18 "github.com/klauspost/pgzip"
19 log "github.com/sirupsen/logrus"
20 "golang.org/x/crypto/blake2b"
23 type tileVariantID uint16 // 1-based
25 type tileLibRef struct {
30 type tileSeq map[string][]tileLibRef
32 func (tseq tileSeq) Variants() ([]tileVariantID, int, int) {
34 for _, refs := range tseq {
35 for _, ref := range refs {
36 if maxtag < int(ref.Tag) {
41 vars := make([]tileVariantID, maxtag+1)
43 for _, refs := range tseq {
44 for _, ref := range refs {
45 if vars[int(ref.Tag)] != 0 {
50 vars[int(ref.Tag)] = ref.Variant
53 return vars, kept, dropped
56 type tileLibrary struct {
59 retainTileSequences bool
62 variant [][][blake2b.Size256]byte
63 refseqs map[string]map[string][]tileLibRef
64 compactGenomes map[string][]tileVariantID
66 seq map[[blake2b.Size256]byte][]byte
68 // if non-nil, write out any tile variants added while tiling
75 func (tilelib *tileLibrary) loadTagSet(newtagset [][]byte) error {
76 // Loading a tagset means either passing it through to the
77 // output (if it's the first one we've seen), or just ensuring
78 // it doesn't disagree with what we already have.
79 if len(newtagset) == 0 {
83 defer tilelib.mtx.Unlock()
84 if tilelib.taglib == nil || tilelib.taglib.Len() == 0 {
85 tilelib.taglib = &tagLibrary{}
86 err := tilelib.taglib.setTags(newtagset)
90 if tilelib.encoder != nil {
91 err = tilelib.encoder.Encode(LibraryEntry{
98 } else if tilelib.taglib.Len() != len(newtagset) {
99 return fmt.Errorf("cannot merge libraries with differing tagsets")
101 current := tilelib.taglib.Tags()
102 for i := range newtagset {
103 if !bytes.Equal(newtagset[i], current[i]) {
104 return fmt.Errorf("cannot merge libraries with differing tagsets")
111 func (tilelib *tileLibrary) loadTileVariants(tvs []TileVariant, variantmap map[tileLibRef]tileVariantID) error {
112 for _, tv := range tvs {
113 // Assign a new variant ID (unique across all inputs)
114 // for each input variant.
115 variantmap[tileLibRef{Tag: tv.Tag, Variant: tv.Variant}] = tilelib.getRef(tv.Tag, tv.Sequence).Variant
120 func (tilelib *tileLibrary) loadCompactGenomes(cgs []CompactGenome, variantmap map[tileLibRef]tileVariantID, onLoadGenome func(CompactGenome)) error {
121 log.Debugf("loadCompactGenomes: %d", len(cgs))
122 var wg sync.WaitGroup
123 errs := make(chan error, 1)
124 for _, cg := range cgs {
129 for i, variant := range cg.Variants {
137 newvariant, ok := variantmap[tileLibRef{Tag: tag, Variant: variant}]
139 err := fmt.Errorf("oops: genome %q has variant %d for tag %d, but that variant was not in its library", cg.Name, variant, tag)
146 log.Tracef("loadCompactGenomes: cg %s tag %d variant %d => %d", cg.Name, tag, variant, newvariant)
147 cg.Variants[i] = newvariant
149 if onLoadGenome != nil {
152 if tilelib.encoder != nil {
153 err := tilelib.encoder.Encode(LibraryEntry{
154 CompactGenomes: []CompactGenome{cg},
164 if tilelib.compactGenomes != nil {
166 defer tilelib.mtx.Unlock()
167 tilelib.compactGenomes[cg.Name] = cg.Variants
176 func (tilelib *tileLibrary) loadCompactSequences(cseqs []CompactSequence, variantmap map[tileLibRef]tileVariantID) error {
177 log.Debugf("loadCompactSequences: %d", len(cseqs))
178 for _, cseq := range cseqs {
179 for _, tseq := range cseq.TileSequences {
180 for i, libref := range tseq {
181 if libref.Variant == 0 {
182 // No variant (e.g., import
183 // dropped tiles with
184 // no-calls) = no translation.
