"strconv"
"strings"
"sync/atomic"
+ "unsafe"
"git.arvados.org/arvados.git/sdk/go/arvados"
"github.com/arvados/lightning/hgvs"
)
type sliceNumpy struct {
- filter filter
- threads int
- chi2CasesFile string
- chi2Cases []bool
- chi2PValue float64
- minCoverage int
- cgnames []string
+ filter filter
+ threads int
+ chi2CaseControlColumn string
+ chi2CaseControlFile string
+ chi2Cases []bool
+ chi2PValue float64
+ minCoverage int
+ cgnames []string
}
func (cmd *sliceNumpy) RunCommand(prog string, args []string, stdin io.Reader, stdout, stderr io.Writer) int {
mergeOutput := flags.Bool("merge-output", false, "merge output into one matrix.npy and one matrix.annotations.csv")
hgvsSingle := flags.Bool("single-hgvs-matrix", false, "also generate hgvs-based matrix")
hgvsChunked := flags.Bool("chunked-hgvs-matrix", false, "also generate hgvs-based matrix per chromosome")
- onehotChunked := flags.Bool("chunked-onehot", false, "generate one-hot tile-based matrix")
+ onehotSingle := flags.Bool("single-onehot", false, "generate one-hot tile-based matrix")
+ onehotChunked := flags.Bool("chunked-onehot", false, "generate one-hot tile-based matrix per input chunk")
flags.IntVar(&cmd.threads, "threads", 16, "number of memory-hungry assembly threads")
- flags.StringVar(&cmd.chi2CasesFile, "chi2-cases-file", "", "text file indicating positive cases (for Χ² test)")
+ flags.StringVar(&cmd.chi2CaseControlFile, "chi2-case-control-file", "", "tsv file or directory indicating cases and controls for Χ² test (if directory, all .tsv files will be read)")
+ flags.StringVar(&cmd.chi2CaseControlColumn, "chi2-case-control-column", "", "name of case/control column in case-control files for Χ² test (value must be 0 for control, 1 for case)")
flags.Float64Var(&cmd.chi2PValue, "chi2-p-value", 1, "do Χ² test and omit columns with p-value above this threshold")
cmd.filter.Flags(flags)
err = flags.Parse(args)
}()
}
- if cmd.chi2CasesFile == "" && cmd.chi2PValue != 1 {
- log.Errorf("cannot use provided -chi2-p-value=%f because -chi2-cases-file= value is empty", cmd.chi2PValue)
+ if cmd.chi2PValue != 1 && (cmd.chi2CaseControlFile == "" || cmd.chi2CaseControlColumn == "") {
+ log.Errorf("cannot use provided -chi2-p-value=%f because -chi2-case-control-file= or -chi2-case-control-column= value is empty", cmd.chi2PValue)
return 2
}
KeepCache: 2,
APIAccess: true,
}
- err = runner.TranslatePaths(inputDir, regionsFilename, &cmd.chi2CasesFile)
+ err = runner.TranslatePaths(inputDir, regionsFilename, &cmd.chi2CaseControlFile)
if err != nil {
return 1
}
"-merge-output=" + fmt.Sprintf("%v", *mergeOutput),
"-single-hgvs-matrix=" + fmt.Sprintf("%v", *hgvsSingle),
"-chunked-hgvs-matrix=" + fmt.Sprintf("%v", *hgvsChunked),
+ "-single-onehot=" + fmt.Sprintf("%v", *onehotSingle),
"-chunked-onehot=" + fmt.Sprintf("%v", *onehotChunked),
- "-chi2-cases-file=" + cmd.chi2CasesFile,
+ "-chi2-case-control-file=" + cmd.chi2CaseControlFile,
+ "-chi2-case-control-column=" + cmd.chi2CaseControlColumn,
"-chi2-p-value=" + fmt.Sprintf("%f", cmd.chi2PValue),
}
runner.Args = append(runner.Args, cmd.filter.Args()...)
