Mercurial > repos > davidvanzessen > shm_csr
view gene_identification.py @ 30:33a7c49d48a7 draft
Uploaded
author | davidvanzessen |
---|---|
date | Mon, 12 Dec 2016 05:19:58 -0500 |
parents | c0ecc31829bd |
children |
line wrap: on
line source
import re import argparse import time starttime= int(time.time() * 1000) parser = argparse.ArgumentParser() parser.add_argument("--input", help="The 1_Summary file from an IMGT zip file") parser.add_argument("--output", help="The annotated output file to be merged back with the summary file") args = parser.parse_args() infile = args.input #infile = "test_VH-Ca_Cg_25nt/1_Summary_test_VH-Ca_Cg_25nt_241013.txt" output = args.output #outfile = "identified.txt" dic = dict() total = 0 first = True IDIndex = 0 seqIndex = 0 with open(infile, 'r') as f: #read all sequences into a dictionary as key = ID, value = sequence for line in f: total += 1 linesplt = line.split("\t") if first: print "linesplt", linesplt IDIndex = linesplt.index("Sequence ID") seqIndex = linesplt.index("Sequence") first = False continue ID = linesplt[IDIndex] if len(linesplt) < 28: #weird rows without a sequence dic[ID] = "" else: dic[ID] = linesplt[seqIndex] print "Number of input sequences:", len(dic) #old cm sequence: gggagtgcatccgccccaacccttttccccctcgtctcctgtgagaattccc #old cg sequence: ctccaccaagggcccatcggtcttccccctggcaccctcctccaagagcacctctgggggcacagcggccctgggctgcctggtcaaggactacttccccgaaccggtgacggtgtcgtggaactcaggcgccctgaccag #lambda/kappa reference sequence searchstrings = {"ca": "catccccgaccagccccaaggtcttcccgctgagcctctgcagcacccagccagatgggaacgtggtcatcgcctgcctgg", "cg": "ctccaccaagggcccatcggtcttccccctggcaccctcctccaagagcacctctgggggcacagcggcc", "ce": "gcctccacacagagcccatccgtcttccccttgacccgctgctgcaaaaacattccctcc", "cm": "gggagtgcatccgccccaacc"} #new (shorter) cm sequence compiledregex = {"ca": [], "cg": [], "ce": [], "cm": []} #lambda/kappa reference sequence variable nucleotides ca1 = {38: 't', 39: 'g', 48: 'a', 49: 'g', 51: 'c', 68: 'a', 73: 'c'} ca2 = {38: 'g', 39: 'a', 48: 'c', 49: 'c', 51: 'a', 68: 'g', 73: 'a'} cg1 = {0: 'c', 33: 'a', 38: 'c', 44: 'a', 54: 't', 56: 'g', 58: 'g', 66: 'g', 132: 'c'} cg2 = {0: 'c', 33: 'g', 38: 'g', 44: 'g', 54: 'c', 56: 'a', 58: 'a', 66: 'g', 132: 't'} cg3 = {0: 't', 33: 'g', 38: 'g', 44: 'g', 54: 't', 56: 'g', 58: 'g', 66: 'g', 132: 'c'} cg4 = {0: 't', 33: 'g', 38: 'g', 44: 'g', 54: 'c', 56: 'a', 58: 'a', 66: 'c', 132: 'c'} #remove last snp for shorter cg sequence --- note, also change varsInCG del cg1[132] del cg2[132] del cg3[132] del cg4[132] #reference sequences are cut into smaller parts of 'chunklength' length, and with 'chunklength' / 2 overlap chunklength = 8 #create the chunks of the reference sequence with regular expressions for the variable nucleotides for i in range(0, len(searchstrings["ca"]) - chunklength, chunklength / 2): pos = i chunk = searchstrings["ca"][i:i+chunklength] result = "" varsInResult = 0 for c in chunk: if pos in ca1.keys(): varsInResult += 1 result += "[" + ca1[pos] + ca2[pos] + "]" else: result += c pos += 1 compiledregex["ca"].append((re.compile(result), varsInResult)) for i in range(0, len(searchstrings["cg"]) - chunklength, chunklength / 2): pos = i chunk = searchstrings["cg"][i:i+chunklength] result = "" varsInResult = 0 