187 v, ok := variantmap[libref]
189 return fmt.Errorf("oops: CompactSequence %q has variant %d for tag %d, but that variant was not in its library", cseq.Name, libref.Variant, libref.Tag)
194 if tilelib.encoder != nil {
195 if err := tilelib.encoder.Encode(LibraryEntry{
196 CompactSequences: []CompactSequence{cseq},
203 defer tilelib.mtx.Unlock()
204 if tilelib.refseqs == nil {
205 tilelib.refseqs = map[string]map[string][]tileLibRef{}
207 for _, cseq := range cseqs {
208 tilelib.refseqs[cseq.Name] = cseq.TileSequences
213 func (tilelib *tileLibrary) LoadDir(ctx context.Context, path string, onLoadGenome func(CompactGenome)) error {
215 var walk func(string) error
216 walk = func(path string) error {
222 fis, err := f.Readdir(-1)
224 files = append(files, path)
227 for _, fi := range fis {
228 if fi.Name() == "." || fi.Name() == ".." {
230 } else if child := path + "/" + fi.Name(); fi.IsDir() {
235 } else if strings.HasSuffix(child, ".gob") || strings.HasSuffix(child, ".gob.gz") {
236 files = append(files, child)
241 log.Infof("LoadDir: walk dir %s", path)
246 ctx, cancel := context.WithCancel(ctx)
249 cgs := []CompactGenome{}
250 cseqs := []CompactSequence{}
251 variantmap := map[tileLibRef]tileVariantID{}
252 errs := make(chan error, len(files))
253 log.Infof("LoadDir: read %d files", len(files))
254 for _, path := range files {
263 defer log.Infof("LoadDir: finished reading %s", path)
264 errs <- DecodeLibrary(f, strings.HasSuffix(path, ".gz"), func(ent *LibraryEntry) error {
265 if ctx.Err() != nil {
268 if len(ent.TagSet) > 0 {
270 if tilelib.taglib == nil || tilelib.taglib.Len() != len(ent.TagSet) {
271 // load first set of tags, or
272 // report mismatch if 2 sets
273 // have different #tags.
274 if err := tilelib.loadTagSet(ent.TagSet); err != nil {
281 variantmapadd := map[tileLibRef]tileVariantID{}
282 for _, tv := range ent.TileVariants {
283 variantmapadd[tileLibRef{Tag: tv.Tag, Variant: tv.Variant}] = tilelib.getRef(tv.Tag, tv.Sequence).Variant
286 cgs = append(cgs, ent.CompactGenomes...)
287 cseqs = append(cseqs, ent.CompactSequences...)
288 for k, v := range variantmapadd {
302 log.Info("LoadDir: loadCompactGenomes")
303 err = tilelib.loadCompactGenomes(cgs, variantmap, onLoadGenome)
307 log.Info("LoadDir: loadCompactSequences")
308 err = tilelib.loadCompactSequences(cseqs, variantmap)
312 log.Info("LoadDir done")
316 func (tilelib *tileLibrary) WriteDir(dir string) error {
318 files := make([]*os.File, nfiles)
319 for i := range files {
320 f, err := os.OpenFile(fmt.Sprintf("%s/library.%04d.gob.gz", dir, i), os.O_CREATE|os.O_WRONLY, 0666)
327 bufws := make([]*bufio.Writer, nfiles)
328 for i := range bufws {
329 bufws[i] = bufio.NewWriterSize(files[i], 1<<26)
331 zws := make([]*pgzip.Writer, nfiles)
333 zws[i] = pgzip.NewWriter(bufws[i])
336 encoders := make([]*gob.Encoder, nfiles)
337 for i := range encoders {
338 encoders[i] = gob.NewEncoder(zws[i])
341 cgnames := make([]string, 0, len(tilelib.compactGenomes))
342 for name := range tilelib.compactGenomes {
343 cgnames = append(cgnames, name)
345 sort.Strings(cgnames)
347 log.Infof("WriteDir: writing %d files", nfiles)
348 ctx, cancel := context.WithCancel(context.Background())
350 errs := make(chan error, nfiles)
351 for start := range files {
354 err := encoders[start].Encode(LibraryEntry{TagSet: tilelib.taglib.Tags()})
360 // For now, just write all the refs to
362 for name, tseqs := range tilelib.refseqs {
363 err := encoders[start].Encode(LibraryEntry{CompactSequences: []CompactSequence{{
365 TileSequences: tseqs,
373 for i := start; i < len(cgnames); i += nfiles {
374 err := encoders[start].Encode(LibraryEntry{CompactGenomes: []CompactGenome{{
376 Variants: tilelib.compactGenomes[cgnames[i]],
383 tvs := []TileVariant{}
384 for tag := start; tag < len(tilelib.variant) && ctx.Err() == nil; tag += nfiles {
386 for idx, hash := range tilelib.variant[tag] {
387 tvs = append(tvs, TileVariant{
389 Variant: tileVariantID(idx + 1),
391 Sequence: tilelib.seq[hash],
394 err := encoders[start].Encode(LibraryEntry{TileVariants: tvs})
409 log.Info("WriteDir: flushing")
411 err := zws[i].Close()
415 err = bufws[i].Flush()
419 err = files[i].Close()
424 log.Info("WriteDir: done")
428 // Load library data from rdr. Tile variants might be renumbered in
429 // the process; in that case, genomes variants will be renumbered to
432 // If onLoadGenome is non-nil, call it on each CompactGenome entry.
