"bufio"
"bytes"
"encoding/gob"
+ "errors"
"flag"
"fmt"
"io"
"git.arvados.org/arvados.git/sdk/go/arvados"
"github.com/arvados/lightning/hgvs"
+ "github.com/james-bowman/nlp"
"github.com/kshedden/gonpy"
+ "github.com/sirupsen/logrus"
log "github.com/sirupsen/logrus"
"golang.org/x/crypto/blake2b"
+ "gonum.org/v1/gonum/mat"
)
+const annotationMaxTileSpan = 100
+
type sliceNumpy struct {
filter filter
threads int
minCoverage int
cgnames []string
includeVariant1 bool
+ debugTag tagID
}
func (cmd *sliceNumpy) RunCommand(prog string, args []string, stdin io.Reader, stdout, stderr io.Writer) int {
- var err error
- defer func() {
- if err != nil {
- fmt.Fprintf(stderr, "%s\n", err)
- }
- }()
+ err := cmd.run(prog, args, stdin, stdout, stderr)
+ if err != nil {
+ fmt.Fprintf(stderr, "%s\n", err)
+ return 1
+ }
+ return 0
+}
+func (cmd *sliceNumpy) run(prog string, args []string, stdin io.Reader, stdout, stderr io.Writer) error {
flags := flag.NewFlagSet("", flag.ContinueOnError)
flags.SetOutput(stderr)
pprof := flags.String("pprof", "", "serve Go profile data at http://`[addr]:port`")
runlocal := flags.Bool("local", false, "run on local host (default: run in an arvados container)")
+ arvadosRAM := flags.Int("arvados-ram", 750000000000, "amount of memory to request for arvados container (`bytes`)")
+ arvadosVCPUs := flags.Int("arvados-vcpus", 96, "number of VCPUs to request for arvados container")
projectUUID := flags.String("project", "", "project `UUID` for output data")
priority := flags.Int("priority", 500, "container request priority")
inputDir := flags.String("input-dir", "./in", "input `directory`")
hgvsChunked := flags.Bool("chunked-hgvs-matrix", false, "also generate hgvs-based matrix per chromosome")
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")
+ onlyPCA := flags.Bool("pca", false, "generate pca matrix")
+ pcaComponents := flags.Int("pca-components", 4, "number of PCA components")
+ maxPCATiles := flags.Int("max-pca-tiles", 0, "maximum tiles to use as PCA input (filter, then drop every 2nd colum pair until below max)")
+ debugTag := flags.Int("debug-tag", -1, "log debugging details about specified tag")
+ flags.IntVar(&cmd.threads, "threads", 16, "number of memory-hungry assembly threads, and number of VCPUs to request for arvados container")
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")
flags.BoolVar(&cmd.includeVariant1, "include-variant-1", false, "include most common variant when building one-hot matrix")
cmd.filter.Flags(flags)
- err = flags.Parse(args)
+ err := flags.Parse(args)
if err == flag.ErrHelp {
- err = nil
- return 0
+ return nil
} else if err != nil {
- return 2
+ return err
}
if *pprof != "" {
}
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
+ return fmt.Errorf("cannot use provided -chi2-p-value=%f because -chi2-case-control-file= or -chi2-case-control-column= value is empty", cmd.chi2PValue)
}
+ cmd.debugTag = tagID(*debugTag)
+
if !*runlocal {
runner := arvadosContainerRunner{
Name: "lightning slice-numpy",
Client: arvados.NewClientFromEnv(),
ProjectUUID: *projectUUID,
- RAM: 750000000000,
- VCPUs: 96,
+ RAM: int64(*arvadosRAM),
+ VCPUs: *arvadosVCPUs,
Priority: *priority,
KeepCache: 2,
APIAccess: true,
}
err = runner.TranslatePaths(inputDir, regionsFilename, &cmd.chi2CaseControlFile)
if err != nil {
- return 1
+ return err
}
runner.Args = []string{"slice-numpy", "-local=true",
"-pprof=:6060",
"-chunked-hgvs-matrix=" + fmt.Sprintf("%v", *hgvsChunked),
"-single-onehot=" + fmt.Sprintf("%v", *onehotSingle),
"-chunked-onehot=" + fmt.Sprintf("%v", *onehotChunked),
+ "-pca=" + fmt.Sprintf("%v", *onlyPCA),
+ "-pca-components=" + fmt.Sprintf("%d", *pcaComponents),
+ "-max-pca-tiles=" + fmt.Sprintf("%d", *maxPCATiles),
"-chi2-case-control-file=" + cmd.chi2CaseControlFile,
"-chi2-case-control-column=" + cmd.chi2CaseControlColumn,
"-chi2-p-value=" + fmt.Sprintf("%f", cmd.chi2PValue),
+ "-include-variant-1=" + fmt.Sprintf("%v", cmd.includeVariant1),
+ "-debug-tag=" + fmt.Sprintf("%d", cmd.debugTag),
}
runner.Args = append(runner.Args, cmd.filter.Args()...)
