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
65 seq2 map[[2]byte]map[[blake2b.Size256]byte][]byte
66 seq2lock map[[2]byte]sync.Locker
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 allcgs := make([][]CompactGenome, len(files))
250 allcseqs := make([][]CompactSequence, len(files))
251 allvariantmap := map[tileLibRef]tileVariantID{}
252 errs := make(chan error, len(files))
253 log.Infof("LoadDir: read %d files", len(files))
254 for fileno, path := range files {
255 fileno, path := fileno, path
263 defer log.Infof("LoadDir: finished reading %s", path)
265 var variantmap = map[tileLibRef]tileVariantID{}
266 var cgs []CompactGenome
267 var cseqs []CompactSequence
268 errs <- DecodeLibrary(f, strings.HasSuffix(path, ".gz"), func(ent *LibraryEntry) error {
269 if ctx.Err() != nil {
272 if len(ent.TagSet) > 0 {
274 if tilelib.taglib == nil || tilelib.taglib.Len() != len(ent.TagSet) {
275 // load first set of tags, or
276 // report mismatch if 2 sets
277 // have different #tags.
278 if err := tilelib.loadTagSet(ent.TagSet); err != nil {
285 for _, tv := range ent.TileVariants {
286 variantmap[tileLibRef{Tag: tv.Tag, Variant: tv.Variant}] = tilelib.getRef(tv.Tag, tv.Sequence).Variant
288 cgs = append(cgs, ent.CompactGenomes...)
289 cseqs = append(cseqs, ent.CompactSequences...)
293 allcseqs[fileno] = cseqs
296 for k, v := range variantmap {
308 log.Info("LoadDir: loadCompactGenomes")
309 var flatcgs []CompactGenome
310 for _, cgs := range allcgs {
311 flatcgs = append(flatcgs, cgs...)
313 err = tilelib.loadCompactGenomes(flatcgs, allvariantmap, onLoadGenome)
318 log.Info("LoadDir: loadCompactSequences")
319 var flatcseqs []CompactSequence
320 for _, cseqs := range allcseqs {
321 flatcseqs = append(flatcseqs, cseqs...)
323 err = tilelib.loadCompactSequences(flatcseqs, allvariantmap)
328 log.Info("LoadDir done")
332 func (tilelib *tileLibrary) WriteDir(dir string) error {
334 files := make([]*os.File, nfiles)
335 for i := range files {
336 f, err := os.OpenFile(fmt.Sprintf("%s/library.%04d.gob.gz", dir, i), os.O_CREATE|os.O_WRONLY, 0666)
343 bufws := make([]*bufio.Writer, nfiles)
344 for i := range bufws {
345 bufws[i] = bufio.NewWriterSize(files[i], 1<<26)
347 zws := make([]*pgzip.Writer, nfiles)
349 zws[i] = pgzip.NewWriter(bufws[i])
352 encoders := make([]*gob.Encoder, nfiles)
353 for i := range encoders {
354 encoders[i] = gob.NewEncoder(zws[i])
357 cgnames := make([]string, 0, len(tilelib.compactGenomes))
358 for name := range tilelib.compactGenomes {
359 cgnames = append(cgnames, name)
361 sort.Strings(cgnames)
363 log.Infof("WriteDir: writing %d files", nfiles)
364 ctx, cancel := context.WithCancel(context.Background())
366 errs := make(chan error, nfiles)
367 for start := range files {
370 err := encoders[start].Encode(LibraryEntry{TagSet: tilelib.taglib.Tags()})
376 // For now, just write all the refs to
378 for name, tseqs := range tilelib.refseqs {
379 err := encoders[start].Encode(LibraryEntry{CompactSequences: []CompactSequence{{
381 TileSequences: tseqs,
389 for i := start; i < len(cgnames); i += nfiles {
390 err := encoders[start].Encode(LibraryEntry{CompactGenomes: []CompactGenome{{
392 Variants: tilelib.compactGenomes[cgnames[i]],
399 tvs := []TileVariant{}
400 for tag := start; tag < len(tilelib.variant) && ctx.Err() == nil; tag += nfiles {
402 for idx, hash := range tilelib.variant[tag] {
403 tvs = append(tvs, TileVariant{
405 Variant: tileVariantID(idx + 1),
407 Sequence: tilelib.hashSequence(hash),
410 err := encoders[start].Encode(LibraryEntry{TileVariants: tvs})
425 log.Info("WriteDir: flushing")
427 err := zws[i].Close()
431 err = bufws[i].Flush()
435 err = files[i].Close()
440 log.Info("WriteDir: done")
444 // Load library data from rdr. Tile variants might be renumbered in
445 // the process; in that case, genomes variants will be renumbered to
448 // If onLoadGenome is non-nil, call it on each CompactGenome entry.
