/* Function: p7_checkptmx_DumpFBRow()
* Synopsis: Dump one row from fwd or bck version of the matrix.
*
* Purpose: Dump current row <dpc> of forward or backward calculations from
* DP matrix <ox> for diagnostics. The index <rowi> is used
* as a row label, along with an additional free-text label
* <pfx>. (The checkpointed backward implementation
* interleaves backward row calculations with recalculated
* fwd rows, both of which it is dumping; they need to be
* labeled something like "fwd" and "bck" to distinguish
* them in the debugging dump.)
*/
int
p7_checkptmx_DumpFBRow(P7_CHECKPTMX *ox, int rowi, __m128 *dpc, char *pfx)
{
union { __m128 v; float x[p7_VNF]; } u;
float *v = NULL; /* */
int Q = ox->Qf;
int M = ox->M;
float *xc = (float *) (dpc + Q*p7C_NSCELLS);
int logify = (ox->dump_flags & p7_SHOW_LOG) ? TRUE : FALSE;
int maxpfx = ox->dump_maxpfx;
int width = ox->dump_width;
int precision = ox->dump_precision;
int k,q,z;
int status;
ESL_ALLOC(v, sizeof(float) * ( (Q*p7_VNF) + 1));
v[0] = 0.;
/* Line 1. M cells: unpack, unstripe, print */
for (q = 0; q < Q; q++) {
u.v = P7C_MQ(dpc, q);
for (z = 0; z < p7_VNF; z++) v[q+Q*z+1] = u.x[z];
}
fprintf(ox->dfp, "%*s %3d M", maxpfx, pfx, rowi);
for (k = 0; k <= M; k++) fprintf(ox->dfp, " %*.*f", width, precision, (logify ? esl_logf(v[k]) : v[k]));
/* a static analyzer may complain about v[k] being uninitialized
* if it isn't smart enough to see that M,Q are linked.
*/
/* Line 1 end: Specials */
for (z = 0; z < p7C_NXCELLS; z++)
fprintf(ox->dfp, " %*.*f", width, precision, (logify ? esl_logf(xc[z]) : xc[z]));
fputc('\n', ox->dfp);
/* Line 2: I cells: unpack, unstripe, print */
for (q = 0; q < Q; q++) {
u.v = P7C_IQ(dpc, q);
for (z = 0; z < p7_VNF; z++) v[q+Q*z+1] = u.x[z];
}
fprintf(ox->dfp, "%*s %3d I", maxpfx, pfx, rowi);
for (k = 0; k <= M; k++) fprintf(ox->dfp, " %*.*f", width, precision, (logify ? esl_logf(v[k]) : v[k]));
fputc('\n', ox->dfp);
/* Line 3. D cells: unpack, unstripe, print */
for (q = 0; q < Q; q++) {
u.v = P7C_DQ(dpc, q);
for (z = 0; z < p7_VNF; z++) v[q+Q*z+1] = u.x[z];
}
fprintf(ox->dfp, "%*s %3d D", maxpfx, pfx, rowi);
for (k = 0; k <= M; k++) fprintf(ox->dfp, " %*.*f", width, precision, (logify ? esl_logf(v[k]) : v[k]));
fputc('\n', ox->dfp);
fputc('\n', ox->dfp);
free(v);
return eslOK;
ERROR:
if (v) free(v);
return status;
}
/* Function: p7_null3_score()
*
* Purpose: Calculate a correction (in log_2 odds) to be applied
* to a sequence, using a null model based on the
* composition of the target sequence.
* The null model is constructed /post hoc/ as the
* distribution of the target sequence; if the target
* sequence is 40% A, 5% C, 5% G, 40% T, then the null
* model is (0.4, 0.05, 0.05, 0.4). This function is
* based heavily on Infernal's ScoreCorrectionNull3(),
* with two important changes:
* - it leaves the log2 conversion from NATS to BITS
* for the calling function.
* - it doesn't include the omega score modifier
* (based on prior probability of using the null3
* model), again leaving this to the calling function.
*
* Args: abc - alphabet for hit (only used to get alphabet size)
* dsq - the sequence the hit resides in
* tr - trace of the alignment, used to find the match states
* (non-match chars are ignored in computing freq, not used if NULL)
* start - start position of hit in dsq
* stop - end position of hit in dsq
* bg - background, used for the default null model's emission freq
* ret_sc - RETURN: the correction to the score (in NATS);
* caller subtracts this from hit score to get
* corrected score.
* Return: void, ret_sc: the log-odds score correction (in NATS).
*/
void
p7_null3_score(const ESL_ALPHABET *abc, const ESL_DSQ *dsq, P7_TRACE *tr, int start, int stop, P7_BG *bg, float *ret_sc)
{
float score = 0.;
int status;
int i;
float *freq;
int dir;
int tr_pos;
ESL_ALLOC(freq, sizeof(float) * abc->K);
esl_vec_FSet(freq, abc->K, 0.0);
/* contract check */
if(abc == NULL) esl_exception(eslEINVAL, FALSE, __FILE__, __LINE__, "p7_null3_score() alphabet is NULL.%s\n", "");
if(dsq == NULL) esl_exception(eslEINVAL, FALSE, __FILE__, __LINE__, "p7_null3_score() dsq alphabet is NULL.%s\n", "");
if(abc->type != eslRNA && abc->type != eslDNA) esl_exception(eslEINVAL, FALSE, __FILE__, __LINE__, "p7_null3_score() expects alphabet of RNA or DNA.%s\n", "");
dir = start < stop ? 1 : -1;
if (tr != NULL) {
/* skip the parts of the trace that precede the first match state */
tr_pos = 2;
i = start;
while (tr->st[tr_pos] != p7T_M) {
if (tr->st[tr_pos] == p7T_N)
i += dir;
tr_pos++;
}
/* tally frequencies from characters hitting match state*/
while (tr->st[tr_pos] != p7T_E) {
if (tr->st[tr_pos] == p7T_M) {
if(esl_abc_XIsGap(abc, dsq[i])) esl_exception(eslEINVAL, FALSE, __FILE__, __LINE__, "in p7_null3_score(), res %d is a gap!%s\n", "");
esl_abc_FCount(abc, freq, dsq[i], 1.);
}
if (tr->st[tr_pos] != p7T_D )
i += dir;
tr_pos++;
}
} else {
/* tally frequencies from the full envelope */
for (i=ESL_MIN(start,stop); i <= ESL_MAX(start,stop); i++)
{
if(esl_abc_XIsGap(abc, dsq[i])) esl_exception(eslEINVAL, FALSE, __FILE__, __LINE__, "in p7_null3_score(), res %d is a gap!%s\n", "");
esl_abc_FCount(abc, freq, dsq[i], 1.);
}
}
esl_vec_FNorm(freq, abc->K);
/* now compute score modifier (nats) - note: even with tr!=NULL, this includes the unmatched characters*/
for (i = 0; i < abc->K; i++)
score += freq[i]==0 ? 0.0 : esl_logf( freq[i]/bg->f[i] ) * freq[i] * ( (stop-start)*dir +1) ;
/* Return the correction to the bit score. */
score = p7_FLogsum(0., score);
*ret_sc = score;
return;
ERROR:
esl_exception(eslEINVAL, FALSE, __FILE__, __LINE__, "p7_null3_score() memory allocation error.%s\n", "");
return; /* never reached */
}