var output string
output, err = runner.Run()
if err != nil {
- return 1
+ return err
}
fmt.Fprintln(stdout, output)
- return 0
+ return nil
}
infiles, err := allFiles(*inputDir, matchGobFile)
if err != nil {
- return 1
+ return err
}
if len(infiles) == 0 {
err = fmt.Errorf("no input files found in %s", *inputDir)
- return 1
+ return err
}
sort.Strings(infiles)
var reftiledata = make(map[tileLibRef][]byte, 11000000)
in0, err := open(infiles[0])
if err != nil {
- return 1
+ return err
}
matchGenome, err := regexp.Compile(cmd.filter.MatchGenome)
if err != nil {
err = fmt.Errorf("-match-genome: invalid regexp: %q", cmd.filter.MatchGenome)
- return 1
+ return err
}
cmd.cgnames = nil
var tagset [][]byte
- DecodeLibrary(in0, strings.HasSuffix(infiles[0], ".gz"), func(ent *LibraryEntry) error {
+ err = DecodeLibrary(in0, strings.HasSuffix(infiles[0], ".gz"), func(ent *LibraryEntry) error {
if len(ent.TagSet) > 0 {
tagset = ent.TagSet
}
return nil
})
if err != nil {
- return 1
+ return err
}
in0.Close()
if refseq == nil {
err = fmt.Errorf("%s: reference sequence not found", infiles[0])
- return 1
+ return err
}
if len(tagset) == 0 {
err = fmt.Errorf("tagset not found")
- return 1
+ return err
}
taglib := &tagLibrary{}
err = taglib.setTags(tagset)
if err != nil {
- return 1
+ return err
}
taglen := taglib.TagLen()
if len(cmd.cgnames) == 0 {
err = fmt.Errorf("no genomes found matching regexp %q", cmd.filter.MatchGenome)
- return 1
+ return err
}
sort.Strings(cmd.cgnames)
err = cmd.useCaseControlFiles()
if err != nil {
- return 1
+ return err
+ }
+ if len(cmd.cgnames) == 0 {
+ err = fmt.Errorf("fatal: 0 cases, 0 controls, nothing to do")
+ return err
+ }
+ if cmd.filter.MinCoverage == 1 {
+ // In the generic formula below, floating point
+ // arithmetic can effectively push the coverage
+ // threshold above 1.0, which is impossible/useless.
+ // 1.0 needs to mean exactly 100% coverage.
+ cmd.minCoverage = len(cmd.cgnames)
+ } else {
+ cmd.minCoverage = int(math.Ceil(cmd.filter.MinCoverage * float64(len(cmd.cgnames))))
}
- cmd.minCoverage = int(math.Ceil(cmd.filter.MinCoverage * float64(len(cmd.cgnames))))
{
labelsFilename := *outputDir + "/samples.csv"
var f *os.File
f, err = os.Create(labelsFilename)
if err != nil {
- return 1
+ return err
}
defer f.Close()
for i, name := range cmd.cgnames {
_, 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
+ return err
}
}
err = f.Close()
if err != nil {
err = fmt.Errorf("close %s: %w", labelsFilename, err)
- return 1
+ return err
}
}
seqname string // chr1
pos int // distance from start of chromosome to starttag
tiledata []byte // acgtggcaa...
