view Contra/baseline.py @ 15:f4345d10e1ad

Uploaded
author fcaramia
date Tue, 11 Dec 2012 18:52:28 -0500
parents 7564f3b1e675
children
line wrap: on
line source

#!/usr/bin/python

import os
import sys
import fnmatch
import shlex
import subprocess
import shutil
import math
from optparse import OptionParser
from scripts.split_chromosome import splitByChromosome3 as splitByChromosome
from scripts.get_chr_length import *
from scripts.convert_targeted_regions import *
from scripts.convert_gene_coordinate import convertGeneCoordinate2 as convertGeneCoordinate
from multiprocessing import Pool

class Params:
	"""
	Class for top-level system parameters:
	"""

	def __init__(self):

		def mult_files(option, opt_str, value, parser):
			args=[]
			for arg in parser.rargs:
				if arg[0] != "-":
					args.append(arg)
				else:
					del parser.rargs[:len(args)]
					break
			if getattr(parser.values, option.dest):
				args.extend(getattr(parser.values, option.dest))
			setattr(parser.values, option.dest, args)


		#command-line option definition
		self.parser = OptionParser()
		self.parser.add_option("-t", "--target",
			help="Target region definition file [REQUIRED] [BED Format]",
			action="store", type="string", dest="target")
		self.parser.add_option("-f", "--files",
			help="Files to be converted to baselines [REQUIRED] [BAM]",
			action="callback", callback=mult_files, dest="files")
		self.parser.add_option("-o", "--output",
			help="Output folder [REQUIRED]",
			action="store", type="string", dest="output")
		self.parser.add_option("-c","--trim",
			help="Portion of outliers to be removed before calculating average [Default: 0.2]",
			action="store", dest="trim", default="0.2")
		self.parser.add_option("-n","--name",
			help="Output baseline name [Default: baseline]",
			action="store", dest="name", default="baseline")

		# required parameters list:
		self.ERRORLIST = []

		#change system parameters based on any command line arguments
		(options, args) = self.parser.parse_args()
		if options.target:
			self.TARGET=options.target
		else:
			self.ERRORLIST.append("target")

		if options.files:
			self.FILES=options.files
		else:
			self.ERRORLIST.append("files")

		if options.output:
			self.OUTPUT=options.output
		else:
			self.ERRORLIST.append("output")

		if options.trim:
			self.TRIM=float(options.trim)
		
		if options.name:
			self.NAME=options.name

		if len(self.ERRORLIST) != 0:
			self.parser.print_help()
			self.parser.error("Missing required parameters")


	def repeat(self):
		#params test
		print "target	:", self.TARGET
		print "files	:", self.FILES
		print "output	:", self.OUTPUT
		print "trim	:", self.TRIM
		print "name	:", self.NAME


# option handling
params = Params() 
#params.repeat()
targetFile	= params.TARGET
infiles		= params.FILES
output_dir	= params.OUTPUT

#Debug
print " ------ baseline.py ------- "
print "Target:", targetFile
for files in infiles:
	print "File:", files
print "Output Directory: ", output_dir

def make_new_directory(outdir):
	print outdir
	if not os.path.exists(outdir):
		os.mkdir(outdir)

print " ----- creating output directory -----"
make_new_directory(output_dir)
outdir	= os.path.join(output_dir, "buf")
make_new_directory(outdir)

targetFile2 = os.path.join(outdir, os.path.basename(targetFile)+".sorted")
os.system("sort -k1,1 -k2n %s > %s" %(targetFile,targetFile2))

def processInFile(infile):
	infilename = os.path.basename(infile)
	s_outdir = os.path.join(outdir, infilename)
	make_new_directory(s_outdir)
        genomeFile = os.path.join(s_outdir,infilename +'.chrsummary')
        get_genome(infile, genomeFile)

        bedgraph = os.path.join(s_outdir, infilename + ".BEDGRAPH")
        args = shlex.split("genomeCoverageBed -ibam %s -bga -g %s" %(infile, genomeFile))
        iOutFile = open(bedgraph, "w")
        output = subprocess.Popen(args, stdout = iOutFile).wait()
        iOutFile.close()