449 func (tilelib *tileLibrary) LoadGob(ctx context.Context, rdr io.Reader, gz bool, onLoadGenome func(CompactGenome)) error {
450 cgs := []CompactGenome{}
451 cseqs := []CompactSequence{}
452 variantmap := map[tileLibRef]tileVariantID{}
453 err := DecodeLibrary(rdr, gz, func(ent *LibraryEntry) error {
454 if ctx.Err() != nil {
457 if err := tilelib.loadTagSet(ent.TagSet); err != nil {
460 if err := tilelib.loadTileVariants(ent.TileVariants, variantmap); err != nil {
463 cgs = append(cgs, ent.CompactGenomes...)
464 cseqs = append(cseqs, ent.CompactSequences...)
470 if ctx.Err() != nil {
473 err = tilelib.loadCompactGenomes(cgs, variantmap, onLoadGenome)
477 err = tilelib.loadCompactSequences(cseqs, variantmap)
484 func (tilelib *tileLibrary) dump(out io.Writer) {
485 printTV := func(tag int, variant tileVariantID) {
487 fmt.Fprintf(out, " -")
488 } else if tag >= len(tilelib.variant) {
489 fmt.Fprintf(out, " (!tag=%d)", tag)
490 } else if int(variant) > len(tilelib.variant[tag]) {
491 fmt.Fprintf(out, " (tag=%d,!variant=%d)", tag, variant)
493 fmt.Fprintf(out, " %x", tilelib.variant[tag][variant-1][:8])
496 for refname, refseqs := range tilelib.refseqs {
497 for seqname, seq := range refseqs {
498 fmt.Fprintf(out, "ref %s %s", refname, seqname)
499 for _, libref := range seq {
500 printTV(int(libref.Tag), libref.Variant)
502 fmt.Fprintf(out, "\n")
505 for name, cg := range tilelib.compactGenomes {
506 fmt.Fprintf(out, "cg %s", name)
507 for tag, variant := range cg {
508 printTV(tag/2, variant)
510 fmt.Fprintf(out, "\n")
514 type importStats struct {
520 DroppedOutOfOrderTiles int
523 func (tilelib *tileLibrary) TileFasta(filelabel string, rdr io.Reader, matchChromosome *regexp.Regexp) (tileSeq, []importStats, error) {
529 todo := make(chan jobT, 1)
530 scanner := bufio.NewScanner(rdr)
536 buf := scanner.Bytes()
537 if len(buf) > 0 && buf[0] == '>' {
538 todo <- jobT{seqlabel, append([]byte(nil), fasta...)}
539 seqlabel, fasta = strings.SplitN(string(buf[1:]), " ", 2)[0], fasta[:0]
540 log.Debugf("%s %s reading fasta", filelabel, seqlabel)
542 fasta = append(fasta, bytes.ToLower(buf)...)
