Skip to content

Instantly share code, notes, and snippets.

What would you like to do?
This is a python script which processes a SeqInterval file and Jaspar matrix file to find all occurrences of the TF binding sites.
Searches through a fasta sequence for the relevant Position Weight Matrices
from the Jaspar Database.
from __future__ import division
from optparse import OptionParser
from itertools import groupby, izip, count, imap, islice
import re, csv
import MOODS
from collections import defaultdict
def isheader(line):
return line.startswith('>')
def PWMIter(handle):
"""Yields name/PWM"""
num_re = re.compile('\d{1,}')
line_gen = imap(lambda x: x.strip(), handle)
name = None
out = None
for header, lines in groupby(line_gen, isheader):
if header:
name =[1:]
out = []
for line in lines:
out.append(map(int, num_re.findall(line)))
yield name, out
def transpose(input):
"""transposes a list-of-lists"""
return map(None, *input)
def ParseHeader(header):
"""Parses the structured header to retrieve relevant info"""
parts = header.split('_')
odict = {}
odict['CHROM'] = parts[1]
odict['START'] = int(parts[2])
odict['STRAND'] = parts[4]
return odict
def ProcessSeqs(SEQ_HANDLE, PWMS, THRESHOLD, WANT_REV = False, bg = None):
"""Yields matches on sequences in an 'interval' formatted dictionary"""
pwm_names = map(lambda x: x[0], PWMS)
pwm_mats = map(lambda x: x[1], PWMS)
thresh = map(lambda x:MOODS.threshold_from_p(x, bg, THRESHOLD), pwm_mats)
for interval in ReadInterval(SEQ_HANDLE):
print interval['NAME']
results =['SEQ'].upper(), pwm_mats, thresh,
both_strands = WANT_REV, algorithm = 'lf',
absolute_threshold = True, bg = bg)
for res, pwm_name, pwm_mat, th in zip(results, pwm_names, pwm_mats, thresh):
width = len(pwm_mat[0])
for position, score in res:
if score > th:
yield {
print 'got bad result'
def ReadInterval(handle):
"""Reads an Interval file and returns a list of dicts for each row"""
headers = ('CHROM', 'START', 'END', 'NAME', 'junk', 'STRAND',
'junk7', 'junk8', 'junk9', 'junk10', 'junk11',
'junk12', 'SEQ')
for row in csv.DictReader(handle, fieldnames = headers,
delimiter = '\t'):
yield row
def GetBG(handle):
d = defaultdict(int)
for row in islice(ReadInterval(handle), 10):
for r in row['SEQ'].upper():
d[r] += 1
s = sum(d.values())
return [d['A']/s, d['C']/s, d['G']/s, d['T']/s]
if __name__ == '__main__':
parser = OptionParser()
(options, args) = parser.parse_args()
seqintervalfile = args[0]
jasparfile = args[1]
threshold = args[2]
output_file = args[3]
print 'Getting Background'
with open(jasparfile) as handle:
PWMS = list(PWMIter(handle))
with open(seqintervalfile) as handle:
bg = GetBG(handle)
fields = ('CHROM', 'START', 'END', 'STRAND', 'NAME', 'PWM', 'SCORE', 'SEQ')
with open(output_file, 'w') as handle:
handle.write('#'+'\t'.join(fields) + '\n')
with open(seqintervalfile) as f_handle:
writer = csv.DictWriter(handle, fields, )
for row in ProcessSeqs(f_handle, PWMS, float(threshold), bg = bg):

This comment has been minimized.

Copy link

commented Dec 15, 2014

I'm confused: does this take a fasta-file (as stated in the comments) or a seqinterval-file as input? And what do you define as a seqinterval-file. Thanks.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
You can’t perform that action at this time.