for c in chunk: if pos in cg1.keys(): varsInResult += 1 result += "[" + "".join(set([cg1[pos], cg2[pos], cg3[pos], cg4[pos]])) + "]" else: result += c pos += 1 compiledregex["cg"].append((re.compile(result), varsInResult)) for i in range(0, len(searchstrings["cm"]) - chunklength, chunklength / 2): compiledregex["cm"].append((re.compile(searchstrings["cm"][i:i+chunklength]), False)) for i in range(0, len(searchstrings["ce"]) - chunklength + 1, chunklength / 2): compiledregex["ce"].append((re.compile(searchstrings["ce"][i:i+chunklength]), False)) def removeAndReturnMaxIndex(x): #simplifies a list comprehension m = max(x) index = x.index(m) x[index] = 0 return index start_location = dict() hits = dict() alltotal = 0 for key in compiledregex.keys(): #for ca/cg/cm/ce regularexpressions = compiledregex[key] #get the compiled regular expressions for ID in dic.keys()[0:]: #for every ID if ID not in hits.keys(): #ensure that the dictionairy that keeps track of the hits for every gene exists hits[ID] = {"ca_hits": 0, "cg_hits": 0, "cm_hits": 0, "ce_hits": 0, "ca1": 0, "ca2": 0, "cg1": 0, "cg2": 0, "cg3": 0, "cg4": 0} currentIDHits = hits[ID] seq = dic[ID] lastindex = 0 start_zero = len(searchstrings[key]) #allows the reference sequence to start before search sequence (start_locations of < 0) start = [0] * (len(seq) + start_zero) for i, regexp in enumerate(regularexpressions): #for every regular expression relativeStartLocation = lastindex - (chunklength / 2) * i if relativeStartLocation >= len(seq): break regex, hasVar = regexp matches = regex.finditer(seq[lastindex:]) for match in matches: #for every match with the current regex, only uses the first hit because of the break at the end of this loop lastindex += match.start() start[relativeStartLocation + start_zero] += 1 if hasVar: #if the regex has a variable nt in it chunkstart = chunklength / 2 * i #where in the reference does this chunk start chunkend = chunklength / 2 * i + chunklength #where in the reference does this chunk end if key == "ca": #just calculate the variable nt score for 'ca', cheaper currentIDHits["ca1"] += len([1 for x in ca1 if chunkstart <= x < chunkend and ca1[x] == seq[lastindex + x - chunkstart]]) currentIDHits["ca2"] += len([1 for x in ca2 if chunkstart <= x < chunkend and ca2[x] == seq[lastindex + x - chunkstart]]) elif key == "cg": #just calculate the variable nt score for 'cg', cheaper currentIDHits["cg1"] += len([1 for x in cg1 if chunkstart <= x < chunkend and cg1[x] == seq[lastindex + x - chunkstart]]) currentIDHits["cg2"] += len([1 for x in cg2 if chunkstart <= x < chunkend and cg2[x] == seq[lastindex + x - chunkstart]]) currentIDHits["cg3"] += len([1 for x in cg3 if chunkstart <= x < chunkend and cg3[x] == seq[lastindex + x - chunkstart]]) currentIDHits["cg4"] += len([1 for x in cg4 if chunkstart <= x < chunkend and cg4[x] == seq[lastindex + x - chunkstart]]) else: #key == "cm" #no variable regions in 'cm' or 'ce' pass break #this only breaks when there was a match with the regex, breaking means the 'else:' clause is skipped else: #only runs if there were no hits continue #print "found ", regex.pattern , "at", lastindex, "adding one to", (lastindex - chunklength / 2 * i), "to the start array of", ID, "gene", key, "it's