+ excluded bool // true if excluded by regions file
+ nexttag tagID // tagID of following tile (-1 for last tag of chromosome)
}
isdup := map[tagID]bool{}
reftile := map[tagID]*reftileinfo{}
for seqname, cseq := range refseq {
pos := 0
+ lastreftag := tagID(-1)
for _, libref := range cseq {
if cmd.filter.MaxTag >= 0 && libref.Tag > tagID(cmd.filter.MaxTag) {
continue
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
+ return err
}
foundthistag := false
taglib.FindAll(tiledata[:len(tiledata)-1], func(tagid tagID, offset, _ int) {
variant: libref.Variant,
tiledata: tiledata,
pos: pos,
+ nexttag: -1,
}
+ if lastreftag >= 0 {
+ reftile[lastreftag].nexttag = libref.Tag
+ }
+ lastreftag = libref.Tag
}
pos += len(tiledata) - taglen
}
log.Printf("loading regions from %s", *regionsFilename)
mask, err = makeMask(*regionsFilename, *expandRegions)
if err != nil {
- return 1
+ return err
}
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 {
+ for _, rt := range reftile {
if !mask.Check(strings.TrimPrefix(rt.seqname, "chr"), rt.pos, rt.pos+len(rt.tiledata)) {
- delete(reftile, tag)
+ rt.excluded = true
}
}
log.Printf("after applying mask, len(reftile) == %d", len(reftile))
var f *os.File
f, err = os.Create(*outputDir + "/tmp." + seqname + ".gob")
if err != nil {
- return 1
+ return err
}
defer os.Remove(f.Name())
bufw := bufio.NewWriterSize(f, 1<<24)
var onehotIndirect [][2][]uint32 // [chunkIndex][axis][index]
var onehotChunkSize []uint32
var onehotXrefs [][]onehotXref
- if *onehotSingle {
+ if *onehotSingle || *onlyPCA {
onehotIndirect = make([][2][]uint32, len(infiles))
onehotChunkSize = make([]uint32, len(infiles))
onehotXrefs = make([][]onehotXref, len(infiles))
}
+ chunkStartTag := make([]tagID, len(infiles))
throttleMem := throttle{Max: cmd.threads} // TODO: estimate using mem and data size
throttleNumpyMem := throttle{Max: cmd.threads/2 + 1}
log.Info("generating annotations and numpy matrix for each slice")
+ var errSkip = errors.New("skip infile")
var done int64
for infileIdx, infile := range infiles {
infileIdx, infile := infileIdx, infile
if tv.Ref {
continue
}
+ // Skip tile with no
+ // corresponding ref tile, if
+ // mask is in play (we can't
+ // determine coordinates for
+ // these)
if mask != nil && reftile[tv.Tag] == nil {
- // Don't waste
- // time/memory on
- // masked-out tiles.
continue
}
+ // Skip tile whose
+ // corresponding ref tile is
+ // outside target regions --
+ // unless it's a potential
+ // spanning tile.
+ if mask != nil && reftile[tv.Tag].excluded &&
+ (int(tv.Tag+1) >= len(tagset) ||
+ (bytes.HasSuffix(tv.Sequence, tagset[tv.Tag+1]) && reftile[tv.Tag+1] != nil && !reftile[tv.Tag+1].excluded)) {
+ continue
+ }
+ if tv.Tag == cmd.debugTag {
+ log.Printf("infile %d %s tag %d variant %d hash %x", infileIdx, infile, tv.Tag, tv.Variant, tv.Blake2b[:3])
+ }
variants := seq[tv.Tag]
if len(variants) == 0 {
variants = make([]TileVariant, 100)
seq[tv.Tag] = variants
}
for _, cg := range ent.CompactGenomes {
+ if cmd.filter.MaxTag >= 0 && cg.StartTag > tagID(cmd.filter.MaxTag) {
+ return errSkip
+ }
if !matchGenome.MatchString(cg.Name) {
continue
}
}
return nil
})
- if err != nil {
- return err
+ if err == errSkip {
+ return nil
+ } else if err != nil {
+ return fmt.Errorf("%04d: DecodeLibrary(%s): err", infileIdx, infile)
}
tagstart := cgs[cmd.cgnames[0]].StartTag
tagend := cgs[cmd.cgnames[0]].EndTag
+ chunkStartTag[infileIdx] = tagstart
// TODO: filters
throttleCPU := throttle{Max: runtime.GOMAXPROCS(0)}
for tag, variants := range seq {
tag, variants := tag, variants
