"net/http"
_ "net/http/pprof"
"os"
+ "regexp"
"runtime"
"sort"
+ "strconv"
"strings"
"sync/atomic"
"git.arvados.org/arvados.git/sdk/go/arvados"
"github.com/arvados/lightning/hgvs"
"github.com/kshedden/gonpy"
- "github.com/sirupsen/logrus"
log "github.com/sirupsen/logrus"
"golang.org/x/crypto/blake2b"
)
ref := flags.String("ref", "", "reference name (if blank, choose last one that appears in input)")
regionsFilename := flags.String("regions", "", "only output columns/annotations that intersect regions in specified bed `file`")
expandRegions := flags.Int("expand-regions", 0, "expand specified regions by `N` base pairs on each side`")
+ mergeOutput := flags.Bool("merge-output", false, "merge output into one matrix.npy and one matrix.annotations.csv")
flags.IntVar(&cmd.threads, "threads", 16, "number of memory-hungry assembly threads")
cmd.filter.Flags(flags)
err = flags.Parse(args)
Name: "lightning slice-numpy",
Client: arvados.NewClientFromEnv(),
ProjectUUID: *projectUUID,
- RAM: 240000000000,
- VCPUs: 64,
+ RAM: 650000000000,
+ VCPUs: 96,
Priority: *priority,
KeepCache: 2,
APIAccess: true,
return 1
}
runner.Args = []string{"slice-numpy", "-local=true",
- "-pprof", ":6060",
- "-input-dir", *inputDir,
- "-output-dir", "/mnt/output",
- "-threads", fmt.Sprintf("%d", cmd.threads),
- "-regions", *regionsFilename,
- "-expand-regions", fmt.Sprintf("%d", *expandRegions),
+ "-pprof=:6060",
+ "-input-dir=" + *inputDir,
+ "-output-dir=/mnt/output",
+ "-threads=" + fmt.Sprintf("%d", cmd.threads),
+ "-regions=" + *regionsFilename,
+ "-expand-regions=" + fmt.Sprintf("%d", *expandRegions),
+ "-merge-output=" + fmt.Sprintf("%v", *mergeOutput),
}
runner.Args = append(runner.Args, cmd.filter.Args()...)
var output string
return 1
}
+ matchGenome, err := regexp.Compile(cmd.filter.MatchGenome)
+ if err != nil {
+ err = fmt.Errorf("-match-genome: invalid regexp: %q", cmd.filter.MatchGenome)
+ return 1
+ }
+
taglen := -1
DecodeLibrary(in0, strings.HasSuffix(infiles[0], ".gz"), func(ent *LibraryEntry) error {
if len(ent.TagSet) > 0 {
}
}
for _, cg := range ent.CompactGenomes {
- cgnames = append(cgnames, cg.Name)
+ if matchGenome.MatchString(cg.Name) {
+ cgnames = append(cgnames, cg.Name)
+ }
}
for _, tv := range ent.TileVariants {
if tv.Ref {
err = fmt.Errorf("tagset not found")
return 1
}
+ if len(cgnames) == 0 {
+ err = fmt.Errorf("no genomes found matching regexp %q", cmd.filter.MatchGenome)
+ return 1
+ }
sort.Strings(cgnames)
{
}
throttleCPU.Wait()
- log.Info("TODO: determining which tiles intersect given regions")
+ var mask *mask
+ if *regionsFilename != "" {
+ mask, err = makeMask(*regionsFilename, *expandRegions)
+ if err != nil {
+ return 1
+ }
+ // Delete reftile entries for masked-out regions.
+ for tag, rt := range reftile {
+ if !mask.Check(strings.TrimPrefix(rt.seqname, "chr"), rt.pos, rt.pos+len(rt.tiledata)) {
+ delete(reftile, tag)
+ }
+ }
+ }
+
+ var toMerge [][]int16
+ if *mergeOutput {
+ toMerge = make([][]int16, len(infiles))
+ }
throttleMem := throttle{Max: cmd.threads} // TODO: estimate using mem and data size
throttleNumpyMem := throttle{Max: cmd.threads/2 + 1}
if tv.Ref {
continue
}
+ if mask != nil && reftile[tv.Tag] == nil {
+ // Don't waste
+ // time/memory on
+ // masked-out tiles.
