Mercurial > repos > fubar > bigwig_outlier_bed
view bigwig_outlier_bed.py @ 7:c8e22efcaeda draft
planemo upload for repository https://github.com/galaxyproject/tools-iuc/tree/master/tools/bigwig_outlier_bed commit 9fa87e27ea819badb876e6d89807a789119b9f53
author | fubar |
---|---|
date | Wed, 24 Jul 2024 08:49:37 +0000 |
parents | eb17eb8a3658 |
children | 032e930ef6a1 |
line wrap: on
line source
""" Ross Lazarus June 2024 for VGP Bigwigs are great, but hard to reliably "see" small low coverage or small very high coverage regions. Colouring in JB2 tracks will need a new plugin, so this code will find bigwig regions above and below a chosen percentile point. 0.99 and 0.01 work well in testing with a minimum span of 10 bp. Multiple bigwigs **with the same reference** can be combined - bed segments will be named appropriately Combining multiple references works but is silly because only display will rely on one reference so others will not be shown... Tricksy numpy method from http://gregoryzynda.com/python/numpy/contiguous/interval/2019/11/29/contiguous-regions.html takes about 95 seconds for a 17MB test wiggle JBrowse2 bed normally displays ignore the score, so could provide separate low/high bed file outputs as an option. Update june 30 2024: wrote a 'no-build' plugin for beds to display red/blue if >0/<0 so those are used for scores Bed interval naming must be short for JB2 but needs input bigwig name and (lo or hi). """ import argparse import os import sys from pathlib import Path import numpy as np import pybigtools class asciihist: def __init__( self, data, bins=10, minmax=None, str_tag="", scale_output=80, generate_only=True, ): """ https://gist.github.com/bgbg/608d9ef4fd75032731651257fe67fc81 Create an ASCII histogram from an interable of numbers. Author: Boris Gorelik boris@gorelik.net. based on http://econpy.googlecode.com/svn/trunk/pytrix/pytrix.py License: MIT """ self.data = data self.minmax = minmax self.str_tag = str_tag self.bins = bins self.generate_only = generate_only self.scale_output = scale_output self.itarray = np.asanyarray(self.data) if self.minmax == "auto": self.minmax = np.percentile(data, [5, 95]) if self.minmax[0] == self.minmax[1]: # for very ugly distributions self.minmax = None if self.minmax is not None: # discard values that are outside minmax range mn = self.minmax[0] mx = self.minmax[1] self.itarray = self.itarray[self.itarray >= mn] self.itarray = self.itarray[self.itarray <= mx] def draw(self): values, counts = np.unique(self.data, return_counts=True) if len(values) <= 20: self.bins = len(values) ret = [] if self.itarray.size: total = len(self.itarray) counts, cutoffs = np.histogram(self.itarray, bins=self.bins) cutoffs = cutoffs[1:] if self.str_tag: self.str_tag = "%s " % self.str_tag else: self.str_tag = "" if self.scale_output is not None: scaled_counts = counts.astype(float) / counts.sum() * self.scale_output else: scaled_counts = counts footerbar = "{:s}{:s} |{:s} |".format(self.str_tag, "-" * 12, "-" * 12,) if self.minmax is not None: ret.append( "Trimmed to range (%s - %s)" % (str(self.minmax[0]), str(self.minmax[1])) ) for cutoff, original_count, scaled_count in zip( cutoffs, counts, scaled_counts ): ret.append( "{:s}{:>12.2f} |{:>12,d} | {:s}".format( self.str_tag, cutoff, original_count, "*" * int(scaled_count) ) ) ret.append(footerbar) ret.append("{:s}{:>12s} |{:>12,d} |".format(self.str_tag, "N=", total)) ret.append(footerbar) ret.append('') else: ret = [] if not self.generate_only: for line in ret: print(line) ret = "\n".join(ret) return ret class findOut: def __init__(self, args): self.bwnames = args.bigwig self.bwlabels = args.bigwiglabels self.bedwin = args.minwin self.outbeds = args.outbeds self.bedouthi = args.bedouthi self.bedoutlo = args.bedoutlo self.bedouthilo = args.bedouthilo self.tableoutfile = args.tableoutfile self.bedwin = args.minwin self.qlo = None self.qhi = None if args.outbeds != "outtab": self.qhi = args.qhi if args.qlo: try: f = float(args.qlo) self.qlo = f except Exception: print('qlo not provided') nbw = len(args.bigwig) nlab = len(args.bigwiglabels) if nlab < nbw: self.bwlabels += ["Nolabel"] * (nbw - nlab) self.makeBed() def processVals(self, bw, isTop): """ idea from http://gregoryzynda.com/python/numpy/contiguous/interval/2019/11/29/contiguous-regions.html Fast segmentation into regions by taking np.diff on the boolean array of over (under) cutpoint indicators in bwex. This only gives non-zero values at the segment boundaries