- throttleCPU.Acquire()
- go func() {
- defer throttleCPU.Release()
+ throttleCPU.Go(func() error {
+ alleleCoverage := 0
count := make(map[[blake2b.Size256]byte]int, len(variants))
rt := reftile[tag]
count[blake2b.Sum256(rt.tiledata)] = 0
}
- for _, cg := range cgs {
+ for cgname, cg := range cgs {
idx := int(tag-tagstart) * 2
for allele := 0; allele < 2; allele++ {
v := cg.Variants[idx+allele]
if v > 0 && len(variants[v].Sequence) > 0 {
count[variants[v].Blake2b]++
+ alleleCoverage++
+ }
+ if v > 0 && tag == cmd.debugTag {
+ log.Printf("tag %d cg %s allele %d tv %d hash %x count is now %d", tag, cgname, allele, v, variants[v].Blake2b[:3], count[variants[v].Blake2b])
}
}
}
+ if alleleCoverage < cmd.minCoverage*2 {
+ idx := int(tag-tagstart) * 2
+ for _, cg := range cgs {
+ cg.Variants[idx] = 0
+ cg.Variants[idx+1] = 0
+ }
+ if tag == cmd.debugTag {
+ log.Printf("tag %d alleleCoverage %d < min %d, sample data wiped", tag, alleleCoverage, cmd.minCoverage*2)
+ }
+ return nil
+ }
+
// hash[i] will be the hash of
// the variant(s) that should
// be at rank i (0-based).
for i, h := range hash {
rank[h] = tileVariantID(i + 1)
}
+ if tag == cmd.debugTag {
+ for h, r := range rank {
+ log.Printf("tag %d rank(%x) = %v", tag, h[:3], r)
+ }
+ }
// remap[v] will be the new
// variant number for original
// variant number v.
for i, tv := range variants {
remap[i] = rank[tv.Blake2b]
}
+ if tag == cmd.debugTag {
+ for in, out := range remap {
+ if out > 0 {
+ log.Printf("tag %d remap %d => %d", tag, in, out)
+ }
+ }
+ }
variantRemap[tag-tagstart] = remap
if rt != nil {
- rt.variant = rank[blake2b.Sum256(rt.tiledata)]
+ refrank := rank[blake2b.Sum256(rt.tiledata)]
+ if tag == cmd.debugTag {
+ log.Printf("tag %d reftile variant %d => %d", tag, rt.variant, refrank)
+ }
+ rt.variant = refrank
}
- }()
+ return nil
+ })
}
throttleCPU.Wait()
var onehotChunk [][]int8
var onehotXref []onehotXref
- annotationsFilename := fmt.Sprintf("%s/matrix.%04d.annotations.csv", *outputDir, infileIdx)
- log.Infof("%04d: writing %s", infileIdx, annotationsFilename)
+ var annotationsFilename string
+ if *onlyPCA {
+ annotationsFilename = "/dev/null"
+ } else {
+ annotationsFilename = fmt.Sprintf("%s/matrix.%04d.annotations.csv", *outputDir, infileIdx)
+ log.Infof("%04d: writing %s", infileIdx, annotationsFilename)
+ }
annof, err := os.Create(annotationsFilename)
if err != nil {
return err
for tag := tagstart; tag < tagend; tag++ {
rt := reftile[tag]
if rt == nil && mask != nil {
- // Excluded by specified regions
+ // With no ref tile, we don't
+ // have coordinates to say
+ // this is in the desired
+ // regions -- so it's not.
+ // TODO: handle ref spanning
+ // tile case.
continue
}
- if cmd.filter.MaxTag >= 0 && tag > tagID(cmd.filter.MaxTag) {
+ if rt != nil && rt.excluded {
+ // TODO: don't skip yet --
+ // first check for spanning
+ // tile variants that
+ // intersect non-excluded ref
+ // tiles.
continue
}
+ if cmd.filter.MaxTag >= 0 && tag > tagID(cmd.filter.MaxTag) {
+ break
+ }
remap := variantRemap[tag-tagstart]
maxv := tileVariantID(0)
for _, v := range remap {
maxv = v
}
}
- if *onehotChunked || *onehotSingle {
- onehot, xrefs := cmd.tv2homhet(cgs, maxv, remap, tag, tagstart)
+ if *onehotChunked || *onehotSingle || *onlyPCA {
+ onehot, xrefs := cmd.tv2homhet(cgs, maxv, remap, tag, tagstart, seq)
+ if tag == cmd.debugTag {
+ log.WithFields(logrus.Fields{
+ "onehot": onehot,
+ "xrefs": xrefs,
+ }).Info("tv2homhet()")
+ }
onehotChunk = append(onehotChunk, onehot...)
onehotXref = append(onehotXref, xrefs...)