+ continue
+ }
variants := seq[tv.Tag]
if len(variants) == 0 {
variants = make([]TileVariant, 100)
seq[tv.Tag] = variants
}
for _, cg := range ent.CompactGenomes {
- cgs[cg.Name] = cg
+ if matchGenome.MatchString(cg.Name) {
+ cgs[cg.Name] = cg
+ }
}
return nil
})
defer throttleCPU.Release()
count := make(map[[blake2b.Size256]byte]int, len(variants))
for _, cg := range cgs {
- idx := (tag - tagstart) * 2
- if int(idx) < len(cg.Variants) {
- count[variants[cg.Variants[idx]].Blake2b]++
- count[variants[cg.Variants[idx+1]].Blake2b]++
+ idx := int(tag-tagstart) * 2
+ if idx < len(cg.Variants) {
+ for allele := 0; allele < 2; allele++ {
+ v := cg.Variants[idx+allele]
+ if v > 0 && len(variants[v].Sequence) > 0 {
+ count[variants[v].Blake2b]++
+ }
+ }
}
}
// hash[i] will be the hash of
return err
}
annow := bufio.NewWriterSize(annof, 1<<20)
- for tag, variants := range seq {
+ outcol := 0
+ for tag := tagstart; tag < tagend; tag++ {
rt, ok := reftile[tag]
if !ok {
- // Reference does not use any
- // variant of this tile.
- // TODO: log this? mention it
- // in annotations?
+ if mask == nil {
+ outcol++
+ }
+ // Excluded by specified
+ // regions, or reference does
+ // not use any variant of this
+ // tile. (TODO: log this?
+ // mention it in annotations?)
+ continue
+ }
+ variants, ok := seq[tag]
+ if !ok {
+ outcol++
continue
}
- outcol := tag - tagID(tagstart)
reftilestr := strings.ToUpper(string(rt.tiledata))
remap := variantRemap[tag-tagstart]
+ done := make([]bool, len(variants))
for v, tv := range variants {
+ v := remap[v]
+ if done[v] {
+ continue
+ } else {
+ done[v] = true
+ }
if len(tv.Sequence) < taglen || !bytes.HasSuffix(rt.tiledata, tv.Sequence[len(tv.Sequence)-taglen:]) {
continue
}
diffs, _ := hgvs.Diff(reftilestr, strings.ToUpper(string(tv.Sequence)), 0)
for _, diff := range diffs {
diff.Position += rt.pos
- fmt.Fprintf(annow, "%d,%d,%d,%s:g.%s\n", tag, outcol, remap[v], rt.seqname, diff.String())
+ fmt.Fprintf(annow, "%d,%d,%d,%s:g.%s,%s,%d,%s,%s,%s\n", tag, outcol, v, rt.seqname, diff.String(), rt.seqname, diff.Position, diff.Ref, diff.New, diff.Left)
}
}
+ outcol++
}
err = annow.Flush()
if err != nil {
return err
}
- throttleNumpyMem.Acquire()
log.Infof("%04d: preparing numpy", infileIdx)
+ throttleNumpyMem.Acquire()
rows := len(cgnames)
- cols := 2 * int(tagend-tagstart)
+ cols := 2 * outcol
out := make([]int16, rows*cols)
for row, name := range cgnames {
out := out[row*cols:]
+ outcol := 0
for col, v := range cgs[name].Variants {
- if variants, ok := seq[tagstart+tagID(col/2)]; ok && len(variants) > int(v) && len(variants[v].Sequence) > 0 {
- out[col] = int16(variantRemap[col/2][v])
+ tag := tagstart + tagID(col/2)
+ if mask != nil && reftile[tag] == nil {
+ continue
+ }
+ if variants, ok := seq[tag]; ok && len(variants) > int(v) && len(variants[v].Sequence) > 0 {
+ out[outcol] = int16(variantRemap[tag-tagstart][v])
} else {
- out[col] = -1
+ out[outcol] = -1
}
+ outcol++
}
}
seq = nil
throttleNumpyMem.Release()