where there's a change, so those zeros are all removed in bwexdnz leaving an array of segment start/end positions. That's twisted around into an array of start/end coordinates. Magical. Fast. Could do the same for means or medians over windows for sparse bigwigs like repeat regions. """ if isTop: bwex = np.r_[False, bw >= self.bwtop, False] # extend with 0s else: bwex = np.r_[False, bw <= self.bwbot, False] bwexd = np.diff(bwex) bwexdnz = bwexd.nonzero()[0] bwregions = np.reshape(bwexdnz, (-1, 2)) return bwregions def writeBed(self, bed, bedfname): """ potentially multiple """ bed.sort() beds = ["%s\t%d\t%d\t%s\t%d" % x for x in bed] with open(bedfname, "w") as bedf: bedf.write("\n".join(beds)) bedf.write("\n") def makeTableRow(self, bw, bwlabel, chr): """ called for every contig, but messy inline """ bwmean = np.mean(bw) bwstd = np.std(bw) bwmax = np.max(bw) nrow = np.size(bw) bwmin = np.min(bw) row = "%s\t%s\t%d\t%f\t%f\t%f\t%f" % ( bwlabel, chr, nrow, bwmean, bwstd, bwmin, bwmax, ) if self.qhi is not None: row += "\t%.2f" % self.bwtop else: row += "\tnoqhi" if self.qlo is not None: row += "\t%.2f" % self.bwbot else: row += "\tnoqlo" return row def makeBed(self): bedhi = [] bedlo = [] restab = [] bwlabels = self.bwlabels bwnames = self.bwnames bwnames.sort() reshead = "bigwig\tcontig\tn\tmean\tstd\tmin\tmax\tqtop\tqbot" for i, bwname in enumerate(bwnames): bwlabel = bwlabels[i].replace(" ", "") fakepath = "in%d.bw" % i if os.path.isfile(fakepath): os.remove(fakepath) p = Path(fakepath) p.symlink_to(bwname) # required by pybigtools (!) bwf = pybigtools.open(fakepath) chrlist = bwf.chroms() chrs = list(chrlist.keys()) for chr in chrs: first_few = None bw = bwf.values(chr) values, counts = np.unique(bw, return_counts=True) nvalues = len(values) if nvalues <= 20: histo = '\n'.join(['%s: %f occurs %d times' % (chr, values[x], counts[x]) for x in range(len(values))]) else: last10 = range(nvalues-10, nvalues) first_few = ['%.2f\t%d' % (values[x],counts[x]) for x in range(10)] first_few += ['%.2f\t%d' % (values[x],counts[x]) for x in last10] first_few.insert(0,'First/Last 10 value counts\nValue\tCount') ha = asciihist(data=bw, bins=20, str_tag=chr) histo = ha.draw() histo = '\n'.join(first_few) + '\nHistogram of bigwig values\n' + histo bw = bw[~np.isnan(bw)] # some have NaN if parts of a contig not covered if self.qhi is not None: self.bwtop = np.quantile(bw, self.qhi) bwhi = self.processVals(bw, isTop=True) for j, seg in enumerate(bwhi): seglen = seg[1] - seg[0] if seglen >= self.bedwin: score = np.sum(bw[seg[0]:seg[1]])/float(seglen) bedhi.append( ( chr, seg[0], seg[1], "%s_%d" % (bwlabel, score), score, ) ) if self.qlo is not None: self.bwbot = np.quantile(bw, self.qlo) bwlo = self.processVals(bw, isTop=False) for j, seg in enumerate(bwlo): if seg[1] - seg[0] >= self.bedwin: score = -1 * np.sum(bw[seg[0]:seg[1]])/float(seglen) bedlo.append( ( chr, seg[0], seg[1], "%s_%d" % (bwlabel, score), score, ) ) if self.tableoutfile: row = self.makeTableRow(bw, bwlabel, chr) resheadl = reshead.split('\t') rowl = row.split() desc = ['%s\t%s' % (resheadl[x], rowl[x]) for x in range(len(rowl))] desc.insert(0, 'Descriptive measures') descn = '\n'.join(desc) restab.append(descn) restab.append(histo) if os.path.isfile(fakepath): os.remove(fakepath) if self.tableoutfile: stable = "\n".join(restab) with open(self.tableoutfile, "w") as t: t.write(stable) t.write("\n") some = False if self.qlo: if self.outbeds in ["outall", "outlo", "outlohi"]: self.writeBed(bedlo, self.bedoutlo) some = True if self.qhi: if self.outbeds in ["outall", "outlohi", "outhi"]: self.writeBed(bedhi, self.bedouthi) some = True if self.outbeds in ["outall", "outhilo"]: allbed = bedlo + bedhi self.writeBed(allbed, self.bedouthilo) some = True if not ((self.outbeds == 'outtab') or some): sys.stderr.write( "Invalid configuration - no output could be created. Was qlo missing and only low output requested for example?" ) sys.exit(2) if __name__ == "__main__": parser = argparse.ArgumentParser() a = parser.add_argument a("-m", "--minwin", default=10, type=int) a("-l", "--qlo", default=None) a("-i", "--qhi", default=None, type=float) a("--bedouthi", default=None) a("--bedoutlo", default=None) a("--bedouthilo", default=None) a("-w", "--bigwig", nargs="+") a("-n", "--bigwiglabels", nargs="+") a("-o", "--outbeds", default="outhilo", help="optional high and low combined bed") a("-t", "--tableoutfile", default=None) args = parser.parse_args() findOut(args)