}
+ if *onlyPCA {
+ outcol++
+ continue
+ }
if rt == nil {
// Reference does not use any
// variant of this tile
+ //
+ // TODO: diff against the
+ // relevant portion of the
+ // ref's spanning tile
outcol++
continue
}
variantDiffs := make([][]hgvs.Variant, maxv+1)
for v, tv := range variants {
v := remap[v]
- if v == rt.variant || done[v] {
+ if v == 0 || v == rt.variant || done[v] {
continue
} else {
done[v] = true
}
- if len(tv.Sequence) < taglen || !bytes.HasSuffix(rt.tiledata, tv.Sequence[len(tv.Sequence)-taglen:]) {
+ if len(tv.Sequence) < taglen {
+ continue
+ }
+ // if reftilestr doesn't end
+ // in the same tag as tv,
+ // extend reftilestr with
+ // following ref tiles until
+ // it does (up to an arbitrary
+ // sanity-check limit)
+ reftilestr := reftilestr
+ endtagstr := strings.ToUpper(string(tv.Sequence[len(tv.Sequence)-taglen:]))
+ for i, rt := 0, rt; i < annotationMaxTileSpan && !strings.HasSuffix(reftilestr, endtagstr) && rt.nexttag >= 0; i++ {
+ rt = reftile[rt.nexttag]
+ if rt == nil {
+ break
+ }
+ reftilestr += strings.ToUpper(string(rt.tiledata[taglen:]))
+ }
+ if mask != nil && !mask.Check(strings.TrimPrefix(rt.seqname, "chr"), rt.pos, rt.pos+len(reftilestr)) {
+ continue
+ }
+ if !strings.HasSuffix(reftilestr, endtagstr) {
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 {
+ if lendiff := len(reftilestr) - len(tv.Sequence); lendiff < -1000 || lendiff > 1000 {
fmt.Fprintf(annow, "%d,%d,%d,,%s,%d,,,\n", tag, outcol, v, rt.seqname, rt.pos)
continue
}
// 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, len(cmd.cgnames), len(onehotChunk), len(cmd.cgnames)*len(onehotChunk))
+ log.Infof("%04d: preparing onehot numpy (rows=%d, cols=%d, mem=%d)", infileIdx, rows, cols, rows*cols)
throttleNumpyMem.Acquire()
out := onehotcols2int8(onehotChunk)
fnm := fmt.Sprintf("%s/onehot.%04d.npy", *outputDir, infileIdx)
debug.FreeOSMemory()
throttleNumpyMem.Release()
}
- if *onehotSingle {
+ if *onehotSingle || *onlyPCA {
onehotIndirect[infileIdx] = onehotChunk2Indirect(onehotChunk)
onehotChunkSize[infileIdx] = uint32(len(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*2)
}
- if !(*onehotSingle || *onehotChunked) || *mergeOutput || *hgvsSingle {
+ if !(*onehotSingle || *onehotChunked || *onlyPCA) || *mergeOutput || *hgvsSingle {
log.Infof("%04d: preparing numpy (rows=%d, cols=%d)", infileIdx, len(cmd.cgnames), 2*outcol)
throttleNumpyMem.Acquire()
rows := len(cmd.cgnames)
cols := 2 * outcol
out := make([]int16, rows*cols)
for row, name := range cmd.cgnames {
- out := out[row*cols:]
- outcol := 0
+ outidx := row * cols
for col, v := range cgs[name].Variants {
tag := tagstart + tagID(col/2)
- if mask != nil && reftile[tag] == nil || tag > tagID(cmd.filter.MaxTag) {
+ if cmd.filter.MaxTag >= 0 && tag > tagID(cmd.filter.MaxTag) {
+ break
+ }
+ if rt := reftile[tag]; rt == nil || rt.excluded {
continue
}
- if variants, ok := seq[tag]; ok && len(variants) > int(v) && len(variants[v].Sequence) > 0 {
- out[outcol] = int16(variantRemap[tag-tagstart][v])