433 func (tilelib *tileLibrary) LoadGob(ctx context.Context, rdr io.Reader, gz bool, onLoadGenome func(CompactGenome)) error {
434 cgs := []CompactGenome{}
435 cseqs := []CompactSequence{}
436 variantmap := map[tileLibRef]tileVariantID{}
437 err := DecodeLibrary(rdr, gz, func(ent *LibraryEntry) error {
438 if ctx.Err() != nil {
441 if err := tilelib.loadTagSet(ent.TagSet); err != nil {
444 if err := tilelib.loadTileVariants(ent.TileVariants, variantmap); err != nil {
447 cgs = append(cgs, ent.CompactGenomes...)
448 cseqs = append(cseqs, ent.CompactSequences...)
454 if ctx.Err() != nil {
457 err = tilelib.loadCompactGenomes(cgs, variantmap, onLoadGenome)
461 err = tilelib.loadCompactSequences(cseqs, variantmap)
468 func (tilelib *tileLibrary) dump(out io.Writer) {
469 printTV := func(tag int, variant tileVariantID) {
471 fmt.Fprintf(out, " -")
472 } else if tag >= len(tilelib.variant) {
473 fmt.Fprintf(out, " (!tag=%d)", tag)
474 } else if int(variant) > len(tilelib.variant[tag]) {
475 fmt.Fprintf(out, " (tag=%d,!variant=%d)", tag, variant)
477 fmt.Fprintf(out, " %x", tilelib.variant[tag][variant-1][:8])
480 for refname, refseqs := range tilelib.refseqs {
481 for seqname, seq := range refseqs {
482 fmt.Fprintf(out, "ref %s %s", refname, seqname)
483 for _, libref := range seq {
484 printTV(int(libref.Tag), libref.Variant)
486 fmt.Fprintf(out, "\n")
489 for name, cg := range tilelib.compactGenomes {
490 fmt.Fprintf(out, "cg %s", name)
491 for tag, variant := range cg {
492 printTV(tag/2, variant)
494 fmt.Fprintf(out, "\n")
498 type importStats struct {
504 DroppedOutOfOrderTiles int
507 func (tilelib *tileLibrary) TileFasta(filelabel string, rdr io.Reader, matchChromosome *regexp.Regexp) (tileSeq, []importStats, error) {
513 todo := make(chan jobT, 1)
514 scanner := bufio.NewScanner(rdr)
520 buf := scanner.Bytes()
521 if len(buf) > 0 && buf[0] == '>' {
522 todo <- jobT{seqlabel, append([]byte(nil), fasta...)}
523 seqlabel, fasta = strings.SplitN(string(buf[1:]), " ", 2)[0], fasta[:0]
524 log.Debugf("%s %s reading fasta", filelabel, seqlabel)
526 fasta = append(fasta, bytes.ToLower(buf)...)