        targetList = convertTarget(targetFile2)

        splitByChromosome(bedgraph)

        convertGeneCoordinate(targetList, os.path.dirname(bedgraph)+"/")
        bedgraph_tgtonly = bedgraph+".TARGETONLY"
        bedgraph_tgtonly_avg = bedgraph+".TARGETONLY.AVERAGE"
        os.rename(os.path.join(os.path.dirname(bedgraph),"geneRefCoverage.txt"),bedgraph_tgtonly)
        os.rename(os.path.join(os.path.dirname(bedgraph),"geneRefCoverage_targetAverage.txt"),bedgraph_tgtonly_avg)
	shutil.copy(bedgraph_tgtonly,outdir)
	shutil.copy(bedgraph_tgtonly_avg,outdir)

print "----- Processing Files -----"
pool = Pool(5)
pool.map(processInFile,infiles)


# STEP 2 -> Create Union BedGraph
allfiles_names 	= [x for x in os.listdir(outdir) if x.endswith("TARGETONLY")]
allfiles_path 	= [os.path.join(outdir,x) for x in allfiles_names]

args=["unionBedGraphs","-header","-i"]+allfiles_path+["-names"]+allfiles_names

print (str(args))
fo	= os.path.join(outdir,"TARGETONLY.union.txt")
foh	= open(fo,"w")

subprocess.Popen(args,stdout=foh).wait()
foh.close()

# STEP 3 -> POOL METHOD
TRIM	= params.TRIM #TRIM 	= 0.2
f	= fo
fh	= open(f)

fo=os.path.join(output_dir, params.NAME+".pooled2_TRIM"+str(TRIM)+".txt")
fo_libsize=fo+".LIBSIZE"


foh=open(fo,'w')
fo_libsize_h=open(fo_libsize,'w')


fh.readline()

Ns = None
nSamples = None

lineCount=0
for line in fh:
        lineCount+=1
        line_elems=line.rstrip().split("\t")
        rangeLen=int(line_elems[2])-int(line_elems[1])
        if not Ns:
                nSamples=len(line_elems)-3
                Ns = [0 for x in range(nSamples)]
        for i in range(nSamples):
                ind=i+3
                Ns[i] += int(line_elems[ind])*rangeLen


fh.close()
fh=open(f)
fh.readline()


lineCount=0
nSamples_exclude = int(math.floor(TRIM*nSamples))

def gm_mean(xs):
        tmpprod = 1
        p = 1.0/len(xs)
        for x in xs:
                tmpprod = tmpprod * math.pow(x,p)
        return tmpprod

Ns_gmean = gm_mean(Ns)

def meanstdv(x):
        from math import sqrt
        n, mean, std = len(x), 0, 0
        for a in x:
                mean = mean+a
        mean = mean / float(n)
        for a in x:
                std = std+(a-mean)**2
        std=sqrt(std/float(n-1))
        return mean, std


libsize = 0
for line in fh:
        lineCount+=1
        line_elems=line.rstrip().split("\t")
        rangeLen=int(line_elems[2])-int(line_elems[1])
        xs_tmp=[int(x) for x in line_elems[3:]]
        xs = [float(xs_tmp[i])*Ns_gmean/Ns[i] for i in range(len(xs_tmp))]
        xs.sort()
        xs_trimmed=xs[nSamples_exclude:(nSamples-nSamples_exclude)]
	#trimmed_mean=sum(xs_trimmed)/float(len(xs_trimmed))
        trimmed_mean,std=meanstdv(xs_trimmed)
        libsize+=rangeLen*trimmed_mean
	foh.write("\t".join(line_elems[0:3]+[str(trimmed_mean),str(std)])+"\n")


fo_libsize_h.write(str(int(round(libsize))))

fh.close()
foh.close()
fo_libsize_h.close()

def removeTempFolder(tempFolderPath):
	import shutil

	shutil.rmtree(tempFolderPath)
	print "Temp Folder Removed"

# Removed Temp Folder
removeTempFolder(outdir)