Convert DeepSEA training sequences a BED file

seqs2bed.py

import h5py

# HDF5 file with two arrays: 'trainxdata' (samples) and 'traindata' (labels)
INFILE_SAMPLES = ''
INFILE_REFERENCE_FASTA = ''
OUTFILE_FASTA = 'deepsea_train10k.fa'
OUTFILE_BED = 'deepsea_train10k.bed'

def onehot2base(onehot):
  if onehot == [1,0,0,0]:
    return 'A'
  elif onehot == [0,1,0,0]:
    return 'G'
  elif onehot == [0,0,1,0]:
    return 'C'
  elif onehot == [0,0,0,1]:
    return 'T'
  elif onehot == [0,0,0,0]:
    return 'N'
  else:
    return 'U'

training_data_file = h5py.File(INFILE_SAMPLES, 'r')
samples_onehot = training_data_file['trainxdata']
samples_fasta = '\n'.join(
  ['>seq' + str(i) + '\n' +
   ''.join(map(lambda x: onehot2base(x), samples_onehot[:,:,i].tolist()))
   for i in samples_onehot.shape[2]]
  )

with open(OUTFILE_FASTA, 'w') as f:
  f.write(samples_fasta)

# Call bwa mem hg19.fasta OUTFILE_FASTA
# Use bedtools bamtobed on BWA MEM output