"flag"
"fmt"
"io"
+ "io/ioutil"
"net/http"
_ "net/http/pprof"
"os"
"regexp"
"runtime"
"sort"
+ "strconv"
"strings"
"sync/atomic"
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)
"-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
type reftileinfo struct {
variant tileVariantID
seqname string // chr1
- pos int // distance from start of chr1 to start of tile
+ pos int // distance from start of chromosome to starttag
tiledata []byte // acgtggcaa...
}
+ isdup := map[tagID]bool{}
reftile := map[tagID]*reftileinfo{}
for seqname, cseq := range refseq {
+ pos := 0
for _, libref := range cseq {
- reftile[libref.Tag] = &reftileinfo{
- seqname: seqname,
- variant: libref.Variant,
- tiledata: reftiledata[libref],
+ tiledata := reftiledata[libref]
+ if len(tiledata) == 0 {
+ err = fmt.Errorf("missing tiledata for tag %d variant %d in %s in ref", libref.Tag, libref.Variant, seqname)
+ return 1
}
- }
- }
-
- throttleCPU := throttle{Max: runtime.GOMAXPROCS(0)}
- log.Info("reconstructing reference sequences")
- for seqname, cseq := range refseq {
- seqname, cseq := seqname, cseq
- throttleCPU.Go(func() error {
- defer log.Printf("... %s done", seqname)
- pos := 0
- for _, libref := range cseq {
- rt := reftile[libref.Tag]
- rt.pos = pos
- if len(rt.tiledata) == 0 {
- return fmt.Errorf("missing tiledata for tag %d variant %d in %s in ref", libref.Tag, libref.Variant, seqname)
+ if isdup[libref.Tag] {
+ log.Printf("dropping reference tile %+v from %s @ %d, tag not unique", libref, seqname, pos)
+ } else if reftile[libref.Tag] != nil {
+ log.Printf("dropping reference tile %+v from %s @ %d, tag not unique", tileLibRef{Tag: libref.Tag, Variant: reftile[libref.Tag].variant}, reftile[libref.Tag].seqname, reftile[libref.Tag].pos)
+ delete(reftile, libref.Tag)
+ log.Printf("dropping reference tile %+v from %s @ %d, tag not unique", libref, seqname, pos)
+ isdup[libref.Tag] = true
+ } else {
+ reftile[libref.Tag] = &reftileinfo{
+ seqname: seqname,
+ variant: libref.Variant,
+ tiledata: tiledata,
+ pos: pos,
}
- pos += len(rt.tiledata) - taglen
}
- return nil
- })
+ pos += len(tiledata) - taglen
+ }
+ log.Printf("... %s done, len %d", seqname, pos+taglen)
}
- throttleCPU.Wait()
var mask *mask
if *regionsFilename != "" {
+ log.Printf("loading regions from %s", *regionsFilename)
mask, err = makeMask(*regionsFilename, *expandRegions)
if err != nil {
return 1
}
- // Delete reftile entries for masked-out regions.
+ log.Printf("before applying mask, len(reftile) == %d", len(reftile))
+ log.Printf("deleting reftile entries for regions outside %d intervals", mask.Len())
for tag, rt := range reftile {
if !mask.Check(strings.TrimPrefix(rt.seqname, "chr"), rt.pos, rt.pos+len(rt.tiledata)) {
delete(reftile, tag)
}
}
+ log.Printf("after applying mask, len(reftile) == %d", len(reftile))
+ }
+
+ var toMerge [][]int16
+ if *mergeOutput {
+ toMerge = make([][]int16, len(infiles))
}
throttleMem := throttle{Max: cmd.threads} // TODO: estimate using mem and data size
}
variants, ok := seq[tag]
if !ok {
- outcol++
- continue
+ // how could we even have a reftile if there is no sequence data??
