static bwa_seq_t *bwa_read_bam(bwa_seqio_t *bs, int n_needed, int *n, int is_comp, int trim_qual)
{
bwa_seq_t *seqs, *p;
int n_seqs, l, i;
long n_trimmed = 0, n_tot = 0;
bam1_t *b;
int res;
b = bam_init1();
n_seqs = 0;
seqs = (bwa_seq_t*)calloc(n_needed, sizeof(bwa_seq_t));
while ((res = bam_read1(bs->fp, b)) >= 0) {
uint8_t *s, *q;
int go = 0;
if ((bs->which & 1) && (b->core.flag & BAM_FREAD1)) go = 1;
if ((bs->which & 2) && (b->core.flag & BAM_FREAD2)) go = 1;
if ((bs->which & 4) && !(b->core.flag& BAM_FREAD1) && !(b->core.flag& BAM_FREAD2))go = 1;
if (go == 0) continue;
l = b->core.l_qseq;
p = &seqs[n_seqs++];
p->tid = -1; // no assigned to a thread
p->qual = 0;
p->full_len = p->clip_len = p->len = l;
n_tot += p->full_len;
s = bam1_seq(b); q = bam1_qual(b);
p->seq = (ubyte_t*)calloc(p->len + 1, 1);
p->qual = (ubyte_t*)calloc(p->len + 1, 1);
for (i = 0; i != p->full_len; ++i) {
p->seq[i] = bam_nt16_nt4_table[(int)bam1_seqi(s, i)];
p->qual[i] = q[i] + 33 < 126? q[i] + 33 : 126;
}
if (bam1_strand(b)) { // then reverse
seq_reverse(p->len, p->seq, 1);
seq_reverse(p->len, p->qual, 0);
}
if (trim_qual >= 1) n_trimmed += bwa_trim_read(trim_qual, p);
p->rseq = (ubyte_t*)calloc(p->full_len, 1);
memcpy(p->rseq, p->seq, p->len);
seq_reverse(p->len, p->seq, 0); // *IMPORTANT*: will be reversed back in bwa_refine_gapped()
seq_reverse(p->len, p->rseq, is_comp);
p->name = strdup((const char*)bam1_qname(b));
if (n_seqs == n_needed) break;
}
if (res < 0 && res != -1) err_fatal_simple("Error reading bam file");
*n = n_seqs;
if (n_seqs && trim_qual >= 1)
fprintf(stderr, "[bwa_read_seq] %.1f%% bases are trimmed.\n", 100.0f * n_trimmed/n_tot);
if (n_seqs == 0) {
free(seqs);
bam_destroy1(b);
return 0;
}
bam_destroy1(b);
return seqs;
}
void BWA::copy_bases_into_sequence(bwa_seq_t* sequence, const char* bases, const unsigned read_length)
{
// seq, rseq will ultimately be freed by bwa_cal_sa_reg_gap
sequence->seq = new ubyte_t[read_length];
sequence->rseq = new ubyte_t[read_length];
for(unsigned i = 0; i < read_length; i++) sequence->seq[i] = nst_nt4_table[(unsigned)bases[i]];
memcpy(sequence->rseq,sequence->seq,read_length);
// BWA expects the read bases to arrive reversed.
seq_reverse(read_length,sequence->seq,0);
seq_reverse(read_length,sequence->rseq,1);
sequence->full_len = sequence->len = read_length;
}
struct rld_t *fml_fmi_gen(int n, bseq1_t *seq, int is_mt)
{
mrope_t *mr;
kstring_t str = {0,0,0};
mritr_t itr;
rlditr_t di;
const uint8_t *block;
rld_t *e = 0;
int k;
for (k = 0; k < n; ++k)
if (seq[k].l_seq > 0)
break;
if (k == n) return 0;
mr = mr_init(ROPE_DEF_MAX_NODES, ROPE_DEF_BLOCK_LEN, MR_SO_RCLO);
for (k = 0; k < n; ++k) {
int i;
bseq1_t *s = &seq[k];
if (s->l_seq == 0) continue;
free(s->qual);
for (i = 0; i < s->l_seq; ++i)
s->seq[i] = seq_nt6_table[(int)s->seq[i]];
for (i = 0; i < s->l_seq; ++i)
if (s->seq[i] == 5) break;
if (i < s->l_seq) {
free(s->seq);
continue;
}
if (is_rev_same(s->l_seq, s->seq))
--s->l_seq, s->seq[s->l_seq] = 0;
seq_reverse(s->l_seq, (uint8_t*)s->seq);
kputsn(s->seq, s->l_seq + 1, &str);
seq_revcomp6(s->l_seq, (uint8_t*)s->seq);
kputsn(s->seq, s->l_seq + 1, &str);
free(s->seq);
}
free(seq);
mr_insert_multi(mr, str.l, (uint8_t*)str.s, is_mt);
free(str.s);
e = rld_init(6, 3);
rld_itr_init(e, &di, 0);
mr_itr_first(mr, &itr, 1);
while ((block = mr_itr_next_block(&itr)) != 0) {
const uint8_t *q = block + 2, *end = block + 2 + *rle_nptr(block);
while (q < end) {
int c = 0;
int64_t l;
rle_dec1(q, c, l);
rld_enc(e, &di, l, c);
}
}
rld_enc_finish(e, &di);
mr_destroy(mr);
return e;
}
// Mostly stolen from bwa_read_bam.
void bam1_to_seq(bam1_t *raw, bwa_seq_t *p, int is_comp, int trim_qual)
{
// long n_trimmed = 0;
uint8_t *s, *q;
int i, l = raw->core.l_qseq;
p->tid = -1; // no assigned to a thread
p->qual = 0;
p->full_len = p->clip_len = p->len = l;
// n_tot += p->full_len;
s = bam1_seq(raw);
q = bam1_qual(raw);
p->seq = (ubyte_t*)calloc(p->len + 1, 1);
p->qual = (ubyte_t*)calloc(p->len + 1, 1);
for (i = 0; i != p->full_len; ++i) {
p->seq[i] = bam_nt16_nt4_table[(int)bam1_seqi(s, i)];
p->qual[i] = q[i] + 33 < 126? q[i] + 33 : 126;
}
if (bam1_strand(raw)) { // then reverse
seq_reverse(p->len, p->seq, 1);
seq_reverse(p->len, p->qual, 0);
}
if (trim_qual >= 1) /* n_trimmed += */ bwa_trim_read(trim_qual, p);
p->rseq = (ubyte_t*)calloc(p->full_len, 1);
memcpy(p->rseq, p->seq, p->len);
seq_reverse(p->len, p->seq, 0); // *IMPORTANT*: will be reversed back in bwa_refine_gapped()
seq_reverse(p->len, p->rseq, is_comp);
p->max_entries = 0 ;
// We don't set a name, it's contained in the original record
// anyway.