+ if v == 0 {
+ out[outidx] = 0 // tag not found / spanning tile
+ } else if variants, ok := seq[tag]; ok && int(v) < len(variants) && len(variants[v].Sequence) > 0 {
+ out[outidx] = int16(variantRemap[tag-tagstart][v])
} else {
- out[outcol] = -1
+ out[outidx] = -1 // low quality tile variant
+ }
+ if tag == cmd.debugTag {
+ log.Printf("tag %d row %d col %d outidx %d v %d out %d", tag, row, col, outidx, v, out[outidx])
}
- outcol++
+ outidx++
}
}
seq = nil
})
}
if err = throttleMem.Wait(); err != nil {
- return 1
+ return err
}
if *hgvsChunked {
}
err = encodeHGVS.Wait()
if err != nil {
- return 1
+ return err
}
for seqname := range refseq {
log.Infof("%s: reading hgvsCols from temp file", seqname)
f := tmpHGVSCols[seqname]
_, err = f.Seek(0, io.SeekStart)
if err != nil {
- return 1
+ return err
}
var hgvsCols hgvsColSet
dec := gob.NewDecoder(bufio.NewReaderSize(f, 1<<24))
err = dec.Decode(&hgvsCols)
}
if err != io.EOF {
- return 1
+ return err
}
log.Infof("%s: sorting %d hgvs variants", seqname, len(hgvsCols))
variants := make([]hgvs.Variant, 0, len(hgvsCols))
}
err = writeNumpyInt8(fmt.Sprintf("%s/hgvs.%s.npy", *outputDir, seqname), out, rows, cols)
if err != nil {
- return 1
+ return err
}
out = nil
}
err = ioutil.WriteFile(fnm, hgvsLabels.Bytes(), 0666)
if err != nil {
- return 1
+ return err
}
}
}
annoFilename := fmt.Sprintf("%s/matrix.annotations.csv", *outputDir)
annof, err = os.Create(annoFilename)
if err != nil {
- return 1
+ return err
}
annow = bufio.NewWriterSize(annof, 1<<20)
}
log.Infof("reading %s", annotationsFilename)
buf, err := os.ReadFile(annotationsFilename)
if err != nil {
- return 1
+ return err
}
if *mergeOutput {
err = os.Remove(annotationsFilename)
if err != nil {
- return 1
+ return err
}
}
for _, line := range bytes.Split(buf, []byte{'\n'}) {
rt, ok := reftile[tagID(tag)]
if !ok {
err = fmt.Errorf("bug: seeing annotations for tag %d, but it has no reftile entry", tag)
- return 1
+ return err
}
for ph := 0; ph < 2; ph++ {
for row := 0; row < rows; row++ {
if *mergeOutput {
err = annow.Flush()
if err != nil {
- return 1
+ return err
}
err = annof.Close()
if err != nil {
- return 1
+ return err
}
err = writeNumpyInt16(fmt.Sprintf("%s/matrix.npy", *outputDir), out, rows, cols)
if err != nil {
- return 1
+ return err
}
}
out = nil
}
err = writeNumpyInt16(fmt.Sprintf("%s/hgvs.npy", *outputDir), out, rows, cols)
if err != nil {
- return 1
+ return err
}
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 err
}
}
}
- if *onehotSingle {
+ if *onehotSingle || *onlyPCA {
nzCount := 0
for _, part := range onehotIndirect {
nzCount += len(part[0])
onehotXrefs[i] = nil
debug.FreeOSMemory()
}
- fnm := fmt.Sprintf("%s/onehot.npy", *outputDir)
- err = writeNumpyUint32(fnm, onehot, 2, nzCount)
+ if *onehotSingle {
+ fnm := fmt.Sprintf("%s/onehot.npy", *outputDir)
+ err = writeNumpyUint32(fnm, onehot, 2, nzCount)
+ if err != nil {
+ return err
+ }
+ fnm = fmt.Sprintf("%s/onehot-columns.npy", *outputDir)