return 0
}
- infiles, err := allGobFiles(*inputDir)
+ infiles, err := allFiles(*inputDir, matchGobFile)
if err != nil {
return 1
}
}
cmd.cgnames = nil
- taglen := -1
+ var tagset [][]byte
DecodeLibrary(in0, strings.HasSuffix(infiles[0], ".gz"), func(ent *LibraryEntry) error {
if len(ent.TagSet) > 0 {
- taglen = len(ent.TagSet[0])
+ tagset = ent.TagSet
}
for _, cseq := range ent.CompactSequences {
if cseq.Name == *ref || *ref == "" {
err = fmt.Errorf("%s: reference sequence not found", infiles[0])
return 1
}
- if taglen < 0 {
+ if len(tagset) == 0 {
err = fmt.Errorf("tagset not found")
return 1
}
+
+ taglib := &tagLibrary{}
+ err = taglib.setTags(tagset)
+ if err != nil {
+ return 1
+ }
+ taglen := taglib.TagLen()
+
if len(cmd.cgnames) == 0 {
err = fmt.Errorf("no genomes found matching regexp %q", cmd.filter.MatchGenome)
return 1
}
sort.Strings(cmd.cgnames)
-
- cmd.minCoverage = int(math.Ceil(cmd.filter.MinCoverage * float64(len(cmd.cgnames))))
-
- if cmd.chi2CasesFile != "" {
- f, err2 := open(cmd.chi2CasesFile)
- if err2 != nil {
- err = err2
- return 1
- }
- buf, err2 := io.ReadAll(f)
- f.Close()
- if err2 != nil {
- err = err2
- return 1
- }
- cmd.chi2Cases = make([]bool, len(cmd.cgnames))
- ncases := 0
- for _, pattern := range bytes.Split(buf, []byte{'\n'}) {
- if len(pattern) == 0 {
- continue
- }
- pattern := string(pattern)
- idx := -1
- for i, name := range cmd.cgnames {
- if !strings.Contains(name, pattern) {
- continue
- }
- cmd.chi2Cases[i] = true
- ncases++
- if idx >= 0 {
- log.Warnf("pattern %q in cases file matches multiple genome IDs: %q, %q", pattern, cmd.cgnames[idx], name)
- } else {
- idx = i
- }
- }
- if idx < 0 {
- log.Warnf("pattern %q in cases file does not match any genome IDs", pattern)
- continue
- }
- }
- log.Printf("%d cases, %d controls", ncases, len(cmd.cgnames)-ncases)
+ err = cmd.useCaseControlFiles()
+ if err != nil {
+ return 1
}
+ cmd.minCoverage = int(math.Ceil(cmd.filter.MinCoverage * float64(len(cmd.cgnames))))
{
- labelsFilename := *outputDir + "/labels.csv"
+ labelsFilename := *outputDir + "/samples.csv"
log.Infof("writing labels to %s", labelsFilename)
var f *os.File
f, err = os.Create(labelsFilename)
}
defer f.Close()
for i, name := range cmd.cgnames {
- _, err = fmt.Fprintf(f, "%d,%q\n", i, trimFilenameForLabel(name))
+ cc := 0
+ if cmd.chi2Cases != nil && cmd.chi2Cases[i] {
+ cc = 1
+ }
+ _, err = fmt.Fprintf(f, "%d,%q,%d\n", i, trimFilenameForLabel(name), cc)
if err != nil {
err = fmt.Errorf("write %s: %w", labelsFilename, err)
return 1
err = fmt.Errorf("missing tiledata for tag %d variant %d in %s in ref", libref.Tag, libref.Variant, seqname)
return 1
}
+ foundthistag := false
+ taglib.FindAll(tiledata[:len(tiledata)-1], func(tagid tagID, offset, _ int) {
+ if !foundthistag && tagid == libref.Tag {
+ foundthistag = true
+ return
+ }
+ if dupref, ok := reftile[tagid]; ok {
+ log.Printf("dropping reference tile %+v from %s @ %d, tag not unique, also found inside %+v from %s @ %d", tileLibRef{Tag: tagid, Variant: dupref.variant}, dupref.seqname, dupref.pos, libref, seqname, pos+offset+1)
+ delete(reftile, tagid)
+ } else {
+ log.Printf("found tag %d at offset %d inside tile variant %+v on %s @ %d", tagid, offset, libref, seqname, pos+offset+1)
+ }
+ isdup[tagid] = true
+ })
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 {
if *mergeOutput || *hgvsSingle {
