#!/usr/bin/python import getopt, sys import base_types from base_types import * verbose = False output = sys.stdout import string import itertools from random import uniform, randint from math import sqrt from operator import itemgetter rpy = None try: import rpy except ImportError: pass class sample(tuple): """Hold sample data from which we will calculate the test stat. """ def __init__(self, *args, **kwargs): # tuple.__init__(self, *args, **kwargs) tuple.__init__(self) def __mul__(self, scalar): """Multiply a sample by a scalar. eg 0.5*(1,1,1,1,2,1) = (0.5,0.5,0.5,0.5,1.0,0.5) """ # BAD python 2.3 compat change return type(self)( [scalar*item for item in self] ) def __add__(self, other): """Add two samples together pointwise. eg (0,0,0,0,0,0) + (1,1,1,1,2,1) = (1,1,1,1,2,1) """ if len(self) != len(other): raise ValueError, "Can only add samples of the same length." return type(self)( [ i1 + i2 for i1, i2 in itertools.izip( self, other ) ] ) class sample_dict(dict): """Store a dictionary of samples. This is generally used to store all of the samples for a region, categorized by their named partitions. An example of this would be looking at an entire, genome, but having it partitioned by chromosome. The keys would be the chr names. """ def __setitem__(self, key, item): #assert isinstance(item, sample) dict.__setitem__(self, key, item) def sum(self): """Returns the sum of all the overlap_sample objects. In example, if the values in this container were (0,0,0,0,2,4) and (1,4,5,7,2,4) this would return (1,4,5,7,4,8). """ if len(self) == 0: raise ValueError, "Can't sum an empty sequence" # define a container to hold the values empty = type(self.values()[0])(0,0,0,0,0,0) # iterate through the items, and add them to the empty container for item in self.values(): empty += item return empty def weightedSum(self, weights): """Returns the sum of all overlap_sample objects weighted by weights. In example, if the values in this container were 'R1': (0,0,0,0,2,4) and 'R2': (1,4,5,7,2,4) and the weights were 'R1': 0.1 and 'R2': 0.9 then this would return (0, 0, 0, 0, .2, .4) + (0.9, 3.6, 4.5, 6.3, 1.8, 3.6) = (0.9, 3.6, 4.5, 6.3, 2.0, 4.0) """ if len(self) == 0: raise ValueError, "Can't sum an empty sequence" # make sure that the keys are identical assert set(weights.keys()) == set(self.keys()) # make sure that the weights sum to 1 assert round( sum( weights.values() ), 15) == 1.0 # define a container to hold the values obj = self.values()[0] obj_type = type(obj) if obj_type in (int,float): empty = obj_type(0) else: empty = obj_type(args=[0]*len(obj)) # BAD python2.3 compatability change return sum( [ self[key]*weights[key] for key in self.keys() ], empty ) def double_overlap( coveredRegion, coveringRegion, outer_regionFraction, inner_regionFraction, callback, number=1 ): """Double sample from a pair of regions and calculate aggregate stats. This is a wrapper for custom GSC statistics. This code samples from the regions with the specified region fractions, and then calls callback with s11,s12, s21, s22 for the ouer and inner samples. The callback should return an object inherited from sample. Then, double overlap sample will return a sample dict containing the samples hashed by region name. It's probably easiest to quickly browse the code and an example. """ osl = outer_regionFraction isl = inner_regionFraction # for loop in the number of samples to take for loop in xrange(number): # first, build a dict of the outer samples # these are what we use to calculate the expectation # of the statistic osample = sample_dict() isample = sample_dict() # for each named region in the coveringRegion... for key in coveringRegion.keys(): # rename the regions to be more readable cA = coveringRegion[key] eA = coveredRegion[key] if cA.length != eA.length: raise ValueError, "Regions must be the same length in bp's." ############################### # Take the outer samples ############################### # find the length of the sample in bp's sample_length = int(eA.length*osl) # take the slices start_bp = randint(0, cA.length - sample_length - 1) os1 = eA.get_subregion( start_bp, start_bp+sample_length, shift_to_zero=True ) start_bp = randint(0, cA.length - sample_length - 1) os2 = cA.get_subregion( start_bp, start_bp+sample_length, shift_to_zero=True ) # save the outer sample stat osample[key] = callback( os1, os2 ) ############################### # Take the (inner) sub samples ############################### # find the length of the sample in bp's sample_length = int(sample_length*isl) # take the slices start_bp = randint(0, os1.length - sample_length - 1) is1 = eA.get_subregion( start_bp, start_bp+sample_length, shift_to_zero=True ) start_bp = randint(0, os1.length - sample_length - 1) is2 = cA.get_subregion( start_bp, start_bp+sample_length, shift_to_zero=True ) # save the inner sample stat isample[key] = callback( is1, is2 ) yield osample, isample def single_overlap( coveredRegion, coveringRegion, regionFraction, callback, number=1 ): """Block sample from a pair of regions and calculate aggregate stats. This is a wrapper for custom GSC statistics. This code samples from the regions with the specified region fractions, and then calls callback with s11,s12, s21, s22 for the otuer and inner samples. The callback should return an object inherited from sample. Then, single overlap sample will return a sample dict containing the samples hashed by region name. It's probably easiest to quickly browse the code and an example - correlation is a good placde to start. """ sl = regionFraction # for loop in the number of samples to take for loop in xrange(number): sample = sample_dict() # for each named region in the coveringRegion... for key in coveringRegion.keys(): # rename the regions to be more readable cA = coveringRegion[key] eA = coveredRegion[key] if cA.length != eA.length: raise ValueError, "Regions must be the same length in bp's." ############################### # Take the outer samples ############################### # find the length of the sample in bp's sample_length = int(eA.length*sl) # take the slices start_bp = randint(0, cA.length - sample_length - 1) s1 = eA.get_subregion( start_bp, start_bp+sample_length, shift_to_zero=True ) start_bp = randint(0, cA.length - sample_length - 1) s2 = cA.get_subregion( start_bp, start_bp+sample_length, shift_to_zero=True ) # save the outer sample stat sample[key] = callback( s1, s2 ) yield sample def conditional_bp_overlap_stat( \ covering_region, covered_region, region_fraction, num_samples ): #% Lastly, scale the bootstrap distribution to get the null distribution of #% the test statistic. # #% 2*frac = 2L/n. So sqrt(2*frac)*T_n has the correct SD. We can compute #% this by pulling the constant out: # #% store the samples from n and N #%SD = sqrt(2*frac)*std(Tn); # Tns = [] Tns_2 = [] # calculate the overlap stat num_samples times samples = random_regions_bp_overlap(covered_region, covering_region, region_fraction, num_samples) # BAD python2.3 compat change lengths = dict([ (key, covered_region[key].length) for key in covered_region.keys() ]) if verbose: print >> output, "#### Sample Distribution Info" print >> output, "Sample #".rjust(8), "Tn".rjust(15) loop = 0 for sample in samples: stats = conditional_overlap_sample_stat(sample, lengths, region_fraction) Tns.append(stats['Tn']) if verbose: print >> output, str(loop).rjust(8), str("%.8f" % stats['Tn']).rjust(15) loop += 1 SD = sqrt(2*region_fraction)*std(Tns) if verbose: print >> output print >> output, 'mean: ', mean(Tns) print >> output, 'SD: ', SD, '\n' real_stats = conditional_overlap_stat(covered_region, covering_region) theoretical_stat_mean = real_stats['theoretical'] observed_stat_mean = real_stats['observed'] test_stat = real_stats['test_stat'] if verbose: print >> output, 'NULL stat mean: ', theoretical_stat_mean print >> output, 'observed stat mean: ', observed_stat_mean print >> output, 'test_stat: ', test_stat # Finally, refer the "mean zero" test statistic "test_stat" to the # distribution Tn. Compute the SD and whatnot. # #%z_score = test_stat/SD; #%p_value = 1 - normcdf(test_stat,0,SD); z_score = test_stat/SD if verbose: print >> output, 'z_score: ', z_score p_value = min(1 - sn_cdf(z_score), sn_cdf(z_score)) print >> output, 'p_value: ', p_value, "\n" return z_score, p_value def double_bootstrap_stat( covering_regions, covered_regions, outer_regionFraction, inner_regionFraction, aggregate_sample_callback, regions_agg_callback, nsamples=1, real_stat_callback=None ): """Wrapper for generic double bootstrap stat Input Parameters: covering_regions - the regions that num regions is calulated from covered_regions - the other regions outer_region_fraction - the ratio of the outer sample inner_region_fraction - the ratio of the subsample to the outer sample aggregate_samples_callback: this is what is called on the block samples that are taken. the quintesential example of this is region overlap. scale_sample_callback: scales the stat that was aggregated from aggregate_samples_callback. regions_agg_callback: How to aggregate the samples. The two implemented are the conditional and marginal version. Marginal adds up all of the segments - the conditional weights them by the segment weights. nsamples - the number of samples to take real_stat_callback - sometimes we need to calculate the 'true' stat differently than for the samples ( for instance, maybe it needsw to be scaled differently. In such cases, pass this in. If it is none, the real stat is calculated by calling the sample functions on the whole set of regions. """ # store the samples o_samples = [] i_samples = [] if verbose: # TODO make this prettier, maybe with a sample header method? print >> output, "#### Samples\n" print >> output, "Sample #".rjust(8), "Outer Region Stat".rjust(23), \ "Inner Region Stat".rjust(23) # take the samples and keep track of the relevant data for loop, item in \ enumerate(double_overlap( covered_regions, covering_regions, outer_regionFraction, inner_regionFraction, aggregate_sample_callback, nsamples )): # unpack the samples tuple outer_samples, inner_samples = item # append the aggregated stats, should be a floats o_samples.append(regions_agg_callback(outer_samples)) i_samples.append(regions_agg_callback(inner_samples)) #print outer_samples, inner_samples #print agg_callback(outer_samples), agg_callback(inner_samples) if verbose: print >> output, str(loop+1).rjust(8), \ str(o_samples[-1]).rjust(23), str(i_samples[-1]).rjust(23) # the expectation of the mean under the NULL # our estimate is the empirical mean of the outer sample Nmean = mean(o_samples) obsMean = mean(i_samples) if rpy != None and verbose: rpy.r.png( "dist.png", width=600, height=600 ) rpy.r.par( mfrow=(2,2) ) rpy.r.hist( o_samples, main="Outer Sample", xlab="", ylab="" ) rpy.r.hist( i_samples, main="Inner Sample", xlab="", ylab="" ) rpy.r.qqnorm( o_samples, main="Outer Sample", xlab="", ylab="" ) rpy.r.qqline( o_samples ) rpy.r.qqnorm( i_samples, main="Inner Sample", xlab="", ylab="" ) rpy.r.qqline( i_samples ) rpy.r.dev_off() dist = [ (sample - Nmean) for sample in i_samples ] dist_sd = sqrt(outer_regionFraction*inner_regionFraction)*std(dist) if verbose: print >> output, "\nSample Dist Info:\n" if verbose: print >> output, "\tThe expected mean under the NULL: ", Nmean if verbose: print >> output, "\tThe sample mean: ", obsMean if verbose: print >> output, "\tThe normalized bootstrap SD: ", dist_sd, "\n" # the stat on each of the segments if None == real_stat_callback: seg_stats = sample_dict([ (key, aggregate_sample_callback( covered_regions[key], covering_regions[key] )) for key in covered_regions.keys() ]) real_stat_expectation = regions_agg_callback( seg_stats ) else: real_stat_expectation = real_stat_callback( covered_regions, covered_regions ) obs_mean = real_stat_expectation - Nmean if verbose: print >> output, "\tThe real stat: ", real_stat_expectation if verbose: print >> output, "\tThe scaled real stat: ", obs_mean, "\n" z_score = obs_mean/dist_sd print "\tZ Score: ", z_score p_value = min( 1 - sn_cdf( z_score ), sn_cdf( z_score ) ) print "\tp-value: ", p_value return z_score, p_value def single_bootstrap_stat( covering_regions, covered_regions, regionFraction, exp_under_null_callback, aggregate_sample_callback, regions_agg_callback, nsamples=1, real_stat_callback=None): """Wrapper for