// p->name = strdup((const char*)bam1_qname(raw));
// No place to put the tally right now.
// if (n_seqs && trim_qual >= 1)
// fprintf(stderr, "[bwa_read_seq] %.1f%% bases are trimmed.\n", 100.0f * n_trimmed/n_tot);
}
bwa_seq_t *bwa_read_seq(bwa_seqio_t *bs, int n_needed, int *n, int mode, int trim_qual)
{
bwa_seq_t *seqs, *p;
kseq_t *seq = bs->ks;
int n_seqs, l, i, is_comp = mode&BWA_MODE_COMPREAD, is_64 = mode&BWA_MODE_IL13, l_bc = mode>>24;
long n_trimmed = 0, n_tot = 0;
if (l_bc > BWA_MAX_BCLEN) {
fprintf(stderr, "[%s] the maximum barcode length is %d.\n", __func__, BWA_MAX_BCLEN);
return 0;
}
if (bs->is_bam) return bwa_read_bam(bs, n_needed, n, is_comp, trim_qual); // l_bc has no effect for BAM input
n_seqs = 0;
seqs = (bwa_seq_t*)calloc(n_needed, sizeof(bwa_seq_t));
while ((l = kseq_read(seq)) >= 0) {
if ((mode & BWA_MODE_CFY) && (seq->comment.l != 0)) {
// skip reads that are marked to be filtered by Casava
char *s = index(seq->comment.s, ':');
if (s && *(++s) == 'Y') {
continue;
}
}
if (is_64 && seq->qual.l)
for (i = 0; i < seq->qual.l; ++i) seq->qual.s[i] -= 31;
if (seq->seq.l <= l_bc) continue; // sequence length equals or smaller than the barcode length
p = &seqs[n_seqs++];
if (l_bc) { // then trim barcode
for (i = 0; i < l_bc; ++i)
p->bc[i] = (seq->qual.l && seq->qual.s[i]-33 < BARCODE_LOW_QUAL)? tolower(seq->seq.s[i]) : toupper(seq->seq.s[i]);
p->bc[i] = 0;
for (; i < seq->seq.l; ++i)
seq->seq.s[i - l_bc] = seq->seq.s[i];
seq->seq.l -= l_bc; seq->seq.s[seq->seq.l] = 0;
if (seq->qual.l) {
for (i = l_bc; i < seq->qual.l; ++i)
seq->qual.s[i - l_bc] = seq->qual.s[i];
seq->qual.l -= l_bc; seq->qual.s[seq->qual.l] = 0;
}
l = seq->seq.l;
} else p->bc[0] = 0;
p->tid = -1; // no assigned to a thread
p->qual = 0;
p->full_len = p->clip_len = p->len = l;
n_tot += p->full_len;
p->seq = (ubyte_t*)calloc(p->len, 1);
for (i = 0; i != p->full_len; ++i)
p->seq[i] = nst_nt4_table[(int)seq->seq.s[i]];
if (seq->qual.l) { // copy quality
p->qual = (ubyte_t*)strdup((char*)seq->qual.s);
if (trim_qual >= 1) n_trimmed += bwa_trim_read(trim_qual, p);
}
p->rseq = (ubyte_t*)calloc(p->full_len, 1);
memcpy(p->rseq, p->seq, p->len);
seq_reverse(p->len, p->seq, 0); // *IMPORTANT*: will be reversed back in bwa_refine_gapped()
seq_reverse(p->len, p->rseq, is_comp);
p->name = strdup((const char*)seq->name.s);
{ // trim /[12]$
int t = strlen(p->name);
if (t > 2 && p->name[t-2] == '/' && (p->name[t-1] == '1' || p->name[t-1] == '2')) p->name[t-2] = '\0';
}
if (n_seqs == n_needed) break;
}
*n = n_seqs;
if (n_seqs && trim_qual >= 1)
fprintf(stderr, "[bwa_read_seq] %.1f%% bases are trimmed. \n", 100.0f * n_trimmed/n_tot);
if (n_seqs == 0) {
free(seqs);
return 0;
}
return seqs;
}
void BWA::generate_alignments_from_paths(const char* bases,
const unsigned read_length,
bwt_aln1_t* paths,
const unsigned num_paths,
const unsigned best_count,
const unsigned second_best_count,
Alignment*& alignments,
unsigned& num_alignments)
{
bwa_seq_t* sequence = create_sequence(bases,read_length);
sequence->aln = paths;
sequence->n_aln = num_paths;
// (Ab)use bwa_aln2seq to propagate values stored in the path out into the sequence itself.
bwa_aln2seq(sequence->n_aln,sequence->aln,sequence);
// But overwrite key parts of the sequence in case the user passed back only a smaller subset
// of the paths.
sequence->c1 = best_count;
sequence->c2 = second_best_count;
sequence->type = sequence->c1 > 1 ? BWA_TYPE_REPEAT : BWA_TYPE_UNIQUE;
num_alignments = 0;
for(unsigned i = 0; i < (unsigned)sequence->n_aln; i++)
num_alignments += (sequence->aln + i)->l - (sequence->aln + i)->k + 1;
alignments = new Alignment[num_alignments];
unsigned alignment_idx = 0;
for(unsigned path_idx = 0; path_idx < (unsigned)num_paths; path_idx++) {
// Stub in a 'working' path, so that only the desired alignment is local-aligned.
const bwt_aln1_t* path = paths + path_idx;
bwt_aln1_t working_path = *path;
// Loop through all alignments, aligning each one individually.
for(unsigned sa_idx = path->k; sa_idx <= path->l; sa_idx++) {
working_path.k = working_path.l = sa_idx;
sequence->aln = &working_path;
sequence->n_aln = 1;
sequence->sa = sa_idx;
sequence->strand = path->a;
sequence->score = path->score;
// Each time through bwa_refine_gapped, seq gets reversed. Revert the reverse.
