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 fi.IsDir() {
231 err = walk(path + "/" + fi.Name())
235 } else if strings.HasSuffix(path, ".gob") || strings.HasSuffix(path, ".gob.gz") {
236 files = append(files, path)
245 ctx, cancel := context.WithCancel(ctx)
248 cgs := []CompactGenome{}
249 cseqs := []CompactSequence{}
250 variantmap := map[tileLibRef]tileVariantID{}
251 errs := make(chan error, len(files))
252 for _, path := range files {
261 errs <- DecodeLibrary(f, strings.HasSuffix(path, ".gz"), func(ent *LibraryEntry) error {
262 if ctx.Err() != nil {
267 if err := tilelib.loadTagSet(ent.TagSet); err != nil {
270 if err := tilelib.loadTileVariants(ent.TileVariants, variantmap); err != nil {
273 cgs = append(cgs, ent.CompactGenomes...)
274 cseqs = append(cseqs, ent.CompactSequences...)
285 err = tilelib.loadCompactGenomes(cgs, variantmap, onLoadGenome)
289 err = tilelib.loadCompactSequences(cseqs, variantmap)
296 func (tilelib *tileLibrary) WriteDir(dir string) error {
298 files := make([]*os.File, nfiles)
299 for i := range files {
300 f, err := os.OpenFile(fmt.Sprintf("%s/library.%04d.gob.gz", dir, i), os.O_CREATE|os.O_WRONLY, 0666)
307 bufws := make([]*bufio.Writer, nfiles)
308 for i := range bufws {
309 bufws[i] = bufio.NewWriterSize(files[i], 1<<26)
311 zws := make([]*pgzip.Writer, nfiles)
313 zws[i] = pgzip.NewWriter(bufws[i])
316 encoders := make([]*gob.Encoder, nfiles)
317 for i := range encoders {
318 encoders[i] = gob.NewEncoder(zws[i])
320 ctx, cancel := context.WithCancel(context.Background())
322 errs := make(chan error, nfiles)
323 for start := range files {
326 err := encoders[start].Encode(LibraryEntry{TagSet: tilelib.taglib.Tags()})
332 // For now, just write all the genomes and refs
334 for name, cg := range tilelib.compactGenomes {
335 err := encoders[start].Encode(LibraryEntry{CompactGenomes: []CompactGenome{{
344 for name, tseqs := range tilelib.refseqs {
345 err := encoders[start].Encode(LibraryEntry{CompactSequences: []CompactSequence{{
347 TileSequences: tseqs,
355 tvs := []TileVariant{}
356 for tag := start; tag < len(tilelib.variant) && ctx.Err() == nil; tag += nfiles {
358 for idx, hash := range tilelib.variant[tag] {
359 tvs = append(tvs, TileVariant{
361 Variant: tileVariantID(idx + 1),
363 Sequence: tilelib.seq[hash],
366 err := encoders[start].Encode(LibraryEntry{TileVariants: tvs})
382 err := zws[i].Close()
386 err = bufws[i].Flush()
390 err = files[i].Close()
398 // Load library data from rdr. Tile variants might be renumbered in
399 // the process; in that case, genomes variants will be renumbered to
402 // If onLoadGenome is non-nil, call it on each CompactGenome entry.
403 func (tilelib *tileLibrary) LoadGob(ctx context.Context, rdr io.Reader, gz bool, onLoadGenome func(CompactGenome)) error {
404 cgs := []CompactGenome{}
405 cseqs := []CompactSequence{}
406 variantmap := map[tileLibRef]tileVariantID{}
407 err := DecodeLibrary(rdr, gz, func(ent *LibraryEntry) error {
408 if ctx.Err() != nil {
411 if err := tilelib.loadTagSet(ent.TagSet); err != nil {
414 if err := tilelib.loadTileVariants(ent.TileVariants, variantmap); err != nil {
417 cgs = append(cgs, ent.CompactGenomes...)
418 cseqs = append(cseqs, ent.CompactSequences...)
424 if ctx.Err() != nil {
427 err = tilelib.loadCompactGenomes(cgs, variantmap, onLoadGenome)
431 err = tilelib.loadCompactSequences(cseqs, variantmap)
438 func (tilelib *tileLibrary) dump(out io.Writer) {
439 printTV := func(tag int, variant tileVariantID) {
441 fmt.Fprintf(out, " -")
442 } else if tag >= len(tilelib.variant) {
443 fmt.Fprintf(out, " (!tag=%d)", tag)
444 } else if int(variant) > len(tilelib.variant[tag]) {
445 fmt.Fprintf(out, " (tag=%d,!variant=%d)", tag, variant)
447 fmt.Fprintf(out, " %x", tilelib.variant[tag][variant-1][:8])
450 for refname, refseqs := range tilelib.refseqs {
451 for seqname, seq := range refseqs {
452 fmt.Fprintf(out, "ref %s %s", refname, seqname)
453 for _, libref := range seq {
454 printTV(int(libref.Tag), libref.Variant)
456 fmt.Fprintf(out, "\n")
459 for name, cg := range tilelib.compactGenomes {
460 fmt.Fprintf(out, "cg %s", name)
461 for tag, variant := range cg {
462 printTV(tag/2, variant)
464 fmt.Fprintf(out, "\n")
468 type importStats struct {
474 DroppedOutOfOrderTiles int
477 func (tilelib *tileLibrary) TileFasta(filelabel string, rdr io.Reader, matchChromosome *regexp.Regexp) (tileSeq, []importStats, error) {
483 todo := make(chan jobT, 1)
484 scanner := bufio.NewScanner(rdr)
490 buf := scanner.Bytes()
491 if len(buf) > 0 && buf[0] == '>' {
492 todo <- jobT{seqlabel, append([]byte(nil), fasta...)}
493 seqlabel, fasta = strings.SplitN(string(buf[1:]), " ", 2)[0], fasta[:0]
494 log.Debugf("%s %s reading fasta", filelabel, seqlabel)
496 fasta = append(fasta, bytes.ToLower(buf)...)