generic double bootstrap stat Input Parameters: covering_regions - the regions that num regions is calulated from covered_regions - the other regions region_fraction - the fraction of the region to sample exp_under_null_callback - determine the expectation of the stat under the null, as a function of the covering and covered regions. aggregate_samples_callback: this is what is called on the block samples that are taken. the quintesential example of this is region overlap. scale_sample_callback: scales the stat that was aggregated from aggregate_samples_callback. regions_agg_callback: How to aggregate the samples. The two implemented are the conditional and marginal version. Marginal adds up all of the segments - the conditional weights them by the segment weights. nsamples - the number of samples to take """ # store the samples samples = [] if verbose: # TODO make this prettier, maybe with a sample header method? print >> output, "#### Samples\n" print >> output, "Sample #".rjust(8), "Stat".rjust(23) # take the samples and keep track of the relevant data for loop, sample in \ enumerate(single_overlap( covered_regions, covering_regions, regionFraction, aggregate_sample_callback, nsamples )): # append the aggregated stats, should be a floats samples.append(regions_agg_callback(sample)) #print outer_samples, inner_samples #print agg_callback(outer_samples), agg_callback(inner_samples) if verbose: print >> output, str(loop+1).rjust(8), \ str(samples[-1]).rjust(23) # the expectation of the mean under the NULL # our estimate is the empirical mean of the outer sample Nmean = exp_under_null_callback( covering_regions, covering_regions ) obsMean = mean(samples) if rpy != None and verbose: rpy.r.png( "dist.png", width=1200, height=600 ) rpy.r.par( mfrow=(1,2) ) rpy.r.hist( samples, main="Sample Hist", xlab="", ylab="", n=nsamples/10.0 ) rpy.r.qqnorm( samples, main="qqplot vs norm quantiles", xlab="", ylab="" ) rpy.r.qqline( samples ) rpy.r.dev_off() dist = [ (sample - Nmean) for sample in samples ] dist_sd = sqrt(regionFraction)*std(dist) if verbose: print >> output, "\nSample Dist Info:\n" if verbose: print >> output, "\tThe expected mean under the NULL: ", Nmean if verbose: print >> output, "\tThe sample mean: ", obsMean if verbose: print >> output, "\tThe normalized bootstrap SD: ", dist_sd, "\n" # the stat on each of the segments # the stat on each of the segments if None == real_stat_callback: seg_stats = sample_dict([ (key, aggregate_sample_callback( covered_regions[key], covering_regions[key] )) for key in covered_regions.keys() ]) real_stat_expectation = regions_agg_callback( seg_stats ) else: real_stat_expectation = real_stat_callback( covered_regions, covering_regions ) obs_mean = real_stat_expectation - Nmean if verbose: print >> output, "\tThe real stat: ", real_stat_expectation if verbose: print >> output, "\tThe scaled real stat: ", obs_mean, "\n" z_score = obs_mean/dist_sd print "\tZ Score: ", z_score p_value = min( 1 - sn_cdf( z_score ), sn_cdf( z_score ) ) print "\tp-value: ", p_value return z_score, p_value def fraction_basepair_overlap( coveredRegionSample, coveringRegionSample ): """ Calculate region overlap statistics. This is intended as a callback for single_overlap. """ # stores the feature length of the covered region ( self ) covered_feature_len = coveredRegionSample.featuresLength() # stores the feature length of the covering region ( coveringRegion ) covering_feature_len = coveringRegionSample.featuresLength() # stores the observed total feature overlap between the region's overlap_feature_len = coveredRegionSample.overlap(coveringRegionSample) return sample( ( overlap_feature_len, covered_feature_len, covering_feature_len ) ) def conditional_bp_overlap_stat( \ covering_regions, covered_regions, \ region_fraction, num_samples ): def agg_callback(sample): total_length = sum( item.length for item in covering_regions.values() ) Fn = 0.0 Jn_num = 0.0 Jn_denom = 0.0 for r_name, values in sample.iteritems(): # unpack the sample values overlap, covered_feature_len, covering_feature_len = values regionLength = float(covering_regions[r_name].length) # lambda