// TODO: Fix the interface to bwa_refine_gapped so its easier to work with.
if(alignment_idx > 0)
seq_reverse(sequence->len, sequence->seq, 0);
// Copy the local alignment data into the alignment object.
*(alignments + alignment_idx) = generate_final_alignment_from_sequence(sequence);
alignment_idx++;
}
}
sequence->aln = NULL;
sequence->n_aln = 0;
bwa_free_read_seq(1,sequence);
}
void bwa_rg_tpx(int iidx, const bntseq_t *bns, int n_seqs1, int n_seqs2,
bwa_seq_t *seqs, ubyte_t *pacseq, bntseq_t *ntbns)
{
ubyte_t *ntpac = 0;
int i, j;
kstring_t *str;
#ifdef _TIMING
struct timeval st;
uint64_t s1, e1;
double pos1_time = 0.0;
#endif
#ifdef _TIMING
gettimeofday(&st, NULL);
s1 = st.tv_sec * 1000000L + (time_t)st.tv_usec;
#endif
for (i = n_seqs1; i < n_seqs2; ++i) {
bwa_seq_t *s = seqs + i;
seq_reverse(s->len, s->seq, 0); // IMPORTANT: s->seq is reversed here!!!
for (j = 0; j < s->n_multi; ++j) {
bwt_multi1_t *q = s->multi + j;
int n_cigar;
if (q->gap == 0) continue;
q->cigar = refine_gapped_core(bns->l_pac, pacseq, s->len, q->strand? s->rseq : s->seq, &q->pos,
(q->strand? 1 : -1) * q->gap, &n_cigar, 1);
q->n_cigar = n_cigar;
}
if (s->type == BWA_TYPE_NO_MATCH || s->type == BWA_TYPE_MATESW || s->n_gapo == 0) continue;
s->cigar = refine_gapped_core(bns->l_pac, pacseq, s->len, s->strand? s->rseq : s->seq, &s->pos,
(s->strand? 1 : -1) * (s->n_gapo + s->n_gape), &s->n_cigar, 1);
}
if (ntbns) { // in color space
for (i = n_seqs1; i < n_seqs2; ++i) {
bwa_seq_t *s = seqs + i;
bwa_cs2nt_core(s, bns->l_pac, ntpac);
for (j = 0; j < s->n_multi; ++j) {
bwt_multi1_t *q = s->multi + j;
int n_cigar;
if (q->gap == 0) continue;
free(q->cigar);
q->cigar = refine_gapped_core(bns->l_pac, ntpac, s->len, q->strand? s->rseq : s->seq, &q->pos,
(q->strand? 1 : -1) * q->gap, &n_cigar, 0);
q->n_cigar = n_cigar;
}
if (s->type != BWA_TYPE_NO_MATCH && s->cigar) { // update cigar again
free(s->cigar);
s->cigar = refine_gapped_core(bns->l_pac, ntpac, s->len, s->strand? s->rseq : s->seq, &s->pos,
(s->strand? 1 : -1) * (s->n_gapo + s->n_gape), &s->n_cigar, 0);
}
}
}
// generate MD tag
str = (kstring_t*)calloc(1, sizeof(kstring_t));
for (i = n_seqs1; i < n_seqs2; ++i) {
bwa_seq_t *s = seqs + i;
if (s->type != BWA_TYPE_NO_MATCH) {
int nm;
s->md = bwa_cal_md1(s->n_cigar, s->cigar, s->len, s->pos, s->strand? s->rseq : s->seq,
bns->l_pac, ntbns? ntpac : pacseq, str, &nm);
s->nm = nm;
}
}
free(str->s); free(str);
// correct for trimmed reads
if (!ntbns) // trimming is only enabled for Illumina reads
for (i = n_seqs1; i < n_seqs2; ++i) bwa_correct_trimmed(seqs + i);
#ifdef _TIMING
gettimeofday(&st, NULL);
e1 = st.tv_sec * 1000000L + (time_t)st.tv_usec;
pos1_time = (double)((double)e1 - (double)s1) / 1000000.0;
# ifdef HAVE_PTHREAD
pthread_mutex_lock(&pe_lock);
# endif // HAVE_PTHREAD
fprintf(stderr,"bwapese1 time = %lf (sec)\n",pos1_time);
# ifdef HAVE_PTHREAD
pthread_mutex_unlock(&pe_lock);
# endif // HAVE_PTHREAD
#endif
return;
}
s_align* ssw_align (const s_profile* prof,
const int8_t* ref,
int32_t refLen,
const uint8_t weight_gapO,
const uint8_t weight_gapE,
const uint8_t flag, // (from high to low) bit 5: return the best alignment beginning position; 6: if (ref_end1 - ref_begin1 <= filterd) && (read_end1 - read_begin1 <= filterd), return cigar; 7: if max score >= filters, return cigar; 8: always return cigar; if 6 & 7 are both setted, only return cigar when both filter fulfilled
const uint16_t filters,
const int32_t filterd,
const int32_t maskLen) {
alignment_end* bests = 0, *bests_reverse = 0;
__m128i* vP = 0;
int32_t word = 0, band_width = 0, readLen = prof->readLen;
int8_t* read_reverse = 0;
cigar* path;
s_align* r = (s_align*)calloc(1, sizeof(s_align));
r->ref_begin1 = -1;
r->read_begin1 = -1;
r->cigar = 0;
r->cigarLen = 0;
if (maskLen < 15) {
fprintf(stderr, "When maskLen < 15, the function ssw_align doesn't return 2nd best alignment information.\n");
}
// Find the alignment scores and ending positions
if (prof->profile_byte) {
bests = sw_sse2_byte(ref, 0, refLen, readLen, weight_gapO, weight_gapE, prof->profile_byte, -1, prof->bias, maskLen);
if (prof->profile_word && bests[0].score == 255) {
free(bests);
bests = sw_sse2_word(ref, 0, refLen, readLen, weight_gapO, weight_gapE, prof->profile_word, -1, maskLen);
word = 1;
} else if (bests[0].score == 255) {
fprintf(stderr, "Please set 2 to the score_size parameter of the function ssw_init, otherwise the alignment results will be incorrect.\n");
free(r);
return NULL;
}
}else if (prof->profile_word) {
bests = sw_sse2_word(ref, 0, refLen, readLen, weight_gapO, weight_gapE, prof->profile_word, -1, maskLen);
word = 1;
}else {
fprintf(stderr, "Please call the function ssw_init before ssw_align.\n");
free(r);
return NULL;
}
r->score1 = bests[0].score;
r->ref_end1 = bests[0].ref;
r->read_end1 = bests[0].read;
if (maskLen >= 15) {
r->score2 = bests[1].score;
r->ref_end2 = bests[1].ref;
} else {
r->score2 = 0;
r->ref_end2 = -1;
}
free(bests);
if (flag == 0 || (flag == 2 && r->score1 < filters)) goto end;
// Find the beginning position of the best alignment.