499 todo <- jobT{seqlabel, fasta}
501 type foundtag struct {
505 found := make([]foundtag, 2000000)
506 path := make([]tileLibRef, 2000000)
509 skippedSequences := 0
510 taglen := tilelib.taglib.TagLen()
511 var stats []importStats
512 for job := range todo {
513 if len(job.fasta) == 0 {
515 } else if !matchChromosome.MatchString(job.label) {
519 log.Debugf("%s %s tiling", filelabel, job.label)
522 tilelib.taglib.FindAll(job.fasta, func(tagid tagID, pos, taglen int) {
523 found = append(found, foundtag{pos: pos, tagid: tagid})
525 totalFoundTags += len(found)
527 log.Warnf("%s %s no tags found", filelabel, job.label)
532 log.Infof("%s %s keeping longest increasing subsequence", filelabel, job.label)
533 keep := longestIncreasingSubsequence(len(found), func(i int) int { return int(found[i].tagid) })
534 for i, x := range keep {
537 skipped = len(found) - len(keep)
538 found = found[:len(keep)]
541 log.Infof("%s %s getting %d librefs", filelabel, job.label, len(found))
542 throttle := &throttle{Max: runtime.NumCPU()}
543 path = path[:len(found)]
545 for i, f := range found {
549 defer throttle.Release()
550 var startpos, endpos int
556 if i == len(found)-1 {
557 endpos = len(job.fasta)
559 endpos = found[i+1].pos + taglen
561 path[i] = tilelib.getRef(f.tagid, job.fasta[startpos:endpos])
562 if countBases(job.fasta[startpos:endpos]) != endpos-startpos {
563 atomic.AddInt64(&lowquality, 1)
569 log.Infof("%s %s copying path", filelabel, job.label)
571 pathcopy := make([]tileLibRef, len(path))
573 ret[job.label] = pathcopy
575 basesIn := countBases(job.fasta)
576 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)
577 stats = append(stats, importStats{
578 InputFile: filelabel,
579 InputLabel: job.label,
580 InputLength: len(job.fasta),
581 InputCoverage: basesIn,
582 PathLength: len(path),
583 DroppedOutOfOrderTiles: skipped,
586 totalPathLen += len(path)
588 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)
589 return ret, stats, scanner.Err()
592 func (tilelib *tileLibrary) Len() int64 {
593 return atomic.LoadInt64(&tilelib.variants)
596 // Return a tileLibRef for a tile with the given tag and sequence,
597 // adding the sequence to the library if needed.
598 func (tilelib *tileLibrary) getRef(tag tagID, seq []byte) tileLibRef {
600 if !tilelib.retainNoCalls {
601 for _, b := range seq {
602 if b != 'a' && b != 'c' && b != 'g' && b != 't' {
608 seqhash := blake2b.Sum256(seq)
609 var vlock sync.Locker
612 if len(tilelib.vlock) > int(tag) {
613 vlock = tilelib.vlock[tag]
615 tilelib.mtx.RUnlock()
619 for i, varhash := range tilelib.variant[tag] {
620 if varhash == seqhash {
622 return tileLibRef{Tag: tag, Variant: tileVariantID(i + 1)}
628 if tilelib.variant == nil && tilelib.taglib != nil {
629 tilelib.variant = make([][][blake2b.Size256]byte, tilelib.taglib.Len())
630 tilelib.vlock = make([]sync.Locker, tilelib.taglib.Len())
631 for i := range tilelib.vlock {
632 tilelib.vlock[i] = new(sync.Mutex)
635 if int(tag) >= len(tilelib.variant) {
636 oldlen := len(tilelib.vlock)
637 for i := 0; i < oldlen; i++ {
638 tilelib.vlock[i].Lock()
640 // If we haven't seen the tag library yet (as
641 // in a merge), tilelib.taglib.Len() is
642 // zero. We can still behave correctly, we
643 // just need to expand the tilelib.variant and
644 // tilelib.vlock slices as needed.
645 if int(tag) >= cap(tilelib.variant) {
646 // Allocate 2x capacity.
