uint8_t *bns_fetch_seq(const bntseq_t *bns, const uint8_t *pac, int64_t *beg, int64_t mid, int64_t *end, int *rid)
{
int64_t far_beg, far_end, len;
int is_rev;
uint8_t *seq;
if (*end < *beg) *end ^= *beg, *beg ^= *end, *end ^= *beg; // if end is smaller, swap
assert(*beg <= mid && mid < *end);
*rid = bns_pos2rid(bns, bns_depos(bns, mid, &is_rev));
far_beg = bns->anns[*rid].offset;
far_end = far_beg + bns->anns[*rid].len;
if (is_rev) { // flip to the reverse strand
int64_t tmp = far_beg;
far_beg = (bns->l_pac<<1) - far_end;
far_end = (bns->l_pac<<1) - tmp;
}
*beg = *beg > far_beg? *beg : far_beg;
*end = *end < far_end? *end : far_end;
seq = bns_get_seq(bns->l_pac, pac, *beg, *end, &len);
if (seq == 0 || *end - *beg != len) {
fprintf(stderr, "[E::%s] begin=%ld, mid=%ld, end=%ld, len=%ld, seq=%p, rid=%d, far_beg=%ld, far_end=%ld\n",
__func__, (long)*beg, (long)mid, (long)*end, (long)len, seq, *rid, (long)far_beg, (long)far_end);
}
assert(seq && *end - *beg == len); // assertion failure should never happen
return seq;
}
void loadTranscriptsFromFMD() {
bwaidx_t* idx_ = salmonIndex_->bwaIndex();
size_t numRecords = idx_->bns->n_seqs;
std::vector<Transcript> transcripts_tmp;
//transcripts_tmp.reserve(numRecords);
//transcripts_.reserve(numRecords);
fmt::print(stderr, "Index contained {} targets\n", numRecords);
//transcripts_.resize(numRecords);
for (auto i : boost::irange(size_t(0), numRecords)) {
uint32_t id = i;
char* name = idx_->bns->anns[i].name;
uint32_t len = idx_->bns->anns[i].len;
// copy over the length, then we're done.
transcripts_tmp.emplace_back(id, name, len);
}
std::sort(transcripts_tmp.begin(), transcripts_tmp.end(),
[](const Transcript& t1, const Transcript& t2) -> bool {
return t1.id < t2.id;
});
double alpha = 0.005;
char nucTab[256];
nucTab[0] = 'A'; nucTab[1] = 'C'; nucTab[2] = 'G'; nucTab[3] = 'T';
for (size_t i = 4; i < 256; ++i) { nucTab[i] = 'N'; }
size_t tnum = 0;
// Load the transcript sequence from file
for (auto& t : transcripts_tmp) {
transcripts_.emplace_back(t.id, t.RefName.c_str(), t.RefLength, alpha);
/* from BWA */
uint8_t* rseq = nullptr;
int64_t tstart, tend, compLen, l_pac = idx_->bns->l_pac;
tstart = idx_->bns->anns[t.id].offset;
tend = tstart + t.RefLength;
rseq = bns_get_seq(l_pac, idx_->pac, tstart, tend, &compLen);
if (compLen != t.RefLength) {
fmt::print(stderr,
"For transcript {}, stored length ({}) != computed length ({}) --- index may be corrupt. exiting\n",
t.RefName, compLen, t.RefLength);
std::exit(1);
}
std::string seq(t.RefLength, ' ');
if (rseq != 0) {
for (int64_t i = 0; i < compLen; ++i) { seq[i] = nucTab[rseq[i]]; }
}
auto& txp = transcripts_.back();
// allocate space for the new copy
char* seqCopy = new char[seq.length()+1];
std::strcpy(seqCopy, seq.c_str());
txp.Sequence = seqCopy;
txp.freeSeqOnDestruct = false;
txp.SAMSequence = salmon::stringtools::encodeSequenceInSAM(seq.c_str(), t.RefLength);
// Length classes taken from
// ======
// Roberts, Adam, et al.
// "Improving RNA-Seq expression estimates by correcting for fragment bias."
// Genome Biol 12.3 (2011): R22.
// ======
// perhaps, define these in a more data-driven way
if (t.RefLength <= 1334) {
txp.lengthClassIndex(0);
} else if (t.RefLength <= 2104) {
txp.lengthClassIndex(0);
} else if (t.RefLength <= 2988) {
txp.lengthClassIndex(0);
} else if (t.RefLength <= 4389) {
txp.lengthClassIndex(0);
} else {
txp.lengthClassIndex(0);
}
/*
std::cerr << "TS = " << t.RefName << " : \n";
std::cerr << seq << "\n VS \n";
for (size_t i = 0; i < t.RefLength; ++i) {
std::cerr << transcripts_.back().charBaseAt(i);
}
std::cerr << "\n\n";
*/
free(rseq);
/* end BWA code */
++tnum;
}
// Since we have the de-coded reference sequences, we no longer need
// the encoded sequences, so free them.
/** TEST OPT **/
// free(idx_->pac); idx_->pac = nullptr;
/** END TEST OPT **/
transcripts_tmp.clear();
// ====== Done loading the transcripts from file
}