read_reverse = seq_reverse(prof->read, r->read_end1);
if (word == 0) {
vP = qP_byte(read_reverse, prof->mat, r->read_end1 + 1, prof->n, prof->bias);
bests_reverse = sw_sse2_byte(ref, 1, r->ref_end1 + 1, r->read_end1 + 1, weight_gapO, weight_gapE, vP, r->score1, prof->bias, maskLen);
} else {
vP = qP_word(read_reverse, prof->mat, r->read_end1 + 1, prof->n);
bests_reverse = sw_sse2_word(ref, 1, r->ref_end1 + 1, r->read_end1 + 1, weight_gapO, weight_gapE, vP, r->score1, maskLen);
}
free(vP);
free(read_reverse);
r->ref_begin1 = bests_reverse[0].ref;
r->read_begin1 = r->read_end1 - bests_reverse[0].read;
free(bests_reverse);
if ((7&flag) == 0 || ((2&flag) != 0 && r->score1 < filters) || ((4&flag) != 0 && (r->ref_end1 - r->ref_begin1 > filterd || r->read_end1 - r->read_begin1 > filterd))) goto end;
// Generate cigar.
refLen = r->ref_end1 - r->ref_begin1 + 1;
readLen = r->read_end1 - r->read_begin1 + 1;
band_width = abs(refLen - readLen) + 1;
path = banded_sw(ref + r->ref_begin1, prof->read + r->read_begin1, refLen, readLen, r->score1, weight_gapO, weight_gapE, band_width, prof->mat, prof->n);
if (path == 0) {
free(r);
r = NULL;
}
else {
r->cigar = path->seq;
r->cigarLen = path->length;
free(path);
}
end:
return r;
}
int bwa_read_seq1(bwa_seqio_t *bs, int iter, int tid, int thrds, bwa_seq_t **_seqs, int *n_avail, int mode, int trim_qual)
{
bwa_seq_t *p;
bwa_seq_t *seqs = *_seqs;
kseq_t *seq = bs->ks;
int n_seqs, l, i, is_comp = mode&BWA_MODE_COMPREAD, is_64 = mode&BWA_MODE_IL13, l_bc = mode>>24;
long n_trimmed = 0, n_tot = 0;
bool first;
if (l_bc > BWA_MAX_BCLEN) {
fprintf(stderr, "[%s] the maximum barcode length is %d.\n", __func__, BWA_MAX_BCLEN);
return 0;
}
if (bs->is_bam) {
fprintf (stderr, "IS BAM! --- Port bwa_read_bam function\n");
bwa_free_read_seq(*n_avail, seqs);
// return bwa_read_bam(bs, n_needed, n, is_comp, trim_qual); // l_bc has no effect for BAM input
exit(0);
}
if (*n_avail == 0) {
if (*n_avail) bwa_free_read_seq(*n_avail, seqs);
seqs = (bwa_seq_t*)calloc(READ_SEQ_SIZE, sizeof(bwa_seq_t));
*_seqs = seqs;
*n_avail = READ_SEQ_SIZE;
}
n_seqs = 0;
first = true;
//err_fwrite("FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF", strlen("FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF"), 1, stdout);
//long cksm = 0;
while ((l = kseq_read1(seq, iter, tid, thrds, &first)) >= 0) {
if ((mode & BWA_MODE_CFY) && (seq->comment.l != 0)) {
// skip reads that are marked to be filtered by Casava
char *s = index(seq->comment.s, ':');
if (s && *(++s) == 'Y') {
continue;
}
}
if (is_64 && seq->qual.l)
for (i = 0; i < seq->qual.l; ++i) seq->qual.s[i] -= 31;
if (seq->seq.l <= l_bc) continue; // sequence length equals or smaller than the barcode length
p = &seqs[n_seqs++];
if(n_seqs > READ_SEQ_SIZE) {
fprintf (stderr, "READ_SEQ_SIZE not big enough\n");
abort();
}
init_bwa_seq_t(p);
if (l_bc) { // then trim barcode
for (i = 0; i < l_bc; ++i)
p->bc[i] = (seq->qual.l && seq->qual.s[i]-33 < BARCODE_LOW_QUAL)? tolower(seq->seq.s[i]) : toupper(seq->seq.s[i]);
p->bc[i] = 0;
for (; i < seq->seq.l; ++i)
seq->seq.s[i - l_bc] = seq->seq.s[i];
seq->seq.l -= l_bc; seq->seq.s[seq->seq.l] = 0;
if (seq->qual.l) {
for (i = l_bc; i < seq->qual.l; ++i)
seq->qual.s[i - l_bc] = seq->qual.s[i];
seq->qual.l -= l_bc; seq->qual.s[seq->qual.l] = 0;
}
l = seq->seq.l;
} else p->bc[0] = 0;
//ComputeChecksum(seq->seq.s,seq->seq.l,&cksm);
p->tid = -1; // no assigned to a thread
p->full_len = p->clip_len = p->len = l;
n_tot += p->full_len;
if (p->llen < p->len) {
if(p->seq) free(p->seq);
p->llen = p->len;
p->seq = (ubyte_t*)calloc(p->len, 1);
p->rseq = (ubyte_t*)calloc(p->full_len, 1);
}
for (i = 0; i != p->full_len; ++i)
p->seq[i] = nst_nt4_table[(int)seq->seq.s[i]];
if (seq->qual.l) { // copy quality
if(p->qual) free(p->qual);
p->qual = (ubyte_t*)strdup((char*)seq->qual.s);
if (trim_qual >= 1) n_trimmed += bwa_trim_read(trim_qual, p);
}
memcpy(p->rseq, p->seq, p->len);
seq_reverse(p->len, p->seq, 0); // *IMPORTANT*: will be reversed back in bwa_refine_gapped()
seq_reverse(p->len, p->rseq, is_comp);
if(p->name) free(p->name);
