm->abc->K, c);
esl_mixdchlet_MPParameters(c, hmm->abc->K, pri->ei, mix, p);
esl_vec_D2F(p, hmm->abc->K, hmm->ins[k]);
}
return eslOK;
}
/* Function: p7_Seqmodel()
* Synopsis: Make a profile HMM from a single sequence.
*
* Purpose: Make a profile HMM from a single sequence, for
* probabilistic Smith/Waterman alignment, HMMER3-style.
*
* The query is digital sequence <dsq> of length <M>
* residues in alphabet <abc>, named <name>.
*
* The scoring system is given by <Q>, <f>, <popen>, and
* <pextend>. <Q> is a $K \times K$ matrix giving
* conditional residue probabilities $P(a \mid b)}$; these
* are typically obtained by reverse engineering a score
* matrix like BLOSUM62. <f> is a vector of $K$ background
* frequencies $p_a$. <popen> and <pextend> are the
* probabilities assigned to gap-open ($t_{MI}$ and
* $t_{MD}$) and gap-extend ($t_{II}$ and $t_{DD}$)
* transitions.
*
* The <p7H_SINGLE> flag is set on the <hmm>. Model
* configuration (<p7_profile_Config(), friends> detects
* this flag. <B->Mk> entry transitions include a match
* state occupancy term for profile HMMs, but for single
* queries, that <occ[]> term is assumed 1.0 for all
* positions. See commentary in modelconfig.c.
*
* Args:
*
* Returns: <eslOK> on success, and a newly allocated HMM is returned
* in <ret_hmm>.
*
* Throws: <eslEMEM> on allocation error, and <*ret_hmm> is <NULL>.
*/
int
p7_Seqmodel(const ESL_ALPHABET *abc, ESL_DSQ *dsq, int M, char *name,
ESL_DMATRIX *Q, float *f, double popen, double pextend,
P7_HMM **ret_hmm)
{
int status;
P7_HMM *hmm = NULL;
char *logmsg = "[HMM created from a query sequence]";
int k;
if ((hmm = p7_hmm_Create(M, abc)) == NULL) { status = eslEMEM; goto ERROR; }
for (k = 0; k <= M; k++)
{
/* Use rows of P matrix as source of match emission vectors */
if (k > 0) esl_vec_D2F(Q->mx[(int) dsq[k]], abc->K, hmm->mat[k]);
/* Set inserts to background for now. This will be improved. */
esl_vec_FCopy(f, abc->K, hmm->ins[k]);
hmm->t[k][p7H_MM] = 1.0 - 2 * popen;
hmm->t[k][p7H_MI] = popen;
hmm->t[k][p7H_MD] = popen;
hmm->t[k][p7H_IM] = 1.0 - pextend;
hmm->t[k][p7H_II] = pextend;
hmm->t[k][p7H_DM] = 1.0 - pextend;
hmm->t[k][p7H_DD] = pextend;
}
/* Deal w/ special stuff at node M, overwriting a little of what we
* just did.
*/
hmm->t[M][p7H_MM] = 1.0 - popen;
hmm->t[M][p7H_MD] = 0.;
hmm->t[M][p7H_DM] = 1.0;
hmm->t[M][p7H_DD] = 0.;
/* Add mandatory annotation
*/
p7_hmm_SetName(hmm, name);
p7_hmm_AppendComlog(hmm, 1, &logmsg);
hmm->nseq = 1;
p7_hmm_SetCtime(hmm);
hmm->checksum = 0;
hmm->flags |= p7H_SINGLE;
*ret_hmm = hmm;
return eslOK;
ERROR:
if (hmm != NULL) p7_hmm_Destroy(hmm);
*ret_hmm = NULL;
return status;
}
/* The DChoose() and FChoose() unit tests.