- fnm := fmt.Sprintf("%s/matrix.%04d.npy", *outputDir, infileIdx)
- output, err := os.Create(fnm)
- if err != nil {
- return err
- }
- defer output.Close()
- bufw := bufio.NewWriterSize(output, 1<<26)
- npw, err := gonpy.NewWriter(nopCloser{bufw})
- if err != nil {
- return err
- }
- log.WithFields(logrus.Fields{
- "filename": fnm,
- "rows": rows,
- "cols": cols,
- }).Infof("%04d: writing numpy", infileIdx)
- npw.Shape = []int{rows, cols}
- npw.WriteInt16(out)
- err = bufw.Flush()
- if err != nil {
- return err
- }
- err = output.Close()
- if err != nil {
- return err
+ if *mergeOutput {
+ log.Infof("%04d: matrix fragment %d rows x %d cols", infileIdx, rows, cols)
+ toMerge[infileIdx] = out
+ } else {
+ fnm := fmt.Sprintf("%s/matrix.%04d.npy", *outputDir, infileIdx)
+ err = writeNumpyInt16(fnm, out, rows, cols)
+ if err != nil {
+ return err
+ }
}
log.Infof("%s: done (%d/%d)", infile, int(atomic.AddInt64(&done, 1)), len(infiles))
return nil
if err = throttleMem.Wait(); err != nil {
return 1
}
+ if *mergeOutput {
+ log.Info("merging output matrix and annotations")
+
+ annoFilename := fmt.Sprintf("%s/matrix.annotations.csv", *outputDir)
+ annof, err := os.Create(annoFilename)
+ if err != nil {
+ return 1
+ }
+ annow := bufio.NewWriterSize(annof, 1<<20)
+
+ rows := len(cgnames)
+ cols := 0
+ for _, chunk := range toMerge {
+ cols += len(chunk) / rows
+ }
+ out := make([]int16, rows*cols)
+ startcol := 0
+ for outIdx, chunk := range toMerge {
+ chunkcols := len(chunk) / rows
+ for row := 0; row < rows; row++ {
+ copy(out[row*cols+startcol:], chunk[row*chunkcols:(row+1)*chunkcols])
+ }
+ toMerge[outIdx] = nil
+
+ annotationsFilename := fmt.Sprintf("%s/matrix.%04d.annotations.csv", *outputDir, outIdx)
+ log.Infof("reading %s", annotationsFilename)
+ buf, err := os.ReadFile(annotationsFilename)
+ if err != nil {
+ return 1
+ }
+ err = os.Remove(annotationsFilename)
+ if err != nil {
+ return 1
+ }
+ for _, line := range bytes.Split(buf, []byte{'\n'}) {
+ if len(line) == 0 {
+ continue
+ }
+ fields := bytes.SplitN(line, []byte{','}, 3)
+ incol, _ := strconv.Atoi(string(fields[1]))
+ fmt.Fprintf(annow, "%s,%d,%s\n", fields[0], incol+startcol/2, fields[2])
+ }
+
+ startcol += chunkcols
+ }
+ err = annow.Flush()
+ if err != nil {
+ return 1
+ }
+ err = annof.Close()
+ if err != nil {
+ return 1
+ }
+ err = writeNumpyInt16(fmt.Sprintf("%s/matrix.npy", *outputDir), out, rows, cols)
+ if err != nil {
+ return 1
+ }
+ }
return 0
}
+
+func writeNumpyInt16(fnm string, out []int16, rows, cols int) error {
+ output, err := os.Create(fnm)
+ if err != nil {
+ return err
+ }
+ defer output.Close()
+ bufw := bufio.NewWriterSize(output, 1<<26)
+ npw, err := gonpy.NewWriter(nopCloser{bufw})
+ if err != nil {
+ return err
+ }
+ log.WithFields(log.Fields{
+ "filename": fnm,
+ "rows": rows,
+ "cols": cols,
+ }).Infof("writing numpy: %s", fnm)
+ npw.Shape = []int{rows, cols}
+ npw.WriteInt16(out)
+ err = bufw.Flush()
+ if err != nil {
+ return err
+ }
+ return output.Close()
+}
projects / lightning.git / blobdiff