in the paper length_frac = regionLength/total_length # the length of the sampled region sampleRegionLength = region_fraction*regionLength I = covered_feature_len/sampleRegionLength J = covering_feature_len/sampleRegionLength IJ = overlap/sampleRegionLength Fn += length_frac*(IJ/max(I, 1.0/sampleRegionLength)) Jn_denom += length_frac*I Jn_num += length_frac*I*J return Fn def analytical_expectation( covering_regions, covered_regions ): totalLength = sum( item.length for item in covering_regions.values() ) Obs_num = 0.0 Obs_den = 0.0 I_num = 0.0 for key in covered_regions.keys(): covered_feature_len = covered_regions[key].featuresLength() covering_feature_len = covering_regions[key].featuresLength() overlap_feature_len = covered_regions[key].overlap(covering_regions[key]) # stores the length of this region for both features region_length = float(covered_regions[key].length) # this is lambda_i in the paper regionFraction = region_length/totalLength Obs_num += regionFraction*overlap_feature_len/region_length Obs_den += regionFraction*covered_feature_len/region_length I_num += regionFraction*covering_feature_len*covered_feature_len/(region_length**2) J_n = I_num/Obs_den; O_n = Obs_num/Obs_den; return { 'expected': J_n, 'observed': O_n } def expectation_under_null( covering_regions, covered_regions ): return analytical_expectation( covering_regions, covered_regions )[ 'expected' ] def real_stat( covering_regions, covered_regions ): return analytical_expectation( covering_regions, covered_regions )[ 'observed' ] z_score, p_value = \ single_bootstrap_stat( covering_regions, covered_regions, \ region_fraction, expectation_under_null, \ fraction_basepair_overlap, agg_callback, \ num_samples, real_stat ) return z_score, p_value def marginal_bp_overlap_stat( \ covering_regions, covered_regions, \ region_fraction, num_samples ): weights = covering_regions.regionFraction() def agg_callback(sample): assert set( sample.keys() ) == set( covering_regions.keys() ) assert set( sample.keys() ) == set( covered_regions.keys() ) tot_overlap = 0 tot_fl = 0 for r_name, values in sample.iteritems(): overlap_feature_len, covered_feature_len, covering_feature_len = values tot_overlap += float(overlap_feature_len) tot_fl += covered_feature_len # If we want to do this conditional on features, uncomment this """ if tot_fl == 0: return 0 else: return tot_overlap/tot_fl """ return tot_overlap/tot_fl def expectation_under_null( covering_regions, covered_regions ): covered_fl = 0 covering_fl = 0 region_lengths = 0 for r_name in covering_regions.keys(): covered_fl += covered_regions[r_name].featuresLength() covering_fl += covering_regions[r_name].featuresLength() region_lengths += covered_regions[r_name].length # Note that this is really (covering*covered/rl**2)/covered, # which reduces algebraically to the below expression return covering_fl/float(region_lengths) z_score, p_value = \ single_bootstrap_stat( covering_regions, covered_regions, \ region_fraction, expectation_under_null, \ fraction_basepair_overlap, agg_callback, \ num_samples ) return z_score, p_value def conditional_resample_region_overlap_stat( covering_regions, covered_regions, outer_regionFraction, inner_regionFraction, nsamples, min_overlap_fraction=0.0 ): """Calculate the probability under the NULL of independence that the region overlap is purely do to chance. I use the machinery in double bootstrap stat to implement this. """ # the region weights - defined as homogenous region length/ total length weights = covering_regions.regionFraction() def agg_callback(sample): assert set( sample.keys() ) == set( covering_regions.keys() ) assert set( sample.keys() ) == set( covered_regions.keys() ) return sample.weightedSum( weights ) def region_overlap( coveredRegionSample, coveringRegionSample ): """ Calculate region overlap statistics. This is intended as a callback for double_overlap. """ percent_overlap = float(coveredRegionSample.regionOverlap(coveringRegionSample, min_overlap_fraction)) / coveredRegionSample.numRegions return percent_overlap return