p->name = strdup((const char*)seq->name.s);
{ // trim /[12]$
int t = strlen(p->name);
if (t > 2 && p->name[t-2] == '/' && (p->name[t-1] == '1' || p->name[t-1] == '2')) p->name[t-2] = '\0';
}
//if (n_seqs == n_needed) break;
if(kseq_end(seq)) break;
}
if (n_seqs && trim_qual >= 1)
fprintf(stderr, "[bwa_read_seq1] %.1f%% bases are trimmed.\n", 100.0f * n_trimmed/n_tot);
//fprintf(stderr, "%d tid: %d sequences: %d cksum: %lx\n", iter, tid, n_seqs, cksm);
return n_seqs;
}
static bwa_seq_t *bwa_read_bam(bwa_seqio_t *bs, int n_needed, int *n, int is_comp, int trim_qual)
{
bwa_seq_t *seqs, *p;
int n_seqs, l, i;
long n_trimmed = 0, n_tot = 0;
bam1_t *b;
int res;
b = bam_init1();
n_seqs = 0;
seqs = (bwa_seq_t*)calloc(n_needed, sizeof(bwa_seq_t));
#ifdef USE_HTSLIB
while ((res = sam_read1(bs->fp, bs->h, b)) >= 0) {
#else
while ((res = bam_read1(bs->fp, b)) >= 0) {
#endif
uint8_t *s, *q;
int go = 0;
if ((bs->which & 1) && (b->core.flag & BAM_FREAD1)) go = 1;
if ((bs->which & 2) && (b->core.flag & BAM_FREAD2)) go = 1;
if ((bs->which & 4) && !(b->core.flag& BAM_FREAD1) && !(b->core.flag& BAM_FREAD2))go = 1;
if (go == 0) continue;
l = b->core.l_qseq;
p = &seqs[n_seqs++];
p->tid = -1; // no assigned to a thread
p->qual = 0;
p->full_len = p->clip_len = p->len = l;
n_tot += p->full_len;
#ifdef USE_HTSLIB
s = bam_get_seq(b); q = bam_get_qual(b);
#else
s = bam1_seq(b); q = bam1_qual(b);
#endif
p->seq = (ubyte_t*)calloc(p->len + 1, 1);
p->qual = (ubyte_t*)calloc(p->len + 1, 1);
for (i = 0; i != p->full_len; ++i) {
#ifdef USE_HTSLIB
p->seq[i] = bam_nt16_nt4_table[(int)bam_seqi(s, i)];
#else
p->seq[i] = bam_nt16_nt4_table[(int)bam1_seqi(s, i)];
#endif
p->qual[i] = q[i] + 33 < 126? q[i] + 33 : 126;
}
#ifdef USE_HTSLIB
if (bam_is_rev(b)) { // then reverse
#else
if (bam1_strand(b)) { // then reverse
#endif
seq_reverse(p->len, p->seq, 1);
seq_reverse(p->len, p->qual, 0);
}
if (trim_qual >= 1) n_trimmed += bwa_trim_read(trim_qual, p);
p->rseq = (ubyte_t*)calloc(p->full_len, 1);
memcpy(p->rseq, p->seq, p->len);
seq_reverse(p->len, p->seq, 0); // *IMPORTANT*: will be reversed back in bwa_refine_gapped()
seq_reverse(p->len, p->rseq, is_comp);
#ifdef USE_HTSLIB
p->name = strdup((const char*)bam_get_qname(b));
#else
p->name = strdup((const char*)bam1_qname(b));
#endif
if (n_seqs == n_needed) break;
}
if (res < 0 && res != -1) err_fatal_simple("Error reading bam file");
*n = n_seqs;
if (n_seqs && trim_qual >= 1)
fprintf(stderr, "[bwa_read_seq] %.1f%% bases are trimmed.\n", 100.0f * n_trimmed/n_tot);
if (n_seqs == 0) {
free(seqs);
bam_destroy1(b);
return 0;
}
bam_destroy1(b);
return seqs;
}
#define BARCODE_LOW_QUAL 13
bwa_seq_t *bwa_read_seq(bwa_seqio_t *bs, int n_needed, int *n, int mode, int trim_qual)
{
bwa_seq_t *seqs, *p;
kseq_t *seq = bs->ks;
int n_seqs, l, i, is_comp = mode&BWA_MODE_COMPREAD, is_64 = mode&BWA_MODE_IL13, l_bc = mode>>24;
long n_trimmed = 0, n_tot = 0;
if (l_bc > BWA_MAX_BCLEN) {
fprintf(stderr, "[%s] the maximum barcode length is %d.\n", __func__, BWA_MAX_BCLEN);
return 0;
}
if (bs->is_bam) return bwa_read_bam(bs, n_needed, n, is_comp, trim_qual); // l_bc has no effect for BAM input
n_seqs = 0;
seqs = (bwa_seq_t*)calloc(n_needed, sizeof(bwa_seq_t));
while ((l = kseq_read(seq)) >= 0) {
if ((mode & BWA_MODE_CFY) && (seq->comment.l != 0)) {
// skip reads that are marked to be filtered by Casava
char *s = index(seq->comment.s, ':');
if (s && *(++s) == 'Y') {
continue;
}
}
if (is_64 && seq->qual.l)
for (i = 0; i < seq->qual.l; ++i) seq->qual.s[i] -= 31;
if (seq->seq.l <= l_bc) continue; // sequence length equals or smaller than the barcode length
p = &seqs[n_seqs++];
if (l_bc) { // then trim barcode
for (i = 0; i < l_bc; ++i)
p->bc[i] = (seq->qual.l && seq->qual.s[i]-33 < BARCODE_LOW_QUAL)? tolower(seq->seq.s[i]) : toupper(seq->seq.s[i]);
p->bc[i] = 0;
for (; i < seq->seq.l; ++i)
seq->seq.s[i - l_bc] = seq->seq.s[i];
seq->seq.l -= l_bc; seq->seq.s[seq->seq.l] = 0;
if (seq->qual.l) {
for (i = l_bc; i < seq->qual.l; ++i)
seq->qual.s[i - l_bc] = seq->qual.s[i];
seq->qual.l -= l_bc; seq->qual.s[seq->qual.l] = 0;
}
l = seq->seq.l;
} else p->bc[0] = 0;