*/
static void
utest_choose(ESL_RANDOMNESS *r, int n, int nbins, int be_verbose)
{
double *pd = NULL;
float *pf = NULL;
int *ct = NULL;
int i;
double X2, diff, exp, X2p;
if ((pd = malloc(sizeof(double) * nbins)) == NULL) esl_fatal("malloc failed");
if ((pf = malloc(sizeof(float) * nbins)) == NULL) esl_fatal("malloc failed");
if ((ct = malloc(sizeof(int) * nbins)) == NULL) esl_fatal("malloc failed");
/* Sample a random multinomial probability vector. */
if (esl_dirichlet_DSampleUniform(r, nbins, pd) != eslOK) esl_fatal("dirichlet sample failed");
esl_vec_D2F(pd, nbins, pf);
/* Sample observed counts using DChoose(). */
esl_vec_ISet(ct, nbins, 0);
for (i = 0; i < n; i++)
ct[esl_rnd_DChoose(r, pd, nbins)]++;
/* X^2 test on those observed counts. */
for (X2 = 0., i=0; i < nbins; i++) {
exp = (double) n * pd[i];
diff = (double) ct[i] - exp;
X2 += diff*diff/exp;
}
if (esl_stats_ChiSquaredTest(nbins, X2, &X2p) != eslOK) esl_fatal("chi square eval failed");
if (be_verbose) printf("DChoose(): \t%g\n", X2p);
if (X2p < 0.01) esl_fatal("chi squared test failed");
/* Repeat above for FChoose(). */
esl_vec_ISet(ct, nbins, 0);
for (i = 0; i < n; i++)
ct[esl_rnd_FChoose(r, pf, nbins)]++;
for (X2 = 0., i=0; i < nbins; i++) {
exp = (double) n * pd[i];
diff = (double) ct[i] - exp;
X2 += diff*diff/exp;
}
if (esl_stats_ChiSquaredTest(nbins, X2, &X2p) != eslOK) esl_fatal("chi square eval failed");
if (be_verbose) printf("FChoose(): \t%g\n", X2p);
if (X2p < 0.01) esl_fatal("chi squared test failed");
free(pd);
free(pf);
free(ct);
return;
}
static void
utest_pvectors(void)
{
char *msg = "pvector unit test failed";
double p1[4] = { 0.25, 0.25, 0.25, 0.25 };
double p2[4];
double p3[4];
float p1f[4];
float p2f[4] = { 0.0, 0.5, 0.5, 0.0 };
float p3f[4];
int n = 4;
double result;
esl_vec_D2F(p1, n, p1f);
esl_vec_F2D(p2f, n, p2);
if (esl_vec_DValidate(p1, n, 1e-12, NULL) != eslOK) esl_fatal(msg);
if (esl_vec_FValidate(p1f, n, 1e-7, NULL) != eslOK) esl_fatal(msg);
result = esl_vec_DEntropy(p1, n); if (esl_DCompare(2.0, result, 1e-9) != eslOK) esl_fatal(msg);
result = esl_vec_FEntropy(p1f, n); if (esl_DCompare(2.0, result, 1e-9) != eslOK) esl_fatal(msg);
result = esl_vec_DEntropy(p2, n); if (esl_DCompare(1.0, result, 1e-9) != eslOK) esl_fatal(msg);
result = esl_vec_FEntropy(p2f, n); if (esl_DCompare(1.0, result, 1e-9) != eslOK) esl_fatal(msg);
result = esl_vec_DRelEntropy(p2, p1, n); if (esl_DCompare(1.0, result, 1e-9) != eslOK) esl_fatal(msg);
result = esl_vec_FRelEntropy(p2f, p1f, n); if (esl_DCompare(1.0, result, 1e-9) != eslOK) esl_fatal(msg);
result = esl_vec_DRelEntropy(p1, p2, n); if (result != eslINFINITY) esl_fatal(msg);
result = esl_vec_FRelEntropy(p1f, p2f, n); if (result != eslINFINITY) esl_fatal(msg);
esl_vec_DLog(p2, n);
if (esl_vec_DLogValidate(p2, n, 1e-12, NULL) != eslOK) esl_fatal(msg);
esl_vec_DExp(p2, n);
if (p2[0] != 0.) esl_fatal(msg);
esl_vec_FLog(p2f, n);
if (esl_vec_FLogValidate(p2f, n, 1e-7, NULL) != eslOK) esl_fatal(msg);
esl_vec_FExp(p2f, n);
if (p2f[0] != 0.) esl_fatal(msg);
esl_vec_DCopy(p2, n, p3);
esl_vec_DScale(p3, n, 10.);
esl_vec_DNorm(p3, n);
if (esl_vec_DCompare(p2, p3, n, 1e-12) != eslOK) esl_fatal(msg);
esl_vec_DLog(p3, n);
result = esl_vec_DLogSum(p3, n); if (esl_DCompare(0.0, result, 1e-12) != eslOK) esl_fatal(msg);
esl_vec_DIncrement(p3, n, 2.0);
esl_vec_DLogNorm(p3, n);
if (esl_vec_DCompare(p2, p3, n, 1e-12) != eslOK) esl_fatal(msg);
esl_vec_FCopy(p2f, n, p3f);
esl_vec_FScale(p3f, n, 10.);
esl_vec_FNorm(p3f, n);
if (esl_vec_FCompare(p2f, p3f, n, 1e-7) != eslOK) esl_fatal(msg);
esl_vec_FLog(p3f, n);
result = esl_vec_FLogSum(p3f, n); if (esl_DCompare(0.0, result, 1e-7) != eslOK) esl_fatal(msg);
esl_vec_FIncrement(p3f, n, 2.0);
esl_vec_FLogNorm(p3f, n);
if (esl_vec_FCompare(p2f, p3f, n, 1e-7) != eslOK) esl_fatal(msg);
return;
}
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;
}