now:", start[lastindex - chunklength / 2 * i] currentIDHits[key + "_hits"] += 1 start_location[ID + "_" + key] = str([(removeAndReturnMaxIndex(start) + 1 - start_zero) for x in range(5) if len(start) > 0 and max(start) > 1]) #start_location[ID + "_" + key] = str(start.index(max(start))) varsInCA = float(len(ca1.keys()) * 2) varsInCG = float(len(cg1.keys()) * 2) - 2 # -2 because the sliding window doesn't hit the first and last nt twice varsInCM = 0 varsInCE = 0 def round_int(val): return int(round(val)) first = True seq_write_count=0 with open(infile, 'r') as f: #read all sequences into a dictionary as key = ID, value = sequence with open(output, 'w') as o: for line in f: total += 1 if first: o.write("Sequence ID\tbest_match\tnt_hit_percentage\tchunk_hit_percentage\tstart_locations\n") first = False continue linesplt = line.split("\t") if linesplt[2] == "No results": pass ID = linesplt[1] currentIDHits = hits[ID] possibleca = float(len(compiledregex["ca"])) possiblecg = float(len(compiledregex["cg"])) possiblecm = float(len(compiledregex["cm"])) possiblece = float(len(compiledregex["ce"])) cahits = currentIDHits["ca_hits"] cghits = currentIDHits["cg_hits"] cmhits = currentIDHits["cm_hits"] cehits = currentIDHits["ce_hits"] if cahits >= cghits and cahits >= cmhits and cahits >= cehits: #its a ca gene ca1hits = currentIDHits["ca1"] ca2hits = currentIDHits["ca2"] if ca1hits >= ca2hits: o.write(ID + "\tIGA1\t" + str(round_int(ca1hits / varsInCA * 100)) + "\t" + str(round_int(cahits / possibleca * 100)) + "\t" + start_location[ID + "_ca"] + "\n") else: o.write(ID + "\tIGA2\t" + str(round_int(ca2hits / varsInCA * 100)) + "\t" + str(round_int(cahits / possibleca * 100)) + "\t" + start_location[ID + "_ca"] + "\n") elif cghits >= cahits and cghits >= cmhits and cghits >= cehits: #its a cg gene cg1hits = currentIDHits["cg1"] cg2hits = currentIDHits["cg2"] cg3hits = currentIDHits["cg3"] cg4hits = currentIDHits["cg4"] if cg1hits >= cg2hits and cg1hits >= cg3hits and cg1hits >= cg4hits: #cg1 gene o.write(ID + "\tIGG1\t" + str(round_int(cg1hits / varsInCG * 100)) + "\t" + str(round_int(cghits / possiblecg * 100)) + "\t" + start_location[ID + "_cg"] + "\n") elif cg2hits >= cg1hits and cg2hits >= cg3hits and cg2hits >= cg4hits: #cg2 gene o.write(ID + "\tIGG2\t" + str(round_int(cg2hits / varsInCG * 100)) + "\t" + str(round_int(cghits / possiblecg * 100)) + "\t" + start_location[ID + "_cg"] + "\n") elif cg3hits >= cg1hits and cg3hits >= cg2hits and cg3hits >= cg4hits: #cg3 gene o.write(ID + "\tIGG3\t" + str(round_int(cg3hits / varsInCG * 100)) + "\t" + str(round_int(cghits / possiblecg * 100)) + "\t" + start_location[ID + "_cg"] + "\n") else: #cg4 gene o.write(ID + "\tIGG4\t" + str(round_int(cg4hits / varsInCG * 100)) + "\t" + str(round_int(cghits / possiblecg * 100)) + "\t" + start_location[ID + "_cg"] + "\n") else: #its a cm or ce gene if cmhits >= cehits: o.write(ID + "\tIGM\t100\t" + str(round_int(cmhits / possiblecm * 100)) + "\t" + start_location[ID + "_cm"] + "\n") else: o.write(ID + "\tIGE\t100\t" + str(round_int(cehits / possiblece * 100)) + "\t" + start_location[ID + "_ce"] + "\n") seq_write_count += 1 print "Time: %i" % (int(time.time() * 1000) - starttime) print "Number of sequences written to file:", seq_write_count