p->tid = -1; // no assigned to a thread
p->qual = 0;
p->full_len = p->clip_len = p->len = l;
n_tot += p->full_len;
p->seq = (ubyte_t*)calloc(p->full_len, 1);
for (i = 0; i != p->full_len; ++i)
p->seq[i] = nst_nt4_table[(int)seq->seq.s[i]];
if (seq->qual.l) { // copy quality
p->qual = (ubyte_t*)strdup((char*)seq->qual.s);
if (trim_qual >= 1) n_trimmed += bwa_trim_read(trim_qual, p);
}
p->rseq = (ubyte_t*)calloc(p->full_len, 1);
memcpy(p->rseq, p->seq, p->len);
seq_reverse(p->len, p->seq, 0); // *IMPORTANT*: will be reversed back in bwa_refine_gapped()
seq_reverse(p->len, p->rseq, is_comp);
p->name = strdup((const char*)seq->name.s);
{ // trim /[12]$
int t = strlen(p->name);
if (t > 2 && p->name[t-2] == '/' && (p->name[t-1] == '1' || p->name[t-1] == '2')) p->name[t-2] = '\0';
}
if (n_seqs == n_needed) break;
}
*n = n_seqs;
if (n_seqs && trim_qual >= 1)
fprintf(stderr, "[bwa_read_seq] %.1f%% bases are trimmed.\n", 100.0f * n_trimmed/n_tot);
if (n_seqs == 0) {
free(seqs);
return 0;
}
return seqs;
}
void bwa_free_read_seq(int n_seqs, bwa_seq_t *seqs)
{
int i, j;
for (i = 0; i != n_seqs; ++i) {
bwa_seq_t *p = seqs + i;
for (j = 0; j < p->n_multi; ++j)
if (p->multi[j].cigar) free(p->multi[j].cigar);
free(p->name);
free(p->seq); free(p->rseq); free(p->qual); free(p->aln); free(p->md); free(p->multi);
free(p->cigar);
}
free(seqs);
}
bwa_seq_t *bwa_read_seq(bwa_seqio_t *bs, unsigned int n_needed, unsigned int *n, int is_comp, int mid)
{
bwa_seq_t *seqs, *p;
kseq_t *seq = bs->ks;
int n_seqs, l, i; //, j;
n_seqs = 0;
seqs = (bwa_seq_t*)calloc(n_needed, sizeof(bwa_seq_t));
while ((l = kseq_read(seq)) >= 0) {
if ( l > MAX_READ_LENGTH ) l = MAX_READ_LENGTH; //put a limit on sequence length
p = &seqs[n_seqs++];
p->tid = -1; // no assigned to a thread
p->qual = 0;
//p->len = l;
p->len = l - mid;
p->seq = (ubyte_t*)calloc(p->len, 1);
for (i = 0; i != p->len; ++i)
p->seq[i] = nst_nt4_table[(int)seq->seq.s[i+mid]];
p->rseq = (ubyte_t*)calloc(p->len, 1);
memcpy(p->rseq, p->seq, p->len);
seq_reverse(p->len, p->seq, 0); // *IMPORTANT*: will be reversed back in bwa_refine_gapped()
seq_reverse(p->len, p->rseq, is_comp);
/*
printf("Options is_comp = opt->mode & BWA_MODE_COMPREAD is %d\n", is_comp);
printf("Forward Sequence:");
int j;
for ( j = 0; j < p->len; ++j)
{
printf("%d", p->seq[j]);
}
printf("\nReverse Sequence:");
for (j = 0; j < p->len; ++j)
{
printf("%d", p->rseq[j]);
}
printf("\n");
*/
p->name = strdup((const char*)seq->name.s);
{ // trim /[12]$
int t = strlen(p->name);
if (t > 2 && p->name[t-2] == '/' && (p->name[t-1] == '1' || p->name[t-1] == '2')) p->name[t-2] = '\0';
}
if (seq->qual.l) // copy quality
{
//p->qual = (ubyte_t*)strdup((char*)seq->qual.s);
p->qual = (ubyte_t*)calloc(p->len, 1);
int i;
for (i = 0; i != p->len; ++i)
{
p->qual[i] = seq->qual.s[mid+i];
}
}
if (n_seqs == n_needed) break;
}
*n = n_seqs;
if (n_seqs == 0) {
free(seqs);
return 0;
}
return seqs;
}
void bsw2_pair1(const bsw2opt_t *opt, int64_t l_pac, const uint8_t *pac, const bsw2pestat_t *st, const bsw2hit_t *h, int l_mseq, const char *mseq, bsw2hit_t *a, int8_t g_mat[25])
{
extern void seq_reverse(int len, ubyte_t *seq, int is_comp);
int64_t k, beg, end;
uint8_t *seq, *ref;
int i;
// compute the region start and end
a->n_seeds = 1; a->flag |= BSW2_FLAG_MATESW; // before calling this routine, *a has been cleared with memset(0); the flag is set with 1<<6/7
if (h->is_rev == 0) {
beg = (int64_t)(h->k + st->avg - EXT_STDDEV * st->std - l_mseq + .499);
if (beg < h->k) beg = h->k;
end = (int64_t)(h->k + st->avg + EXT_STDDEV * st->std + .499);
a->is_rev = 1; a->flag |= 16;
} else {
beg = (int64_t)(h->k + h->end - h->beg - st->avg - EXT_STDDEV * st->std + .499);
end = (int64_t)(h->k + h->end - h->beg - st->avg + EXT_STDDEV * st->std + l_mseq + .499);
if (end > h->k + (h->end - h->beg)) end = h->k + (h->end - h->beg);
a->is_rev = 0;
}
if (beg < 1) beg = 1;
if (end > l_pac) end = l_pac;
if (end - beg < l_mseq) return;
// generate the sequence
seq = malloc(l_mseq + (end - beg));
ref = seq + l_mseq;
for (k = beg; k < end; ++k)
ref[k - beg] = pac[k>>2] >> ((~k&3)<<1) & 0x3;