529 todo <- jobT{seqlabel, fasta}
531 type foundtag struct {
535 found := make([]foundtag, 2000000)
536 path := make([]tileLibRef, 2000000)
539 skippedSequences := 0
540 taglen := tilelib.taglib.TagLen()
541 var stats []importStats
542 for job := range todo {
543 if len(job.fasta) == 0 {
545 } else if !matchChromosome.MatchString(job.label) {
549 log.Debugf("%s %s tiling", filelabel, job.label)
552 tilelib.taglib.FindAll(job.fasta, func(tagid tagID, pos, taglen int) {
553 found = append(found, foundtag{pos: pos, tagid: tagid})
555 totalFoundTags += len(found)
557 log.Warnf("%s %s no tags found", filelabel, job.label)
562 log.Infof("%s %s keeping longest increasing subsequence", filelabel, job.label)
563 keep := longestIncreasingSubsequence(len(found), func(i int) int { return int(found[i].tagid) })
564 for i, x := range keep {
567 skipped = len(found) - len(keep)
568 found = found[:len(keep)]
571 log.Infof("%s %s getting %d librefs", filelabel, job.label, len(found))
572 throttle := &throttle{Max: runtime.NumCPU()}
573 path = path[:len(found)]
575 for i, f := range found {
579 defer throttle.Release()
580 var startpos, endpos int
586 if i == len(found)-1 {
587 endpos = len(job.fasta)
589 endpos = found[i+1].pos + taglen
591 path[i] = tilelib.getRef(f.tagid, job.fasta[startpos:endpos])
592 if countBases(job.fasta[startpos:endpos]) != endpos-startpos {
593 atomic.AddInt64(&lowquality, 1)
599 log.Infof("%s %s copying path", filelabel, job.label)
601 pathcopy := make([]tileLibRef, len(path))
603 ret[job.label] = pathcopy
605 basesIn := countBases(job.fasta)
606 log.Infof("%s %s fasta in %d coverage in %d path len %d low-quality %d skipped-out-of-order %d", filelabel, job.label, len(job.fasta), basesIn, len(path), lowquality, skipped)
607 stats = append(stats, importStats{
608 InputFile: filelabel,
609 InputLabel: job.label,
610 InputLength: len(job.fasta),
611 InputCoverage: basesIn,
612 PathLength: len(path),
613 DroppedOutOfOrderTiles: skipped,
616 totalPathLen += len(path)
618 log.Printf("%s tiled with total path len %d in %d sequences (skipped %d sequences that did not match chromosome regexp, skipped %d out-of-order tags)", filelabel, totalPathLen, len(ret), skippedSequences, totalFoundTags-totalPathLen)
619 return ret, stats, scanner.Err()
622 func (tilelib *tileLibrary) Len() int64 {
623 return atomic.LoadInt64(&tilelib.variants)
626 // Return a tileLibRef for a tile with the given tag and sequence,
627 // adding the sequence to the library if needed.
628 func (tilelib *tileLibrary) getRef(tag tagID, seq []byte) tileLibRef {
630 if !tilelib.retainNoCalls {
631 for _, b := range seq {
632 if b != 'a' && b != 'c' && b != 'g' && b != 't' {
638 seqhash := blake2b.Sum256(seq)
639 var vlock sync.Locker
642 if len(tilelib.vlock) > int(tag) {
643 vlock = tilelib.vlock[tag]
645 tilelib.mtx.RUnlock()
649 for i, varhash := range tilelib.variant[tag] {
650 if varhash == seqhash {
652 return tileLibRef{Tag: tag, Variant: tileVariantID(i + 1)}
658 if tilelib.variant == nil && tilelib.taglib != nil {
659 tilelib.variant = make([][][blake2b.Size256]byte, tilelib.taglib.Len())
660 tilelib.vlock = make([]sync.Locker, tilelib.taglib.Len())
661 for i := range tilelib.vlock {
662 tilelib.vlock[i] = new(sync.Mutex)
665 if int(tag) >= len(tilelib.variant) {
666 oldlen := len(tilelib.vlock)
667 for i := 0; i < oldlen; i++ {
668 tilelib.vlock[i].Lock()
670 // If we haven't seen the tag library yet (as
671 // in a merge), tilelib.taglib.Len() is
672 // zero. We can still behave correctly, we
673 // just need to expand the tilelib.variant and
674 // tilelib.vlock slices as needed.
675 if int(tag) >= cap(tilelib.variant) {
676 // Allocate 2x capacity.