toMerge = make([][]int16, len(infiles))
}
+ var onehotIndirect [][2][]uint32 // [chunkIndex][axis][index]
+ var onehotXrefs [][]onehotXref
+ if *onehotSingle {
+ onehotIndirect = make([][2][]uint32, len(infiles))
+ onehotXrefs = make([][]onehotXref, len(infiles))
+ }
throttleMem := throttle{Max: cmd.threads} // TODO: estimate using mem and data size
throttleNumpyMem := throttle{Max: cmd.threads/2 + 1}
throttleCPU.Wait()
var onehotChunk [][]int8
- var onehotXrefs []onehotXref
+ var onehotXref []onehotXref
annotationsFilename := fmt.Sprintf("%s/matrix.%04d.annotations.csv", *outputDir, infileIdx)
log.Infof("%04d: writing %s", infileIdx, annotationsFilename)
maxv = v
}
}
- if *onehotChunked {
+ if *onehotChunked || *onehotSingle {
onehot, xrefs := cmd.tv2homhet(cgs, maxv, remap, tag, tagstart)
onehotChunk = append(onehotChunk, onehot...)
- onehotXrefs = append(onehotXrefs, xrefs...)
+ onehotXref = append(onehotXref, xrefs...)
}
fmt.Fprintf(annow, "%d,%d,%d,=,%s,%d,,,\n", tag, outcol, rt.variant, rt.seqname, rt.pos)
variants := seq[tag]
// transpose onehotChunk[col][row] to numpy[row*ncols+col]
rows := len(cmd.cgnames)
cols := len(onehotChunk)
- log.Infof("%04d: preparing onehot numpy (rows=%d, cols=%d, mem=%d)", infileIdx, rows, cols, rows*cols)
+ log.Infof("%04d: preparing onehot numpy (rows=%d, cols=%d, mem=%d)", infileIdx, len(cmd.cgnames), len(onehotChunk), len(cmd.cgnames)*len(onehotChunk))
throttleNumpyMem.Acquire()
- out := make([]int8, rows*cols)
- for row := range cmd.cgnames {
- out := out[row*cols:]
- for colnum, values := range onehotChunk {
- out[colnum] = values[row]
- }
- }
- seq = nil
- cgs = nil
- debug.FreeOSMemory()
- throttleNumpyMem.Release()
-
+ out := onehotcols2int8(onehotChunk)
fnm := fmt.Sprintf("%s/onehot.%04d.npy", *outputDir, infileIdx)
err = writeNumpyInt8(fnm, out, rows, cols)
if err != nil {
return err
}
-
fnm = fmt.Sprintf("%s/onehot-columns.%04d.npy", *outputDir, infileIdx)
- xcols := len(onehotXrefs)
- xdata := make([]int32, 4*xcols)
- for i, xref := range onehotXrefs {
- xdata[i] = int32(xref.tag)
- xdata[xcols+i] = int32(xref.variant)
- if xref.het {
- xdata[xcols*2+i] = 1
- }
- xdata[xcols*3+i] = int32(xref.pvalue * 1000000)
- }
- err = writeNumpyInt32(fnm, xdata, 4, xcols)
+ err = writeNumpyInt32(fnm, onehotXref2int32(onehotXref), 4, len(onehotXref))
if err != nil {
return err
}
+ debug.FreeOSMemory()
+ throttleNumpyMem.Release()
+ }
+ if *onehotSingle {
+ onehotIndirect[infileIdx] = onehotChunk2Indirect(onehotChunk)
+ onehotXrefs[infileIdx] = onehotXref
+ n := len(onehotIndirect[infileIdx][0])
+ log.Infof("%04d: keeping onehot coordinates in memory (n=%d, mem=%d)", infileIdx, n, n*8)
}
- if !*onehotChunked || *mergeOutput || *hgvsSingle {
+ if !(*onehotSingle || *onehotChunked) || *mergeOutput || *hgvsSingle {
log.Infof("%04d: preparing numpy", infileIdx)
throttleNumpyMem.Acquire()
rows := len(cmd.cgnames)
log.Infof("%04d: matrix fragment %d rows x %d cols", infileIdx, rows, cols)
toMerge[infileIdx] = out
}
- if !*mergeOutput && !*onehotChunked {
+ if !*mergeOutput && !*onehotChunked && !*onehotSingle {
fnm := fmt.Sprintf("%s/matrix.%04d.npy", *outputDir, infileIdx)
err = writeNumpyInt16(fnm, out, rows, cols)
if err != nil {
}
}
}
+ if *onehotSingle {
+ nzCount := 0
+ for _, part := range onehotIndirect {
+ nzCount += len(part[0])
+ }
+ onehot := make([]uint32, nzCount*2) // [r,r,r,...,c,c,c,...]