+ err = writeNumpyInt32(fnm, onehotXref2int32(xrefs), 5, len(xrefs))
+ if err != nil {
+ return err
+ }
+ }
+ if *onlyPCA {
+ cols := 0
+ for _, c := range onehot[nzCount:] {
+ if int(c) >= cols {
+ cols = int(c) + 1
+ }
+ }
+ if cols == 0 {
+ return fmt.Errorf("cannot do PCA: one-hot matrix is empty")
+ }
+ log.Printf("have %d one-hot cols", cols)
+ stride := 1
+ for *maxPCATiles > 0 && cols > *maxPCATiles*2 {
+ cols = cols / 2
+ stride = stride * 2
+ }
+ log.Printf("creating matrix: %d rows, %d cols, stride %d", len(cmd.cgnames), cols, stride)
+ mtx := mat.NewDense(len(cmd.cgnames), cols, nil)
+ for i, c := range onehot[nzCount:] {
+ if int(c/2)%stride == 0 {
+ mtx.Set(int(onehot[i]), int(c/2)/stride*2+int(c)%2, 1)
+ }
+ }
+ log.Print("fitting")
+ transformer := nlp.NewPCA(*pcaComponents)
+ transformer.Fit(mtx.T())
+ log.Printf("transforming")
+ pca, err := transformer.Transform(mtx.T())
+ if err != nil {
+ return err
+ }
+ pca = pca.T()
+ outrows, outcols := pca.Dims()
+ log.Printf("copying result to numpy output array: %d rows, %d cols", outrows, outcols)
+ out := make([]float64, outrows*outcols)
+ for i := 0; i < outrows; i++ {
+ for j := 0; j < outcols; j++ {
+ out[i*outcols+j] = pca.At(i, j)
+ }
+ }
+ fnm := fmt.Sprintf("%s/pca.npy", *outputDir)
+ log.Printf("writing numpy: %s", fnm)
+ output, err := os.OpenFile(fnm, os.O_CREATE|os.O_TRUNC|os.O_WRONLY, 0777)
+ if err != nil {
+ return err
+ }
+ npw, err := gonpy.NewWriter(nopCloser{output})
+ if err != nil {
+ return fmt.Errorf("gonpy.NewWriter: %w", err)
+ }
+ npw.Shape = []int{outrows, outcols}
+ err = npw.WriteFloat64(out)
+ if err != nil {
+ return fmt.Errorf("WriteFloat64: %w", err)
+ }
+ err = output.Close()
+ if err != nil {
+ return err
+ }
+ log.Print("done")
+ }
+ }
+ if !*mergeOutput && !*onehotChunked && !*onehotSingle && !*onlyPCA {
+ tagoffsetFilename := *outputDir + "/chunk-tag-offset.csv"
+ log.Infof("writing tag offsets to %s", tagoffsetFilename)
+ var f *os.File
+ f, err = os.Create(tagoffsetFilename)
if err != nil {
- return 1
+ return err
}
- fnm = fmt.Sprintf("%s/onehot-columns.npy", *outputDir)
- err = writeNumpyInt32(fnm, onehotXref2int32(xrefs), 4, len(xrefs))
+ defer f.Close()
+ for idx, offset := range chunkStartTag {
+ _, err = fmt.Fprintf(f, "%q,%d\n", fmt.Sprintf("matrix.%04d.npy", idx), offset)
+ if err != nil {
+ err = fmt.Errorf("write %s: %w", tagoffsetFilename, err)
+ return err
+ }
+ }
+ err = f.Close()
if err != nil {
- return 1
+ err = fmt.Errorf("close %s: %w", tagoffsetFilename, err)
+ return err
}
}
- return 0
+ return nil
}
// Read case/control files, remove non-case/control entries from
type onehotXref struct {
tag tagID
variant tileVariantID
- het bool
+ hom bool
pvalue float64
}
const onehotXrefSize = unsafe.Sizeof(onehotXref{})
-// Build onehot matrix (m[variant*2+isHet][genome] == 0 or 1) for all
+// Build onehot matrix (m[tileVariantIndex][genome] == 0 or 1) for all
// variants of a single tile/tag#.
//
// Return nil if no tile variant passes Χ² filter.
