pseudo.py

def deepSv_predict(reads, reference, representation, model):
  stats = generate_read_stats(reads) # we grab std and mean for insert and read size
  candidates = paired_read_analysis(reads, stats) # generate candidate variants
  
  calls = set()
  for (variant, supporting_reads) in candidates:
    proba = model.predict(representation.generate(variant, supporting_reads))
    if proba > 0.5:
      calls.add((variant, supporting_reads))
      
  refined_calls = set()
  for (variant, supporting_reads) in refined_calls:
    located_call = split_read_location(reads, variant, reference)
    if located_call is None:
      located_call = supporting_reads_bounding(supporting_reads, variant)
    refined_calls.add(located_call)
  
  return format_to_vcf(refined_calls)