677 newslice := make([][][blake2b.Size256]byte, int(tag)+1, (int(tag)+1)*2)
678 copy(newslice, tilelib.variant)
679 tilelib.variant = newslice[:int(tag)+1]
680 newvlock := make([]sync.Locker, int(tag)+1, (int(tag)+1)*2)
681 copy(newvlock, tilelib.vlock)
682 tilelib.vlock = newvlock[:int(tag)+1]
684 // Use previously allocated capacity,
686 tilelib.variant = tilelib.variant[:int(tag)+1]
687 tilelib.vlock = tilelib.vlock[:int(tag)+1]
689 for i := oldlen; i < len(tilelib.vlock); i++ {
690 tilelib.vlock[i] = new(sync.Mutex)
692 for i := 0; i < oldlen; i++ {
693 tilelib.vlock[i].Unlock()
696 vlock = tilelib.vlock[tag]
701 for i, varhash := range tilelib.variant[tag] {
702 if varhash == seqhash {
704 return tileLibRef{Tag: tag, Variant: tileVariantID(i + 1)}
707 atomic.AddInt64(&tilelib.variants, 1)
708 tilelib.variant[tag] = append(tilelib.variant[tag], seqhash)
709 variant := tileVariantID(len(tilelib.variant[tag]))
712 if tilelib.retainTileSequences && !dropSeq {
714 if tilelib.seq == nil {
715 tilelib.seq = map[[blake2b.Size256]byte][]byte{}
717 tilelib.seq[seqhash] = append([]byte(nil), seq...)
721 if tilelib.encoder != nil {
724 // Save the hash, but not the sequence
727 tilelib.encoder.Encode(LibraryEntry{
728 TileVariants: []TileVariant{{
736 return tileLibRef{Tag: tag, Variant: variant}
739 func (tilelib *tileLibrary) TileVariantSequence(libref tileLibRef) []byte {
740 if libref.Variant == 0 || len(tilelib.variant) <= int(libref.Tag) || len(tilelib.variant[libref.Tag]) < int(libref.Variant) {
743 return tilelib.seq[tilelib.variant[libref.Tag][libref.Variant-1]]
746 // Tidy deletes unreferenced tile variants and renumbers variants so
747 // more common variants have smaller IDs.
748 func (tilelib *tileLibrary) Tidy() {
749 log.Print("Tidy: compute inref")
750 inref := map[tileLibRef]bool{}
751 for _, refseq := range tilelib.refseqs {
752 for _, librefs := range refseq {
753 for _, libref := range librefs {
758 log.Print("Tidy: compute remap")
759 remap := make([][]tileVariantID, len(tilelib.variant))
760 throttle := throttle{Max: runtime.NumCPU() + 1}
761 for tag, oldvariants := range tilelib.variant {
762 tag, oldvariants := tagID(tag), oldvariants
763 if tag%1000000 == 0 {
764 log.Printf("Tidy: tag %d", tag)
768 defer throttle.Release()
769 uses := make([]int, len(oldvariants))
770 for _, cg := range tilelib.compactGenomes {
771 for phase := 0; phase < 2; phase++ {
772 cgi := int(tag)*2 + phase
773 if cgi < len(cg) && cg[cgi] > 0 {
779 // Compute desired order of variants:
780 // neworder[x] == index in oldvariants that
781 // should move to position x.
782 neworder := make([]int, len(oldvariants))
783 for i := range neworder {
786 sort.Slice(neworder, func(i, j int) bool {
787 if cmp := uses[neworder[i]] - uses[neworder[j]]; cmp != 0 {
790 return bytes.Compare(oldvariants[neworder[i]][:], oldvariants[neworder[j]][:]) < 0
794 // Replace tilelib.variant[tag] with a new
795 // re-ordered slice of hashes, and make a
796 // mapping from old to new variant IDs.
797 remaptag := make([]tileVariantID, len(oldvariants)+1)
798 newvariants := make([][blake2b.Size256]byte, 0, len(neworder))
799 for _, oldi := range neworder {
800 if uses[oldi] > 0 || inref[tileLibRef{Tag: tag, Variant: tileVariantID(oldi + 1)}] {
801 newvariants = append(newvariants, oldvariants[oldi])
802 remaptag[oldi+1] = tileVariantID(len(newvariants))
805 tilelib.variant[tag] = newvariants
806 remap[tag] = remaptag
811 // Apply remap to genomes and reference sequences, so they
812 // refer to the same tile variants using the changed IDs.
813 log.Print("Tidy: apply remap")
814 var wg sync.WaitGroup
815 for _, cg := range tilelib.compactGenomes {
820 for idx, variant := range cg {
821 cg[idx] = remap[tagID(idx/2)][variant]
825 for _, refcs := range tilelib.refseqs {
826 for _, refseq := range refcs {
831 for i, tv := range refseq {
832 refseq[i].Variant = remap[tv.Tag][tv.Variant]
838 log.Print("Tidy: done")
841 func countBases(seq []byte) int {
843 for _, c := range seq {