+ return fmt.Errorf("bug: have no variants for tag %d but reftile is %+v", tag, rt)
}
reftilestr := strings.ToUpper(string(rt.tiledata))
remap := variantRemap[tag-tagstart]
done[v] = true
}
if len(tv.Sequence) < taglen || !bytes.HasSuffix(rt.tiledata, tv.Sequence[len(tv.Sequence)-taglen:]) {
+ fmt.Fprintf(annow, "%d,%d,%d,.,%s,%d,.,,\n", tag, outcol, v, rt.seqname, rt.pos)
continue
}
if lendiff := len(rt.tiledata) - len(tv.Sequence); lendiff < -1000 || lendiff > 1000 {
+ fmt.Fprintf(annow, "%d,%d,%d,,%s,%d,,,\n", tag, outcol, v, rt.seqname, rt.pos)
continue
}
diffs, _ := hgvs.Diff(reftilestr, strings.ToUpper(string(tv.Sequence)), 0)
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(log.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)
+ hgvs := map[string][2][]int16{} // hgvs -> [[g0,g1,g2,...], [g0,g1,g2,...]] (slice of genomes for each phase)
+ 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{','}, 8)
+ incol, _ := strconv.Atoi(string(fields[1]))
+ tileVariant, _ := strconv.Atoi(string(fields[2]))
+ hgvsID := string(fields[3])
+ seqname := string(fields[4])
+ pos, _ := strconv.Atoi(string(fields[5]))
+ refseq := fields[6]
+ if mask != nil && !mask.Check(strings.TrimPrefix(seqname, "chr"), pos, pos+len(refseq)) {
+ // The tile intersects one of
+ // the selected regions, but
+ // this particular HGVS
+ // variant does not.
+ continue
+ }
+ fmt.Fprintf(annow, "%s,%d,%d,%s,%s,%d,%s,%s\n", fields[0], incol+startcol/2, tileVariant, hgvsID, seqname, pos, refseq, fields[7])
+ hgvscols := hgvs[hgvsID]
+ if hgvscols[0] == nil {
+ hgvscols = [2][]int16{make([]int16, len(cgnames)), make([]int16, len(cgnames))}
+ hgvs[hgvsID] = hgvscols
+ }
+ for ph := 0; ph < 2; ph++ {
+ for row := 0; row < rows; row++ {
+ v := chunk[row*chunkcols+incol*2+ph]
+ if int(v) == tileVariant {
+ if len(hgvsID) == 0 {
+ // we have the tile variant sequence, but the diff against ref didn't work out (see lendiff above)
+ hgvscols[ph][row] = -2
+ } else {
+ hgvscols[ph][row] = 1
+ }
+ } else if v < 0 {
+ hgvscols[ph][row] = -1
+ }
+ }
+ }
+ }
+
+ 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
+ }
+ out = nil
+
+ cols = len(hgvs) * 2
+ log.Printf("building hgvs-based matrix: %d rows x %d cols", rows, cols)
+ out = make([]int16, rows*cols)
+ hgvsIDs := make([]string, 0, len(hgvs))
+ for hgvsID := range hgvs {
+ hgvsIDs = append(hgvsIDs, hgvsID)
+ }
+ sort.Strings(hgvsIDs)
+ var hgvsLabels bytes.Buffer
+ for idx, hgvsID := range hgvsIDs {
+ fmt.Fprintf(&hgvsLabels, "%d,%s\n", idx, hgvsID)
+ for ph := 0; ph < 2; ph++ {
+ hgvscol := hgvs[hgvsID][ph]
+ for row, val := range hgvscol {
+ out[row*cols+idx*2+ph] = val
+ }
+ }
+ }
+ err = writeNumpyInt16(fmt.Sprintf("%s/hgvs.npy", *outputDir), out, rows, cols)
+ if err != nil {
+ return 1
+ }
+
+ fnm := fmt.Sprintf("%s/hgvs.annotations.csv", *outputDir)
+ log.Printf("writing hgvs labels: %s", fnm)
+ err = ioutil.WriteFile(fnm, hgvsLabels.Bytes(), 0777)
+ 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