+ var xrefs []onehotXref
+ outcol := 0
+ for i, part := range onehotIndirect {
+ for i := range part[1] {
+ part[1][i] += uint32(outcol)
+ }
+ copy(onehot[outcol:], part[0])
+ copy(onehot[outcol+nzCount:], part[1])
+ outcol += len(part[0])
+ xrefs = append(xrefs, onehotXrefs[i]...)
+
+ part[0] = nil
+ part[1] = nil
+ onehotXrefs[i] = nil
+ debug.FreeOSMemory()
+ }
+ fnm := fmt.Sprintf("%s/onehot.npy", *outputDir)
+ err = writeNumpyUint32(fnm, onehot, 2, nzCount)
+ if err != nil {
+ return 1
+ }
+ fnm = fmt.Sprintf("%s/onehot-columns.npy", *outputDir)
+ err = writeNumpyInt32(fnm, onehotXref2int32(xrefs), 4, len(xrefs))
+ if err != nil {
+ return 1
+ }
+ }
return 0
}
+// Read case/control files, remove non-case/control entries from
+// cmd.cgnames, and build cmd.chi2Cases.
+func (cmd *sliceNumpy) useCaseControlFiles() error {
+ if cmd.chi2CaseControlFile == "" {
+ return nil
+ }
+ infiles, err := allFiles(cmd.chi2CaseControlFile, nil)
+ if err != nil {
+ return err
+ }
+ // index in cmd.cgnames => case(true) / control(false)
+ cc := map[int]bool{}
+ for _, infile := range infiles {
+ f, err := open(infile)
+ if err != nil {
+ return err
+ }
+ buf, err := io.ReadAll(f)
+ f.Close()
+ if err != nil {
+ return err
+ }
+ ccCol := -1
+ for _, tsv := range bytes.Split(buf, []byte{'\n'}) {
+ if len(tsv) == 0 {
+ continue
+ }
+ split := strings.Split(string(tsv), "\t")
+ if ccCol < 0 {
+ // header row
+ for col, name := range split {
+ if name == cmd.chi2CaseControlColumn {
+ ccCol = col
+ break
+ }
+ }
+ if ccCol < 0 {
+ return fmt.Errorf("%s: no column named %q in header row %q", infile, cmd.chi2CaseControlColumn, tsv)
+ }
+ continue
+ }
+ if len(split) <= ccCol {
+ continue
+ }
+ pattern := split[0]
+ found := -1
+ for i, name := range cmd.cgnames {
+ if strings.Contains(name, pattern) {
+ if found >= 0 {
+ log.Warnf("pattern %q in %s matches multiple genome IDs (%qs, %q)", pattern, infile, cmd.cgnames[found], name)
+ }
+ found = i
+ }
+ }
+ if found < 0 {
+ log.Warnf("pattern %q in %s does not match any genome IDs", pattern, infile)
+ continue
+ }
+ if split[ccCol] == "0" {
+ cc[found] = false
+ }
+ if split[ccCol] == "1" {
+ cc[found] = true
+ }
+ }
+ }
+ allnames := cmd.cgnames
+ cmd.cgnames = nil
+ cmd.chi2Cases = nil
+ ncases := 0
+ for i, name := range allnames {
+ if cc, ok := cc[i]; ok {
+ cmd.cgnames = append(cmd.cgnames, name)
+ cmd.chi2Cases = append(cmd.chi2Cases, cc)
+ if cc {
+ ncases++
+ }
+ }
+ }
+ log.Printf("%d cases, %d controls, %d neither (dropped)", ncases, len(cmd.cgnames)-ncases, len(allnames)-len(cmd.cgnames))