-func (cmd *sliceNumpy) tv2homhet(cgs map[string]CompactGenome, maxv tileVariantID, remap []tileVariantID, tag, chunkstarttag tagID) ([][]int8, []onehotXref) {
- if maxv < 2 {
- // everyone has the most common variant
+func (cmd *sliceNumpy) tv2homhet(cgs map[string]CompactGenome, maxv tileVariantID, remap []tileVariantID, tag, chunkstarttag tagID, seq map[tagID][]TileVariant) ([][]int8, []onehotXref) {
+ if tag == cmd.debugTag {
+ tv := make([]tileVariantID, len(cmd.cgnames)*2)
+ for i, name := range cmd.cgnames {
+ copy(tv[i*2:(i+1)*2], cgs[name].Variants[(tag-chunkstarttag)*2:])
+ }
+ log.WithFields(logrus.Fields{
+ "cgs[i].Variants[tag*2+j]": tv,
+ "maxv": maxv,
+ "remap": remap,
+ "tag": tag,
+ "chunkstarttag": chunkstarttag,
+ }).Info("tv2homhet()")
+ }
+ if maxv < 1 || (maxv < 2 && !cmd.includeVariant1) {
+ // everyone has the most common variant (of the variants we don't drop)
return nil, nil
}
tagoffset := tag - chunkstarttag
coverage := 0
for _, cg := range cgs {
- if cg.Variants[tagoffset*2] > 0 && cg.Variants[tagoffset*2+1] > 0 {
+ alleles := 0
+ for _, v := range cg.Variants[tagoffset*2 : tagoffset*2+2] {
+ if v > 0 && int(v) < len(seq[tag]) && len(seq[tag][v].Sequence) > 0 {
+ alleles++
+ }
+ }
+ if alleles == 2 {
coverage++
}
}
obs[i] = make([]bool, len(cmd.cgnames))
}
for cgid, name := range cmd.cgnames {
- cgvars := cgs[name].Variants
- for v := tileVariantID(2); v <= maxv; v++ {
- if remap[cgvars[tagoffset*2]] == v && remap[cgvars[tagoffset*2+1]] == v {
+ cgvars := cgs[name].Variants[tagoffset*2:]
+ tv0, tv1 := remap[cgvars[0]], remap[cgvars[1]]
+ for v := tileVariantID(1); v <= maxv; v++ {
+ if tv0 == v && tv1 == v {
obs[v*2][cgid] = true
- } else if remap[cgvars[tagoffset*2]] == v || remap[cgvars[tagoffset*2+1]] == v {
+ } else if tv0 == v || tv1 == v {
obs[v*2+1][cgid] = true
}
}
}
var onehot [][]int8
var xref []onehotXref
- for homcol := 2; homcol < len(obs); homcol += 2 {
- // homcol 0,1 correspond to tile variant 0, i.e.,
- // no-call; homcol 2,3 correspond to the most common
- // variant; so we (normally) start at homcol 4.
- if homcol < 4 && !cmd.includeVariant1 {
+ for col := 2; col < len(obs); col++ {
+ // col 0,1 correspond to tile variant 0, i.e.,
+ // no-call; col 2,3 correspond to the most common
+ // variant; so we (normally) start at col 4.
+ if col < 4 && !cmd.includeVariant1 {
continue
}
- for het := 0; het < 2; het++ {
- p := pvalue(obs[homcol+het], cmd.chi2Cases)
- if cmd.chi2PValue < 1 && !(p < cmd.chi2PValue) {
- continue
- }
- onehot = append(onehot, bool2int8(obs[homcol+het]))
- xref = append(xref, onehotXref{
- tag: tag,
- variant: tileVariantID(homcol / 2),
- het: het == 1,
- pvalue: p,
- })
+ p := pvalue(obs[col], cmd.chi2Cases)
+ if cmd.chi2PValue < 1 && !(p < cmd.chi2PValue) {
+ continue
}
+ onehot = append(onehot, bool2int8(obs[col]))
+ xref = append(xref, onehotXref{
+ tag: tag,
+ variant: tileVariantID(col >> 1),
+ hom: col&1 == 0,
+ pvalue: p,
+ })
}
return onehot, xref
}
// P-value row contains 1000000x actual p-value.
func onehotXref2int32(xrefs []onehotXref) []int32 {
xcols := len(xrefs)
- xdata := make([]int32, 4*xcols)
+ xdata := make([]int32, 5*xcols)
for i, xref := range xrefs {
xdata[i] = int32(xref.tag)
xdata[xcols+i] = int32(xref.variant)
- if xref.het {
+ if xref.hom {
xdata[xcols*2+i] = 1
}
xdata[xcols*3+i] = int32(xref.pvalue * 1000000)
+ xdata[xcols*4+i] = int32(-math.Log10(xref.pvalue) * 1000000)
}
return xdata
}
projects / lightning.git / blobdiff