if (h->is_rev == 0) {
for (i = 0; i < l_mseq; ++i) { // on the reverse strand
int c = nst_nt4_table[(int)mseq[i]];
seq[l_mseq - 1 - i] = c > 3? 4 : 3 - c;
}
} else {
for (i = 0; i < l_mseq; ++i) // on the forward strand
seq[i] = nst_nt4_table[(int)mseq[i]];
}
#ifndef _NO_SSE2
{
ksw_query_t *q;
ksw_aux_t aux[2];
// forward Smith-Waterman
aux[0].T = opt->t; aux[0].gapo = opt->q; aux[0].gape = opt->r; aux[1] = aux[0];
q = ksw_qinit(l_mseq * g_mat[0] < 250? 1 : 2, l_mseq, seq, 5, g_mat);
ksw_sse2(q, end - beg, ref, &aux[0]);
free(q);
if (aux[0].score < opt->t) {
free(seq);
return;
}
++aux[0].qe; ++aux[0].te;
// reverse Smith-Waterman
seq_reverse(aux[0].qe, seq, 0);
seq_reverse(aux[0].te, ref, 0);
q = ksw_qinit(aux[0].qe * g_mat[0] < 250? 1 : 2, aux[0].qe, seq, 5, g_mat);
ksw_sse2(q, aux[0].te, ref, &aux[1]);
free(q);
++aux[1].qe; ++aux[1].te;
// write output
a->G = aux[0].score;
a->G2 = aux[0].score2 > aux[1].score2? aux[0].score2 : aux[1].score2;
if (a->G2 < opt->t) a->G2 = 0;
if (a->G2) a->flag |= BSW2_FLAG_TANDEM;
a->k = beg + (aux[0].te - aux[1].te);
a->len = aux[1].te;
a->beg = aux[0].qe - aux[1].qe;
a->end = aux[0].qe;
}
#else
{
AlnParam ap;
path_t path[2];
int matrix[25];
for (i = 0; i < 25; ++i) matrix[i] = g_mat[i];
ap.gap_open = opt->q; ap.gap_ext = opt->r; ap.gap_end = opt->r;
ap.matrix = matrix; ap.row = 5; ap.band_width = 50;
a->G = aln_local_core(ref, end - beg, seq, l_mseq, &ap, path, 0, opt->t, &a->G2);
if (a->G < opt->t) a->G = 0;
if (a->G2 < opt->t) a->G2 = 0;
a->k = beg + path[0].i - 1;
a->len = path[1].i - path[0].i + 1;
a->beg = path[0].j - 1;
a->end = path[1].j;
}
#endif
if (a->is_rev) i = a->beg, a->beg = l_mseq - a->end, a->end = l_mseq - i;
free(seq);
}
void bwa_print_sam1(const bntseq_t *bns, bwa_seq_t *p, const bwa_seq_t *mate, int mode, int max_top2, const char *bwa_rg_id)
{
int j;
//if (strcmp (p->name, "HWUSI-EAS1600:WT2_250_read_1:11_30_09:3:1:83:1066#0") == 0)
//{
// fprintf (stderr, "found %s\n", p->name);
//}
if (p->type != BWA_TYPE_NO_MATCH || (mate && mate->type != BWA_TYPE_NO_MATCH)) {
int seqid, nn, am = 0, flag = p->extra_flag;
char XT;
if (p->type == BWA_TYPE_NO_MATCH) {
p->pos = mate->pos;
p->strand = mate->strand;
flag |= SAM_FSU;
j = 1;
} else j = pos_end(p) - p->pos; // j is the length of the reference in the alignment
// get seqid
nn = bns_coor_pac2real(bns, p->pos, j, &seqid);
if (p->type != BWA_TYPE_NO_MATCH && p->pos + j - bns->anns[seqid].offset > bns->anns[seqid].len)
flag |= SAM_FSU; // flag UNMAP as this alignment bridges two adjacent reference sequences
// update flag and print it
if (p->strand) flag |= SAM_FSR;
if (mate) {
if (mate->type != BWA_TYPE_NO_MATCH) {
if (mate->strand) flag |= SAM_FMR;
} else flag |= SAM_FMU;
}
printf("%s\t%d\t%s\t", p->name, flag, bns->anns[seqid].name);
printf("%d\t%d\t", (int)(p->pos - bns->anns[seqid].offset + 1), p->mapQ);
// print CIGAR
if (p->cigar) {
for (j = 0; j != p->n_cigar; ++j)
printf("%d%c", __cigar_len(p->cigar[j]), "MIDSN"[__cigar_op(p->cigar[j])]);
} else if (p->type == BWA_TYPE_NO_MATCH) printf("*");
else printf("%dM", p->len);
// print mate coordinate
if (mate && mate->type != BWA_TYPE_NO_MATCH) {
int m_seqid, m_is_N;
long long isize;
am = mate->seQ < p->seQ? mate->seQ : p->seQ; // smaller single-end mapping quality
// redundant calculation here, but should not matter too much
m_is_N = bns_coor_pac2real(bns, mate->pos, mate->len, &m_seqid);
printf("\t%s\t", (seqid == m_seqid)? "=" : bns->anns[m_seqid].name);
isize = (seqid == m_seqid)? pos_5(mate) - pos_5(p) : 0;
if (p->type == BWA_TYPE_NO_MATCH) isize = 0;
printf("%d\t%lld\t", (int)(mate->pos - bns->anns[m_seqid].offset + 1), isize);
} else if (mate) printf("\t=\t%d\t0\t", (int)(p->pos - bns->anns[seqid].offset + 1));
else printf("\t*\t0\t0\t");
// print sequence and quality
if (p->strand == 0)
for (j = 0; j != p->full_len; ++j) putchar("ACGTN"[(int)p->seq[j]]);
else for (j = 0; j != p->full_len; ++j) putchar("TGCAN"[p->seq[p->full_len - 1 - j]]);
putchar('\t');
if (p->qual) {
if (p->strand) seq_reverse(p->len, p->qual, 0); // reverse quality