+ return nil
+}
+
func (cmd *sliceNumpy) filterHGVScolpair(colpair [2][]int8) bool {
if cmd.chi2PValue >= 1 {
return true
cases = append(cases, c)
}
return len(cases) >= cmd.minCoverage &&
- (pvalue(cases, col0) <= cmd.chi2PValue || pvalue(cases, col1) <= cmd.chi2PValue)
+ (pvalue(col0, cases) <= cmd.chi2PValue || pvalue(col1, cases) <= cmd.chi2PValue)
+}
+
+func writeNumpyUint32(fnm string, out []uint32, 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,
+ "bytes": rows * cols * 4,
+ }).Infof("writing numpy: %s", fnm)
+ npw.Shape = []int{rows, cols}
+ npw.WriteUint32(out)
+ err = bufw.Flush()
+ if err != nil {
+ return err
+ }
+ return output.Close()
}
func writeNumpyInt32(fnm string, out []int32, rows, cols int) error {
pvalue float64
}
+const onehotXrefSize = unsafe.Sizeof(onehotXref{})
+
// Build onehot matrix (m[variant*2+isHet][genome] == 0 or 1) for all
// variants of a single tile/tag#.
//
var xref []onehotXref
for homcol := 4; homcol < len(obs); homcol += 2 {
p := [2]float64{
- pvalue(cmd.chi2Cases, obs[homcol]),
- pvalue(cmd.chi2Cases, obs[homcol+1]),
+ pvalue(obs[homcol], cmd.chi2Cases),
+ pvalue(obs[homcol+1], cmd.chi2Cases),
}
if cmd.chi2PValue < 1 && !(p[0] < cmd.chi2PValue || p[1] < cmd.chi2PValue) {
continue
}
return out
}
+
+// convert a []onehotXref with length N to a numpy-style []int32
+// matrix with N columns, one row per field of onehotXref struct.
+//
+// Hom/het row contains hom=0, het=1.
+//
+// P-value row contains 1000000x actual p-value.
+func onehotXref2int32(xrefs []onehotXref) []int32 {
+ xcols := len(xrefs)
+ xdata := make([]int32, 4*xcols)
+ for i, xref := range xrefs {
+ xdata[i] = int32(xref.tag)
+ xdata[xcols+i] = int32(xref.variant)
+ if xref.het {
+ xdata[xcols*2+i] = 1
+ }
+ xdata[xcols*3+i] = int32(xref.pvalue * 1000000)
+ }
+ return xdata
+}
+
+// transpose onehot data from in[col][row] to numpy-style
+// out[row*cols+col].
+func onehotcols2int8(in [][]int8) []int8 {
+ if len(in) == 0 {
+ return nil
+ }
+ cols := len(in)
+ rows := len(in[0])
+ out := make([]int8, rows*cols)
+ for row := 0; row < rows; row++ {
+ outrow := out[row*cols:]
+ for col, incol := range in {
+ outrow[col] = incol[row]
+ }
+ }
+ return out
+}
+
+// Return [2][]uint32{rowIndices, colIndices} indicating which
+// elements of matrixT[c][r] have non-zero values.
+func onehotChunk2Indirect(matrixT [][]int8) [2][]uint32 {
+ var nz [2][]uint32
+ for c, col := range matrixT {
+ for r, val := range col {
+ if val != 0 {
+ nz[0] = append(nz[0], uint32(r))
+ nz[1] = append(nz[1], uint32(c))
+ }
+ }
+ }
+ return nz
+}