647 newslice := make([][][blake2b.Size256]byte, int(tag)+1, (int(tag)+1)*2)
648 copy(newslice, tilelib.variant)
649 tilelib.variant = newslice[:int(tag)+1]
650 newvlock := make([]sync.Locker, int(tag)+1, (int(tag)+1)*2)
651 copy(newvlock, tilelib.vlock)
652 tilelib.vlock = newvlock[:int(tag)+1]
654 // Use previously allocated capacity,
656 tilelib.variant = tilelib.variant[:int(tag)+1]
657 tilelib.vlock = tilelib.vlock[:int(tag)+1]
659 for i := oldlen; i < len(tilelib.vlock); i++ {
660 tilelib.vlock[i] = new(sync.Mutex)
662 for i := 0; i < oldlen; i++ {
663 tilelib.vlock[i].Unlock()
666 vlock = tilelib.vlock[tag]
671 for i, varhash := range tilelib.variant[tag] {
672 if varhash == seqhash {
674 return tileLibRef{Tag: tag, Variant: tileVariantID(i + 1)}
677 atomic.AddInt64(&tilelib.variants, 1)
678 tilelib.variant[tag] = append(tilelib.variant[tag], seqhash)
679 variant := tileVariantID(len(tilelib.variant[tag]))
682 if tilelib.retainTileSequences && !dropSeq {
684 if tilelib.seq == nil {
685 tilelib.seq = map[[blake2b.Size256]byte][]byte{}
687 tilelib.seq[seqhash] = append([]byte(nil), seq...)
691 if tilelib.encoder != nil {
694 // Save the hash, but not the sequence
697 tilelib.encoder.Encode(LibraryEntry{
698 TileVariants: []TileVariant{{
706 return tileLibRef{Tag: tag, Variant: variant}
709 func (tilelib *tileLibrary) TileVariantSequence(libref tileLibRef) []byte {
710 if libref.Variant == 0 || len(tilelib.variant) <= int(libref.Tag) || len(tilelib.variant[libref.Tag]) < int(libref.Variant) {
713 return tilelib.seq[tilelib.variant[libref.Tag][libref.Variant-1]]
716 // Tidy deletes unreferenced tile variants and renumbers variants so
717 // more common variants have smaller IDs.
718 func (tilelib *tileLibrary) Tidy() {
719 log.Print("Tidy: compute inref")
720 inref := map[tileLibRef]bool{}
721 for _, refseq := range tilelib.refseqs {
722 for _, librefs := range refseq {
723 for _, libref := range librefs {
728 log.Print("Tidy: compute remap")
729 remap := make([][]tileVariantID, len(tilelib.variant))
730 throttle := throttle{Max: runtime.NumCPU() + 1}
731 for tag, oldvariants := range tilelib.variant {
732 tag, oldvariants := tagID(tag), oldvariants
733 if tag%1000000 == 0 {
734 log.Printf("Tidy: tag %d", tag)
738 defer throttle.Release()
739 uses := make([]int, len(oldvariants))
740 for _, cg := range tilelib.compactGenomes {
741 for phase := 0; phase < 2; phase++ {
742 cgi := int(tag)*2 + phase
743 if cgi < len(cg) && cg[cgi] > 0 {
749 // Compute desired order of variants:
750 // neworder[x] == index in oldvariants that
751 // should move to position x.
752 neworder := make([]int, len(oldvariants))
753 for i := range neworder {
756 sort.Slice(neworder, func(i, j int) bool {
757 if cmp := uses[neworder[i]] - uses[neworder[j]]; cmp != 0 {
760 return bytes.Compare(oldvariants[neworder[i]][:], oldvariants[neworder[j]][:]) < 0
764 // Replace tilelib.variant[tag] with a new
765 // re-ordered slice of hashes, and make a
766 // mapping from old to new variant IDs.
767 remaptag := make([]tileVariantID, len(oldvariants)+1)
768 newvariants := make([][blake2b.Size256]byte, 0, len(neworder))
769 for _, oldi := range neworder {
770 if uses[oldi] > 0 || inref[tileLibRef{Tag: tag, Variant: tileVariantID(oldi + 1)}] {
771 newvariants = append(newvariants, oldvariants[oldi])
772 remaptag[oldi+1] = tileVariantID(len(newvariants))
775 tilelib.variant[tag] = newvariants
776 remap[tag] = remaptag
781 // Apply remap to genomes and reference sequences, so they
782 // refer to the same tile variants using the changed IDs.
783 log.Print("Tidy: apply remap")
784 var wg sync.WaitGroup
785 for _, cg := range tilelib.compactGenomes {
790 for idx, variant := range cg {
791 cg[idx] = remap[tagID(idx/2)][variant]
795 for _, refcs := range tilelib.refseqs {
796 for _, refseq := range refcs {
801 for i, tv := range refseq {
802 refseq[i].Variant = remap[tv.Tag][tv.Variant]
808 log.Print("Tidy: done")
811 func countBases(seq []byte) int {
813 for _, c := range seq {