Mercurial > repos > yufei-luo > s_mart
view SMART/Java/Python/GetDistribution.py @ 68:85e80c21b1f7 draft
Uploaded
author | m-zytnicki |
---|---|
date | Mon, 16 Nov 2015 12:00:32 -0500 |
parents | 169d364ddd91 |
children |
line wrap: on
line source
#! /usr/bin/env python # # Copyright INRA-URGI 2009-2012 # # This software is governed by the CeCILL license under French law and # abiding by the rules of distribution of free software. You can use, # modify and/ or redistribute the software under the terms of the CeCILL # license as circulated by CEA, CNRS and INRIA at the following URL # "http://www.cecill.info". # # As a counterpart to the access to the source code and rights to copy, # modify and redistribute granted by the license, users are provided only # with a limited warranty and the software's author, the holder of the # economic rights, and the successive licensors have only limited # liability. # # In this respect, the user's attention is drawn to the risks associated # with loading, using, modifying and/or developing or reproducing the # software by the user in light of its specific status of free software, # that may mean that it is complicated to manipulate, and that also # therefore means that it is reserved for developers and experienced # professionals having in-depth computer knowledge. Users are therefore # encouraged to load and test the software's suitability as regards their # requirements in conditions enabling the security of their systems and/or # data to be ensured and, more generally, to use and operate it in the # same conditions as regards security. # # The fact that you are presently reading this means that you have had # knowledge of the CeCILL license and that you accept its terms. # import os from optparse import OptionParser from commons.core.parsing.ParserChooser import ParserChooser from commons.core.parsing.FastaParser import FastaParser from SMART.Java.Python.structure.Transcript import Transcript from commons.core.writer.Gff3Writer import Gff3Writer from SMART.Java.Python.misc.RPlotter import RPlotter from SMART.Java.Python.misc.MultipleRPlotter import MultipleRPlotter from SMART.Java.Python.misc.UnlimitedProgress import UnlimitedProgress from SMART.Java.Python.misc.Progress import Progress TWOSTRANDS = {True: [1, -1], False: [0]} STRANDTOSTR = {1: "(+)", -1: "(-)", 0: ""} class GetDistribution(object): def __init__(self, verbosity): self.verbosity = verbosity self.sizes = None self.nbBins = None self.sliceSize = None self.twoStrands = False self.start = 1 self.names = ["nbElements"] self.average = False self.nbValues = {} self.height = 300 self.width = 600 self.dots = False self.colors = None self.gffFileName = None self.csvFileName = None self.yMin = None self.yMax = None self.chromosome = None self.merge = False self.nbTranscripts = None self.factors = None self.thicknessCurve = 1 self.sizePoliceLegend = 1.5 def setInputFiles(self, fileNames, format): self.fileNames = fileNames self.format = format def setReferenceFile(self, fileName): if fileName == None: return fastaParser = FastaParser(fileName, self.verbosity) self.chromosomes = fastaParser.getRegions() self.sizes = dict([region, fastaParser.getSizeOfRegion(region)] for region in self.chromosomes) self.maxSize = max(self.sizes.values()) def setRegion(self, chromosome, start, end): if chromosome == None or start == None or end == None: return self.maxSize = options.end self.sizes = {chromosome: end} self.chromosomes = [chromosome] self.chromosome = chromosome self.start = start self.end = end def setOutputFile(self, fileName): self.outputFileName = fileName def setNbBins(self, nbBins): if nbBins != None: self.nbBins = int(nbBins) def setBinSize(self, binSize): if binSize != None: self.sliceSize = int(binSize) def set2Strands(self, twoStrands): self.twoStrands = twoStrands def setNames(self, names): self.names = names if len(self.names) == 1 and len(self.fileNames) > 1: self.names = ["file %d" % (i+1) for i in range(len(self.fileNames))] def setAverage(self, average): self.average = average def setNormalization(self, normalization): self.normalization = normalization def setNormalizationFactors(self, factors): self.factors = dict([name, 1.0] for name in self.names) if factors == None else dict(zip(self.names, factors)) def setImageSize(self, height, width): self.height = height self.width = width def setDots(self, dots): self.dots = dots def setYLimits(self, yMin, yMax): self.yMin = yMin self.yMax = yMax def setColors(self, colors): self.colors = colors def writeGff(self, fileName): self.gffFileName = fileName def writeCsv(self, fileName): self.csvFileName = fileName def mergePlots(self, merge): self.merge = merge def setThicknessCurve(self, thickness) : self.thickness = thickness def setSizePoliceLegend(self, sizePoliceLegend): self.sizePoliceLegend = sizePoliceLegend def _estimateSizes(self): self.sizes = {} self.nbTranscripts = {} for fileName in self.fileNames: progress = UnlimitedProgress(10000, "Reading %s for chromosome size estimate" % (fileName), self.verbosity) parserChooser = ParserChooser(self.verbosity) parserChooser.findFormat(self.format) parser = parserChooser.getParser(fileName) for nbTranscripts, transcript in enumerate(parser.getIterator()): if transcript.__class__.__name__ == "Mapping": transcript = transcript.getTranscript() chromosome = transcript.getChromosome() start = transcript.getStart() self.sizes[chromosome] = max(start, self.sizes.get(chromosome, 0)) progress.inc() progress.done() self.nbTranscripts[fileName] = nbTranscripts def _computeSliceSize(self): if self.nbBins == 0: return tmp1 = int(max(self.sizes.values()) / float(self.nbBins)) tmp2 = 10 ** (len("%d" % (tmp1))-2) self.sliceSize = max(1, int((tmp1 / tmp2) * tmp2)) if self.verbosity > 0: print "choosing bin size of %d" % (self.sliceSize) def _initBins(self): self.bins = {} for chromosome in self.sizes: self.bins[chromosome] = {} for name in self.names: self.bins[chromosome][name] = {} for strand in TWOSTRANDS[self.twoStrands]: if self.nbBins == 0: self.bins[chromosome][name][strand] = {} else: self.bins[chromosome][name][strand] = dict([(i * self.sliceSize + 1, 0.0) for i in range(self.start / self.sliceSize, self.sizes[chromosome] / self.sliceSize + 1)]) def _populateBins(self): for id, fileName in enumerate(self.fileNames): if self.nbTranscripts == None: progress = UnlimitedProgress(10000, "Counting data", self.verbosity) else: progress = Progress(self.nbTranscripts[fileName], "Counting data", self.verbosity) parserChooser = ParserChooser(self.verbosity) parserChooser.findFormat(self.format) parser = parserChooser.getParser(fileName) for transcript in parser.getIterator(): if transcript.__class__.__name__ == "Mapping": transcript = transcript.getTranscript() progress.inc() chromosome = transcript.getChromosome() start = transcript.getStart() if self.chromosome and (chromosome != self.chromosome or start < self.start or start > self.end): continue strand = transcript.getDirection() if self.twoStrands else 0 if self.nbBins != 0: bin = (start / self.sliceSize) * self.sliceSize + 1 else: bin = start if len(self.fileNames) > 1: nbElements = transcript.getTagValue("nbElements") if "nbElements" in transcript.getTagNames() else 1 name = self.names[id] self.bins[chromosome][name][strand][bin] = self.bins[chromosome][name][strand].get(bin, 0) + nbElements self.nbValues[name] = self.nbValues.get(name, 0) + nbElements else: for name in self.names: value = float(transcript.tags.get(name, 1)) self.bins[chromosome][name][strand][bin] = self.bins[chromosome][name][strand].get(bin, 0) + value self.nbValues[name] = self.nbValues.get(name, 0) + value progress.done() def _normalizeFactors(self): for chromosome in self.bins: for name in self.bins[chromosome]: for strand in self.bins[chromosome][name]: for bin in self.bins[chromosome][name][strand]: self.bins[chromosome][name][strand][bin] *= self.factors[name] def _normalize(self): average = float(sum(self.nbValues.values())) / len(self.nbValues.keys()) self.factors = dict([name, float(average) / self.nbValues[name]] for name in self.nbValues) self._normalizeFactors() def _computeAverage(self): for chromosome in self.bins: for name in self.bins[chromosome]: for strand in self.bins[chromosome][name]: for bin in self.bins[chromosome][name][strand]: self.bins[chromosome][name][strand][bin] = float(self.bins[chromosome][name][strand][bin]) / self.sliceSize def _getPlotter(self, chromosome): plot = RPlotter("%s_%s.png" % (os.path.splitext(self.outputFileName)[0], chromosome), self.verbosity) plot.setImageSize(self.width, self.height) plot.setLineWidth(self.thickness) plot.setSizePoliceLegend(self.sizePoliceLegend) if self.dots: plot.setPoints(True) if self.sizes[chromosome] <= 1000: unit = "nt." ratio = 1.0 elif self.sizes[chromosome] <= 