545 todo <- jobT{seqlabel, fasta}
547 type foundtag struct {
551 found := make([]foundtag, 2000000)
552 path := make([]tileLibRef, 2000000)
555 skippedSequences := 0
556 taglen := tilelib.taglib.TagLen()
557 var stats []importStats
558 for job := range todo {
559 if len(job.fasta) == 0 {
561 } else if !matchChromosome.MatchString(job.label) {
565 log.Debugf("%s %s tiling", filelabel, job.label)
568 tilelib.taglib.FindAll(job.fasta, func(tagid tagID, pos, taglen int) {
569 found = append(found, foundtag{pos: pos, tagid: tagid})
571 totalFoundTags += len(found)
573 log.Warnf("%s %s no tags found", filelabel, job.label)
578 log.Infof("%s %s keeping longest increasing subsequence", filelabel, job.label)
579 keep := longestIncreasingSubsequence(len(found), func(i int) int { return int(found[i].tagid) })
580 for i, x := range keep {
583 skipped = len(found) - len(keep)
584 found = found[:len(keep)]
587 log.Infof("%s %s getting %d librefs", filelabel, job.label, len(found))
588 throttle := &throttle{Max: runtime.NumCPU()}
589 path = path[:len(found)]
591 for i, f := range found {
595 defer throttle.Release()
596 var startpos, endpos int
602 if i == len(found)-1 {
603 endpos = len(job.fasta)
605 endpos = found[i+1].pos + taglen
607 path[i] = tilelib.getRef(f.tagid, job.fasta[startpos:endpos])
608 if countBases(job.fasta[startpos:endpos]) != endpos-startpos {
609 atomic.AddInt64(&lowquality, 1)
615 log.Infof("%s %s copying path", filelabel, job.label)
617 pathcopy := make([]tileLibRef, len(path))
619 ret[job.label] = pathcopy
621 basesIn := countBases(job.fasta)
622 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)
623 stats = append(stats, importStats{
624 InputFile: filelabel,
625 InputLabel: job.label,
626 InputLength: len(job.fasta),
627 InputCoverage: basesIn,
628 PathLength: len(path),
629 DroppedOutOfOrderTiles: skipped,
632 totalPathLen += len(path)
634 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)
635 return ret, stats, scanner.Err()
638 func (tilelib *tileLibrary) Len() int64 {
639 return atomic.LoadInt64(&tilelib.variants)
642 // Return a tileLibRef for a tile with the given tag and sequence,
643 // adding the sequence to the library if needed.
644 func (tilelib *tileLibrary) getRef(tag tagID, seq []byte) tileLibRef {
646 if !tilelib.retainNoCalls {
647 for _, b := range seq {
648 if b != 'a' && b != 'c' && b != 'g' && b != 't' {
654 seqhash := blake2b.Sum256(seq)
655 var vlock sync.Locker
658 if len(tilelib.vlock) > int(tag) {
659 vlock = tilelib.vlock[tag]
661 tilelib.mtx.RUnlock()
665 for i, varhash := range tilelib.variant[tag] {
666 if varhash == seqhash {
668 return tileLibRef{Tag: tag, Variant: tileVariantID(i + 1)}
674 if tilelib.variant == nil && tilelib.taglib != nil {
675 tilelib.variant = make([][][blake2b.Size256]byte, tilelib.taglib.Len())
676 tilelib.vlock = make([]sync.Locker, tilelib.taglib.Len())
677 for i := range tilelib.vlock {
678 tilelib.vlock[i] = new(sync.Mutex)
681 if int(tag) >= len(tilelib.variant) {
682 oldlen := len(tilelib.vlock)
683 for i := 0; i < oldlen; i++ {
684 tilelib.vlock[i].Lock()
686 // If we haven't seen the tag library yet (as
687 // in a merge), tilelib.taglib.Len() is
688 // zero. We can still behave correctly, we
689 // just need to expand the tilelib.variant and
690 // tilelib.vlock slices as needed.
691 if int(tag) >= cap(tilelib.variant) {
692 // Allocate 2x capacity.
693 newslice := make([][][blake2b.Size256]byte, int(tag)+1, (int(tag)+1)*2)
694 copy(newslice, tilelib.variant)
695 tilelib.variant = newslice[:int(tag)+1]
696 newvlock := make([]sync.Locker, int(tag)+1, (int(tag)+1)*2)
697 copy(newvlock, tilelib.vlock)
698 tilelib.vlock = newvlock[:int(tag)+1]
700 // Use previously allocated capacity,
702 tilelib.variant = tilelib.variant[:int(tag)+1]
703 tilelib.vlock = tilelib.vlock[:int(tag)+1]
705 for i := oldlen; i < len(tilelib.vlock); i++ {
706 tilelib.vlock[i] = new(sync.Mutex)
708 for i := 0; i < oldlen; i++ {
709 tilelib.vlock[i].Unlock()
712 vlock = tilelib.vlock[tag]
717 for i, varhash := range tilelib.variant[tag] {
718 if varhash == seqhash {
720 return tileLibRef{Tag: tag, Variant: tileVariantID(i + 1)}
723 atomic.AddInt64(&tilelib.variants, 1)
724 tilelib.variant[tag] = append(tilelib.variant[tag], seqhash)
725 variant := tileVariantID(len(tilelib.variant[tag]))
728 if tilelib.retainTileSequences && !dropSeq {
729 seqCopy := append([]byte(nil), seq...)
