// structureToSequence uses structural fragments to categorize a segment
// of alpha-carbon atoms, and adds the corresponding residues to a
// corresponding sequence fragment.
func structureToSequence(
lib fragbag.StructureLibrary,
chain *pdb.Chain,
nullChan chan seq.Sequence,
seqChans []chan seq.Sequence,
) {
sequence := chain.AsSequence()
fragSize := lib.FragmentSize()
// If the chain is shorter than the fragment size, we can do nothing
// with it.
if sequence.Len() < fragSize {
util.Verbosef("Sequence '%s' is too short (length: %d)",
sequence.Name, sequence.Len())
return
}
// If we're accumulating a null model, add this sequence to it.
if nullChan != nil {
nullChan <- sequence
}
// This bit of trickery here is all about getting the call to
// SequenceCaAtoms outside of the loop. In particular, it's a very
// expensive call since it has to reconcile inconsistencies between
// SEQRES and ATOM records in PDB files.
limit := sequence.Len() - fragSize
atoms := chain.SequenceCaAtoms()
atomSlice := make([]structure.Coords, fragSize)
noGaps := func(atoms []*structure.Coords) []structure.Coords {
for i, atom := range atoms {
if atom == nil {
return nil
}
atomSlice[i] = *atom
}
return atomSlice
}
for start := 0; start <= limit; start++ {
end := start + fragSize
cas := noGaps(atoms[start:end])
if cas == nil {
// Nothing contiguous was found (a "disordered" residue perhaps).
// So skip this part of the chain.
continue
}
bestFrag := lib.BestStructureFragment(atomSlice)
sliced := sequence.Slice(start, end)
seqChans[bestFrag] <- sliced
}
}
func mkSeqHMM(c *command) {
c.assertLeastNArg(3)
structLib := util.StructureLibrary(c.flags.Arg(0))
outPath := c.flags.Arg(1)
entries := c.flags.Args()[2:]
util.AssertOverwritable(outPath, flagOverwrite)
saveto := util.CreateFile(outPath)
// Stores intermediate files produced by hhmake.
tempDir, err := ioutil.TempDir("", "mk-seqlib-hmm")
util.Assert(err, "Could not create temporary directory.")
defer os.RemoveAll(tempDir)
// Initialize a MSA for each structural fragment.
var msas []seq.MSA
var msaChans []chan seq.Sequence
for i := 0; i < structLib.Size(); i++ {
msa := seq.NewMSA()
msa.SetLen(structLib.FragmentSize())
msas = append(msas, msa)
msaChans = append(msaChans, make(chan seq.Sequence))
}
// Now spin up a goroutine for each fragment that is responsible for
// adding a sequence slice to itself.
for i := 0; i < structLib.Size(); i++ {
addToMSA(msaChans[i], &msas[i])
}
// Create a channel that sends the PDB entries given.
entryChan := make(chan string)
go func() {
for _, fp := range entries {
entryChan <- fp
}
close(entryChan)
}()
progress := util.NewProgress(len(entries))
for i := 0; i < flagCpu; i++ {
wgPDBChains.Add(1)
go func() {
for entryPath := range entryChan {
_, chains, err := util.PDBOpen(entryPath)
progress.JobDone(err)
if err != nil {
continue
}
for _, chain := range chains {
structureToSequence(structLib, chain, nil, msaChans)
}
}
wgPDBChains.Done()
}()
}
wgPDBChains.Wait()
progress.Close()
// We've finishing reading all the PDB inputs. Now close the channels
// and let the sequence fragments finish.
for i := 0; i < structLib.Size(); i++ {
close(msaChans[i])
}
wgSeqFragments.Wait()
util.Verbosef("Building profile HMMs from MSAs...")
// Finally, add the sequence fragments to a new sequence fragment
// library and save.
hmms := make([]*seq.HMM, structLib.Size())
hhmake := func(i int) struct{} {
fname := path.Join(tempDir, fmt.Sprintf("%d.fasta", i))
f := util.CreateFile(fname)
util.Assert(msa.WriteFasta(f, msas[i]))
hhm, err := hhsuite.HHMakePseudo.Run(fname)
util.Assert(err)
hmms[i] = hhm.HMM
return struct{}{} // my unifier sucks, i guess
}
fun.ParMap(hhmake, fun.Range(0, structLib.Size()))
lib, err := fragbag.NewSequenceHMM(structLib.Name(), hmms)
util.Assert(err)
util.Assert(fragbag.Save(saveto, lib))
}
func main() {
outDir := util.Arg(0)
fasInps := util.Args()[1:]
util.Assert(os.MkdirAll(outDir, 0777))
fastaChan := make(chan string)
wg := new(sync.WaitGroup)
for i := 0; i < max(1, runtime.GOMAXPROCS(0)); i++ {
go func() {
wg.Add(1)
for fasta := range fastaChan {
util.Verbosef("Computing map for '%s'...", fasta)
fmap := util.GetFmap(fasta)
outF := path.Join(outDir, fmt.Sprintf("%s.fmap", fmap.Name))
util.FmapWrite(util.CreateFile(outF), fmap)
}
wg.Done()
}()
}
for _, fasta := range fasInps {
fastaChan <- fasta
}
close(fastaChan)
wg.Wait()
}