1000000: unit = "kb" ratio = 1000.0 else: unit = "Mb" ratio = 1000000.0 if self.yMin != None: plot.setMinimumY(self.yMin) if self.yMax != None: plot.setMaximumY(self.yMax) plot.setXLabel("Position on %s (in %s)" % (chromosome.replace("_", " "), unit)) if len(self.names) > 1: plot.setLegend(True, True) for i, name in enumerate(self.bins[chromosome]): for strand in self.bins[chromosome][name]: #fullName = "%s %s" % (name.replace("_", " ")[:6], STRANDTOSTR[strand]) fullName = name.replace("_", " ")[:6] factor = 1 if strand == 0 else strand correctedLine = dict([(key / ratio, value * factor) for key, value in self.bins[chromosome][name][strand].iteritems()]) plot.addLine(correctedLine, fullName, self.colors[i] if self.colors else None) return plot def _plot(self): if self.merge: multiplePlot = MultipleRPlotter(self.outputFileName, self.verbosity) multiplePlot.setImageSize(self.width, self.height * len(self.bins.keys())) progress = Progress(len(self.bins.keys()), "Plotting", options.verbosity) for chromosome in sorted(self.bins.keys()): plot = self._getPlotter(chromosome) if self.merge: multiplePlot.addPlot(plot) else: plot.plot() progress.inc() if self.merge: multiplePlot.plot() progress.done() def _writeCsv(self): if self.verbosity > 1: print "Writing CSV file..." csvHandle = open(self.csvFileName, "w") csvHandle.write("chromosome;tag;strand") if self.nbBins != 0: xValues = range(self.start / self.sliceSize, max(self.sizes.values()) / self.sliceSize + 1) for value in xValues: csvHandle.write(";%d-%d" % (value * self.sliceSize + 1, (value+1) * self.sliceSize)) csvHandle.write("\n") else: xValues = [] for chromosome in self.bins: for name in self.bins[chromosome]: for strand in self.bins[chromosome][name]: for bin in self.bins[chromosome][name][strand]: xValues.extend(self.bins[chromosome][name][strand].keys()) xValues = sorted(list(set(xValues))) for value in xValues: csvHandle.write(";%d" % (value)) csvHandle.write("\n") for chromosome in self.bins: csvHandle.write("%s" % (chromosome)) for name in self.bins[chromosome]: csvHandle.write(";%s" % (name)) for strand in self.bins[chromosome][name]: csvHandle.write(";%s" % (STRANDTOSTR[strand])) for bin in xValues: csvHandle.write(";%.2f" % (self.bins[chromosome][name][strand].get(bin, 0))) csvHandle.write("\n") csvHandle.write(";") csvHandle.write(";") csvHandle.close() if self.verbosity > 1: print "...done" def _writeGff(self): if self.verbosity > 1: print "Writing GFF file..." writer = Gff3Writer(self.gffFileName, self.verbosity) cpt = 1 for chromosome in self.bins: for name in self.bins[chromosome]: for strand in self.bins[chromosome][name]: for bin in self.bins[chromosome][name][strand]: transcript = Transcript() transcript.setChromosome(chromosome) transcript.setStart(bin) if self.nbBins > 0: transcript.setEnd(bin + self.sliceSize) else: transcript.setEnd(self.start) transcript.setDirection(1 if strand == 0 else strand) transcript.setTagValue("ID", "region%d" % (cpt)) cpt += 1 writer.write() if self.verbosity > 1: print "...done" def run(self): if self.sizes == None: self._estimateSizes() if self.sliceSize == None: self._computeSliceSize() self._initBins() self._populateBins() if self.normalization: self._normalize() if self.factors != None: self._normalizeFactors() if self.average: self._computeAverage() self._plot() if self.csvFileName != None: self._writeCsv() if self.gffFileName != None: self._writeGff() if __name__ == "__main__": description = "Get Distribution v1.0.2: Get the distribution of the genomic coordinates on a genome. [Category: Visualization]" parser = OptionParser(description = description) parser.add_option("-i", "--input", dest="inputFileNames", action="store", type="string", help="input files separated by commas [compulsory] [format: string]") parser.add_option("-f", "--format", dest="format", action="store", type="string", help="format of the input file [compulsory] [format: transcript file format]") parser.add_option("-o", "--output", dest="outputFileName", action="store", type="string", help="output file [compulsory] [format: output file in GFF3 format]") parser.add_option("-r", "--reference", dest="referenceFileName", action="store", default=None, type="string", help="file containing the genome [format: file in FASTA format]") parser.add_option("-b", "--nbBins", dest="nbBins", action="store", default=1000, type="int", help="number of bins [default: 