printf("%s", p->qual);
} else printf("*");
if (bwa_rg_id) printf("\tRG:Z:%s", bwa_rg_id);
if (p->clip_len < p->full_len) printf("\tXC:i:%d", p->clip_len);
if (p->type != BWA_TYPE_NO_MATCH) {
int i;
// calculate XT tag
XT = "NURM"[p->type];
if (nn > 10) XT = 'N';
// print tags
printf("\tXT:A:%c\t%s:i:%d", XT, (mode & BWA_MODE_COMPREAD)? "NM" : "CM", p->nm);
// print XS tag, to be compatible with Cufflinks
if (p->sense_strand != 2 ) printf("\tXS:A:%c", p->sense_strand ? '-':'+' );
else printf("\tXS:A:.");
if (nn) printf("\tXN:i:%d", nn);
if (mate) printf("\tSM:i:%lu\tAM:i:%d", p->seQ, am);
if (p->type != BWA_TYPE_MATESW) { // X0 and X1 are not available for this type of alignment
printf("\tX0:i:%lu", p->c1);
if (p->c1 <= max_top2) printf("\tX1:i:%lu", p->c2);
}
printf("\tXM:i:%d\tXO:i:%d\tXG:i:%d", p->n_mm, p->n_gapo_t + p->n_gapo_q, p->n_gapo_t+p->n_gape_t+p->n_gapo_q+p->n_gape_q);
if (p->md) printf("\tMD:Z:%s", p->md);
// print multiple hits
if (p->n_multi) {
bool header_printed = 0;
for (i = 0; i < p->n_multi; ++i) {
bwt_multi1_t *q = p->multi + i;
j = pos_end_multi(q, p->len) - q->pos;
nn = bns_coor_pac2real(bns, q->pos, j, &seqid);
if(pos_end_multi(q, p->len) - bns->anns[seqid].offset > bns->anns[seqid].len) continue; //the alignment bridges adjacent sequences (chroms)
//TODO: need to avoid this at the first place in the junction discovery step, but this should be rare for mm or human
if (! header_printed) {
header_printed = 1;
printf("\tXA:Z:");
}
int k;
printf("%s,%c%d,", bns->anns[seqid].name, q->strand? '-' : '+',
(int)(q->pos - bns->anns[seqid].offset + 1));
if (q->cigar) {
for (k = 0; k < q->n_cigar; ++k)
printf("%d%c", __cigar_len(q->cigar[k]), "MIDSN"[__cigar_op(q->cigar[k])]);
} else printf("%dM", p->len);
printf(",%d", q->nm); //q->gap_t + q->gap_q + q->mm);
if (q->sense_strand != 2) printf(",%c;", q->sense_strand? '-' : '+' );
else printf(",.;");
}
}
}
putchar('\n');
} else { // this read has no match
ubyte_t *s = p->strand? p->rseq : p->seq;
int flag = p->extra_flag | SAM_FSU;
if (mate && mate->type == BWA_TYPE_NO_MATCH) flag |= SAM_FMU;
printf("%s\t%d\t*\t0\t0\t*\t*\t0\t0\t", p->name, flag);
for (j = 0; j != p->len; ++j) putchar("ACGTN"[(int)s[j]]);
putchar('\t');
if (p->qual) {
if (p->strand) seq_reverse(p->len, p->qual, 0); // reverse quality
printf("%s", p->qual);
} else printf("*");
if (p->clip_len < p->full_len) printf("\tXC:i:%d", p->clip_len);
putchar('\n');
}
}
extern "C" bwa_seq_t * bwa_seed2genome_map(const char* read, int read_len, int strand, uint64_t *num, uint64_t *sa_k, uint64_t *sa_l)
{
bwa_seq_t *p=(bwa_seq_t*)calloc(1, sizeof(bwa_seq_t)) ;
int n_seqs=1 ;
int l = read_len ;
//fprintf(stdout, "read=%s\n", read) ;
p->tid = -1; // no assigned to a thread
p->qual = NULL ;
p->full_len = p->clip_len = p->len = l;
p->seq = (ubyte_t*)calloc(p->len, 1);
for (int i = 0; i != p->full_len; ++i)
{
p->seq[i] = nst_nt4_table[(int)read[i]];
//fprintf(stdout, "seq[%i]=%i\n", i, p->seq[i]) ;
}
p->rseq = (ubyte_t*)calloc(p->full_len, 1);
memcpy(p->rseq, p->seq, p->len);
seq_reverse(p->len, p->seq, 0); // *IMPORTANT*: will be reversed back in bwa_refine_gapped()
seq_reverse(p->len, p->rseq, bwt_opt->mode & BWA_MODE_COMPREAD);
p->name = strdup("seq") ;
p->cigar=NULL ;
mybwa_cal_sa_reg_gap(0, bwt_bwt, n_seqs, p, bwt_opt);
//fprintf(stdout, "n_aln=%i\n", p->n_aln) ;
if (p->n_aln>0)
{
assert(p->n_aln<=2) ;
if (p->aln[0].a==strand)
{
*sa_k=p->aln[0].k ;
*sa_l=p->aln[0].l ;
*num=*sa_l-*sa_k+1 ;
} else
if (p->n_aln>=2 && p->aln[1].a==strand)
{
*sa_k=p->aln[1].k ;
*sa_l=p->aln[1].l ;
*num=*sa_l-*sa_k+1 ;
}
else
{
*sa_k=1 ;
*sa_l=0 ;
*num=0 ;
}
//fprintf(stdout, "k=%lld, l=%lld\n", *sa_k, *sa_l) ;
}
else
{
*sa_k=1 ;
*sa_l=0 ;
*num=0 ;
}
/*p->sa = *sa_k ;
p->c1 = 1 ;
p->type=BWA_TYPE_UNIQUE ;
mybwa_cal_pac_pos_core(bwt_bwt[0], bwt_bwt[1], p, 0, 0);
int len = pos_end(p) - p->pos;
int seq_id=0 ;
bns_coor_pac2real(bwt_bns, p->pos, len, &seq_id) ;
int pos = (int)(p->pos - bwt_bns->anns[seq_id].offset) ;
fprintf(stdout, "seq_id=%i, pos=%i, n_aln=%i, multi=%i\n", seq_id, pos, p->n_aln, p->n_multi) ;
*/
return p ;
}