730 if tilelib.seq2 == nil {
732 if tilelib.seq2 == nil {
733 tilelib.seq2lock = map[[2]byte]sync.Locker{}
734 m := map[[2]byte]map[[blake2b.Size256]byte][]byte{}
736 for i := 0; i < 256; i++ {
738 for j := 0; j < 256; j++ {
740 m[k] = map[[blake2b.Size256]byte][]byte{}
741 tilelib.seq2lock[k] = &sync.Mutex{}
749 copy(k[:], seqhash[:])
750 locker := tilelib.seq2lock[k]
752 tilelib.seq2[k][seqhash] = seqCopy
756 if tilelib.encoder != nil {
759 // Save the hash, but not the sequence
762 tilelib.encoder.Encode(LibraryEntry{
763 TileVariants: []TileVariant{{
771 return tileLibRef{Tag: tag, Variant: variant}
774 func (tilelib *tileLibrary) hashSequence(hash [blake2b.Size256]byte) []byte {
775 var partition [2]byte
776 copy(partition[:], hash[:])
777 return tilelib.seq2[partition][hash]
780 func (tilelib *tileLibrary) TileVariantSequence(libref tileLibRef) []byte {
781 if libref.Variant == 0 || len(tilelib.variant) <= int(libref.Tag) || len(tilelib.variant[libref.Tag]) < int(libref.Variant) {
784 return tilelib.hashSequence(tilelib.variant[libref.Tag][libref.Variant-1])
787 // Tidy deletes unreferenced tile variants and renumbers variants so
788 // more common variants have smaller IDs.
789 func (tilelib *tileLibrary) Tidy() {
790 log.Print("Tidy: compute inref")
791 inref := map[tileLibRef]bool{}
792 for _, refseq := range tilelib.refseqs {
793 for _, librefs := range refseq {
794 for _, libref := range librefs {
799 log.Print("Tidy: compute remap")
800 remap := make([][]tileVariantID, len(tilelib.variant))
801 throttle := throttle{Max: runtime.NumCPU() + 1}
802 for tag, oldvariants := range tilelib.variant {
803 tag, oldvariants := tagID(tag), oldvariants
804 if tag%1000000 == 0 {
805 log.Printf("Tidy: tag %d", tag)
809 defer throttle.Release()
810 uses := make([]int, len(oldvariants))
811 for _, cg := range tilelib.compactGenomes {
812 for phase := 0; phase < 2; phase++ {
813 cgi := int(tag)*2 + phase
814 if cgi < len(cg) && cg[cgi] > 0 {
820 // Compute desired order of variants:
821 // neworder[x] == index in oldvariants that
822 // should move to position x.
823 neworder := make([]int, len(oldvariants))
824 for i := range neworder {
827 sort.Slice(neworder, func(i, j int) bool {
828 if cmp := uses[neworder[i]] - uses[neworder[j]]; cmp != 0 {
831 return bytes.Compare(oldvariants[neworder[i]][:], oldvariants[neworder[j]][:]) < 0
835 // Replace tilelib.variant[tag] with a new
836 // re-ordered slice of hashes, and make a
837 // mapping from old to new variant IDs.
838 remaptag := make([]tileVariantID, len(oldvariants)+1)
839 newvariants := make([][blake2b.Size256]byte, 0, len(neworder))
840 for _, oldi := range neworder {
841 if uses[oldi] > 0 || inref[tileLibRef{Tag: tag, Variant: tileVariantID(oldi + 1)}] {
842 newvariants = append(newvariants, oldvariants[oldi])
843 remaptag[oldi+1] = tileVariantID(len(newvariants))
846 tilelib.variant[tag] = newvariants
847 remap[tag] = remaptag
852 // Apply remap to genomes and reference sequences, so they
853 // refer to the same tile variants using the changed IDs.
854 log.Print("Tidy: apply remap")
855 var wg sync.WaitGroup
856 for _, cg := range tilelib.compactGenomes {
861 for idx, variant := range cg {
862 cg[idx] = remap[tagID(idx/2)][variant]
866 for _, refcs := range tilelib.refseqs {
867 for _, refseq := range refcs {
872 for i, tv := range refseq {
873 refseq[i].Variant = remap[tv.Tag][tv.Variant]
879 log.Print("Tidy: done")
882 func countBases(seq []byte) int {
884 for _, c := range seq {