1000] [format: int]") parser.add_option("-B", "--binSize", dest="binSize", action="store", default=None, type="int", help="bin size [default: None] [format: int]") parser.add_option("-2", "--bothStrands", dest="bothStrands", action="store_true", default=False, help="plot one curve per strand [format: bool] [default: false]") parser.add_option("-c", "--chromosome", dest="chromosome", action="store", default=None, type="string", help="plot only a chromosome [format: string]") parser.add_option("-s", "--start", dest="start", action="store", default=None, type="int", help="start from a given region [format: int]") parser.add_option("-e", "--end", dest="end", action="store", default=None, type="int", help="end from a given region [format: int]") parser.add_option("-y", "--yMin", dest="yMin", action="store", default=None, type="int", help="minimum value on the y-axis to plot [format: int]") parser.add_option("-Y", "--yMax", dest="yMax", action="store", default=None, type="int", help="maximum value on the y-axis to plot [format: int]") parser.add_option("-x", "--csv", dest="csv", action="store", default=None, help="write a .csv file [format: output file in CSV format] [default: None]") parser.add_option("-g", "--gff", dest="gff", action="store", default=None, help="also write GFF3 file [format: output file in GFF format] [default: None]") parser.add_option("-H", "--height", dest="height", action="store", default=500, type="int", help="height of the graphics [format: int] [default: 300]") parser.add_option("-W", "--width", dest="width", action="store", default=800, type="int", help="width of the graphics [format: int] [default: 1000]") parser.add_option("-t", "--thickness", dest="lineThickness", action="store", default=1, type="int", help="thickness of the lines [format : int] [default : 1]") parser.add_option("-d", "--policeLegend", dest="sizePoliceLegend", action="store", default=1.5, type="float", help="size of the police of the legend [format : float] [default : 1.5]") parser.add_option("-D", "--dots", dest="dots", action="store_true", default=False, help="plot dots instead of lines [format : bool] [default : false]") parser.add_option("-a", "--average", dest="average", action="store_true", default=False, help="plot average (instead of sum) [default: false] [format: boolean]") parser.add_option("-n", "--names", dest="names", action="store", default="nbElements", type="string", help="name for the tags (separated by commas and no space) [default: None] [format: string]") parser.add_option("-l", "--color", dest="colors", action="store", default=None, type="string", help="color of the lines (separated by commas and no space) [format: string]") parser.add_option("-z", "--normalize", dest="normalize", action="store_true", default=False, help="normalize data (when panels are different) [format: bool] [default: false]") parser.add_option("-Z", "--normalizeFac", dest="normalizeFactors", action="store", default=None, help="normalize data with given factors (when panels are different) [format: string]") parser.add_option("-m", "--merge", dest="mergePlots", action="store_true", default=False, help="merge all plots in one figure [format: bool] [default: false]") parser.add_option("-v", "--verbosity", dest="verbosity", action="store", default=1, type="int", help="trace level [default: 1] [format: int]") (options, args) = parser.parse_args() gt = GetDistribution(options.verbosity) gt.setInputFiles(options.inputFileNames.split(","), options.format) gt.setOutputFile(options.outputFileName) gt.setReferenceFile(options.referenceFileName) gt.setNbBins(options.nbBins) gt.setBinSize(options.binSize) gt.set2Strands(options.bothStrands) gt.setRegion(options.chromosome, options.start, options.end) gt.setNormalization(options.normalize) gt.setAverage(options.average) gt.setYLimits(options.yMin, options.yMax) gt.writeCsv(options.csv) gt.writeGff(options.gff) gt.setImageSize(options.height, options.width) gt.setNames(options.names.split(",")) gt.setThicknessCurve(options.lineThickness) gt.setSizePoliceLegend(options.sizePoliceLegend) gt.setColors(None if options.colors == None else options.colors.split(",")) gt.setDots(options.dots) gt.setNormalization(options.normalize) gt.setNormalizationFactors(None if options.normalizeFactors == None else [float(factor) for factor in options.normalizeFactors.split(",")]) gt.mergePlots(options.mergePlots) gt.run()