double_bootstrap_stat(covering_regions, covered_regions, outer_regionFraction, inner_regionFraction, region_overlap, agg_callback, nsamples=nsamples) def marginal_resample_region_overlap_stat( covering_regions, covered_regions, outer_regionFraction, inner_regionFraction, nsamples, min_overlap_fraction=0.0 ): """Calculate the probability under the NULL of independence that the region overlap is purely do to chance. I use the machinery in double bootstrap stat to implement this. """ def agg_callback(sample): assert set( sample.keys() ) == set( covering_regions.keys() ) assert set( sample.keys() ) == set( covered_regions.keys() ) overlaps = sum([ item[0] for item in sample.values() ]) num_regions = sum([ item[1] for item in sample.values() ]) return float(overlaps)/num_regions def region_overlap( coveredRegionSample, coveringRegionSample ): """ Calculate region overlap statistics. This is intended as a callback for double_overlap. """ region_overlap = ( float(coveredRegionSample.regionOverlap(coveringRegionSample, min_overlap_fraction)), coveredRegionSample.numRegions ) return region_overlap return double_bootstrap_stat(covering_regions, covered_regions, outer_regionFraction, inner_regionFraction, region_overlap, agg_callback, nsamples=nsamples) def conditional_pearson_correlation( covering_regions, covered_regions, regionFraction, nsamples ): """Calculate the probability under the NULL of independence that the pearson correlation is purely do to chance. I use the machinery in double bootstrap stat to implement this. """ def agg_callback(sample): """Correlation is asmytotically normal, so we weight the correlation by 1/sqrt(region length), since length is the number of points that contribute to the correlation. """ stat = 0 for key, val in sample.iteritems(): length = covering_regions[key].length stat += val/sqrt( length ) return stat def pearson_correlation( coveredRegionSample, coveringRegionSample ): """ Calculate region overlap statistics. This is intended as a callback for double_overlap. """ correlation = coveredRegionSample.regionCorrelation(coveringRegionSample) return correlation def mean_under_null( covering_regions, covered_regions ): return 0.0 return single_bootstrap_stat(covering_regions, covered_regions, regionFraction, mean_under_null, pearson_correlation, agg_callback, nsamples=nsamples) def conditional_mean_fold_enrichment( covering_regions, covered_regions, outer_regionFraction, inner_regionFraction, nsamples ): """Calculate the probability under the NULL of independence that the fold of region 1 WRT region2 is due to chance. I use the machinery in double bootstrap stat to implement this. """ # the region weights - defined as homogenous region length/ total length weights = covering_regions.regionFraction() def agg_callback(sample): assert set( sample.keys() ) == set( covering_regions.keys() ) assert set( sample.keys() ) == set( covered_regions.keys() ) return sample.weightedSum( weights ) def fold( coveredRegionSample, coveringRegionSample ): # special case for empty intervals if len(coveredRegionSample) == 0 or len(coveringRegionSample) == 0: return 0 totalEnrichment = 0 currentMatches = [] thisIter = coveredRegionSample.iter_intervals_and_values() nextCoveredMatch = thisIter.next() for cg_iv, cg_val in coveringRegionSample.iter_intervals_and_values(): # first, get rid of any current matches that dont match # because of the ordering, these should start not matching at the begining for cd_iv, cd_val in currentMatches: if cd_iv.end >= cg_iv.start: break else: del currentMatches[0] # next, add any new items to currentMatches while nextCoveredMatch != None and nextCoveredMatch[0].start <= cg_iv.end: currentMatches.append(nextCoveredMatch) try: nextCoveredMatch = thisIter.next() except StopIteration: nextCoveredMatch = None # finally, calculate the fold enrivchment of every item in currentMatches for cd_iv, cd_val in currentMatches: totalEnrichment += cd_iv.overlap(cg_iv)*( float(cg_val)/cd_val ) return totalEnrichment/coveredRegionSample.featuresLength() return double_bootstrap_stat( covering_regions, covered_regions, outer_regionFraction, inner_regionFraction, fold, agg_callback, nsamples=nsamples ) def usage(): print >> output, """ This calculates a p-value under the null hypothesis that the observed instance-coverage of Feature_1 by Feature_2 is due to chance. INPUTS: -1 : --feature_1 : the 'covered' input data file -2 : --feature_2 : the 'covering' input data file Input files are accepted in either the wiggle or bed format. If the filename extension is .wig, the wiggle file is parsed. Otherwise, the file is assumed to be a bed file. For asymmetric stats, the covered file is what the numerator in the test statistic should be. ie, for fold enrichment, the stat is the quotient covering_region/covered_region. Details are in the doc strings. -d : --domain : the domain of the data files, the portion of the genome over which these features (-1 and -2) are defined. The support of the statistics. Usually determined by array coverage or "alignibility". If the features are defined everywhere (e.g. such as may be the case in C. elegans data), then this file contains one line for each chromosome, with: "chr_name 0 chr_length" on each line. The file formats are described in the README under INPUT FILE FORMATS. -r --region_fraction : the fraction of each region (e.g. chromosome) to take in each block-wise sample. The product of this number, and the minimum segment length (e.g. for no segmentation, using whole human chromosomes, the minimum segment length would be about 50Mb due to chromosome Y), should be larger than the mixing distance of features -1 and -2. For human, if the minimum segment length is around 5Mb (some segmentation has been done, or the features of interest are not defined everywhere throughout the genome, e.g. due to mappability issues), then this number should be at least 0.01 for most features. This number can be fitted under a stability criterion. -s --subregion_fraction: only used for the double bootstrap tests, this specifies the fraction of the subsample to take. For instance, a specified region fraction (-r) of 0.20 and a subregion fraction (-s) of 0.20 would yield a net sample length of 0.04 of each named region. -n --num_samples : the number of bootstrap samples to take. 100 will be sufficient to get an idea of significance, but at least 10K should be used for publication. -t --test: Determine the test to run. Accepts one of the following types basepair_overlap_conditional ( bc ) ( default ) basepair_overlap_marginal ( bm ) region_overlap_conditional ( rc ) region_overlap_marginal ( rm ) pearson_correlation_conditional ( cc ) fold_enrichment_conditional ( fc ) The particulars of the tests are discussed in the code ( docstrings ). -fg --filter_covering: Filter covering bed file by a group name. -fd --filter_covered : Filter covered bed file by a group name. In the bed4 format, the 4th column stores a label for the given region. When one of the above two options are chosen, the file will filter out regions that dont match 'name' as given. Defaults to no filtering. -B --force_binary: Treat the input data as binary regardless of whether or not there is a value. This is useful for bed files which have value's for the purpose of plotting at the UCSC genome browser ( ie all 1000 ). --min_overlap_fraction: This is only used for the region overlap test. If the covered region isn't overlapped by at least (X*100)% of the covering region, then we dont consider this an overlap. Defaults to 0.0. That is, if the covered region is overlapped by even 1 basepair, then we say it is covered. -o --output_file: A file to output all non-error output into. Will append to this file if it already exists. defaults to standard out. """ def main(): try: long_args = [ "help", "feature-1=", "feature-2=", "domain=", "region-fraction=", "subregion-fraction=", "num-samples=", "test=", "output-file=", "filter-covering=", "filter-covered=", "force-binary", "verbose", "min-overlap-fraction=" ] opts, args = getopt.getopt(sys.argv[1:], "1:2:d:r:s:n:t:o:fg:fd:Bv", long_args) except getopt.GetoptError, err: usage() print "Error: \n", str(err) print sys.exit(2) ## make percent_basepair_overlap the default test type test_type = 'bc' filter_covering = None filter_covered = None force_binary = False min_overlap_fraction = 0.0 for o, a in opts: if o == "-v": global verbose verbose = True base_types.verbose = True elif o in ("-h", "--help"): usage() sys.exit() elif o in ("-1", "--feature-1"): covered_file = open(a) elif o in ("-2", "--feature-2"): covering_file = open(a) elif o in ("-d", "--domain"): lengths_file = open(a) elif o in ("-r", "--region-fraction"): region_fraction = float(a) elif o in ("-s", "--subregion-fraction"): subregion_fraction = float(a) elif o in ("-n", "--num-samples"): num_samples = int(a) elif o in ("-t", "--test"): test_type = a elif o in ("-B", "--force-binary"): force_binary = True elif o in ("-o", "--output_file"): global output output = open(a, 'a') elif o in ("-fg", "--filter-covering"): filter_covering = a elif o in ("-fd", "--filter-covered"): filter_covered = a elif o in ("--min-overlap-fraction", ): min_overlap_fraction = float( a ) if min_overlap_fraction < 0 or min_overlap_fraction > 1.0: raise ValueError, "min_overlap_fraction must be between 0 and 1.0, inclusive." else: assert False, "unhandled option %s" % o try: covered_file, covering_file, lengths_file, region_fraction, num_samples except Exception, inst: usage() sys.exit() if verbose: import time startTime = time.time() if covered_file.name.endswith(".wig"): coveredAnnotations = parse_wiggle_file(covered_file, lengths_file, force_binary) else: coveredAnnotations = parse_bed_file(covered_file, lengths_file, filter_covered, force_binary) if covering_file.name.endswith(".wig"): coveringAnnotations = parse_wiggle_file(covering_file, lengths_file, force_binary) else: coveringAnnotations = parse_bed_file(covering_file, lengths_file, filter_covering, force_binary) covered_file.close() covering_file.close() lengths_file.close() if verbose: print "Input Files Parse Time: ", time.time() - startTime, "\n" startTime = time.time() if not vars().has_key('region_fraction'): raise ValueError, 'The region_fraction setting is not set - it is mandatory.' #### Tests that work as for continuous data *or* for binary data if test_type in ( 'pearson_correlation_conditional', 'cc' ): conditional_pearson_correlation( coveringAnnotations, coveredAnnotations, region_fraction, num_samples ) elif test_type in ( 'fold_enrichment_conditional', 'fc' ): if not vars().has_key('subregion_fraction'): raise ValueError, 'The fold_enrichment_conditional test requires that the subregion_fraction option be set.' conditional_mean_fold_enrichment( coveringAnnotations, coveredAnnotations, region_fraction, subregion_fraction, num_samples ) else: #### Tests that ONLY work for binary data # test for the correct object if type(coveredAnnotations.values()[0]) != binary_region \ or type(coveringAnnotations.values()[0]) != binary_region: raise TypeError, "The %s test only works for binary data ( You can force this with the -B option )" % test_type if test_type in ( 'basepair_overlap_marginal', 'bm' ) : marginal_bp_overlap_stat( coveringAnnotations, coveredAnnotations, region_fraction, num_samples ) elif test_type in ( 'basepair_overlap_conditional', 'bc' ) : conditional_bp_overlap_stat( coveringAnnotations, coveredAnnotations, region_fraction, num_samples ) elif test_type in ( 'region_overlap_marginal', 'rm' ): if not vars().has_key('subregion_fraction'): raise ValueError, 'The region_overlap_marginal test requires that the subregion_fraction option be set.' marginal_resample_region_overlap_stat( coveringAnnotations, coveredAnnotations, region_fraction, subregion_fraction, num_samples, min_overlap_fraction ) elif test_type in ( 'region_overlap_conditional', 'rc' ): if not vars().has_key('subregion_fraction'): raise ValueError, 'The region_overlap_conditional test requires that the subregion_fraction option be set.' conditional_resample_region_overlap_stat( coveringAnnotations, coveredAnnotations, region_fraction, subregion_fraction, num_samples, min_overlap_fraction ) else: raise ValueError, "Unrecognized test '%s'" % test_type if verbose: print >> output, "\nExecution Time: ", time.time()-startTime output.close() if __name__ == "__main__": main()