def compute_pval_thresh(real_pvals, null_pvals, verbose):
''' given an observed p-value and a collection of real and random p-values,
calculate the empirical false discovery rate (FDR) for the p-value.
returns: dictionary of floats'''
pval_thresh = dict()
if verbose:
label = ">> computing thresholds: "
progress = ProgressBar(len(real_pvals), label=label)
for obs_pval in real_pvals:
if verbose: progress.next()
if obs_pval in pval_thresh: continue
real_pvals_better = num_pvals_better(obs_pval, real_pvals, verbose)
null_pvals_better = num_pvals_better(obs_pval, null_pvals, verbose)
if real_pvals_better == 0:
qval = 0.0
else:
qval = null_pvals_better / real_pvals_better
pval_thresh[obs_pval] = qval
if verbose: progress.end()
return pval_thresh
def find_trimmed_peaks(peak_data, genome, chrom, trackname, max_width, verbose):
''' trim peaks to min width with max signal.
returns: data with same form as peak_data, trimmed'''
# TODO: recalculate pvalue for the trimmed and return that
# XXX: there is likely a faster way to do this with native numpy
# functions. speed scales exponentially with window size, may want to
# add a "max_width" param to prevent trimming of peaks e.g. >1000 bp
trimmed_peaks = []
if verbose:
progress = ProgressBar(len(peak_data),
label=">> trimming peaks: ")
for peak_start, peak_end, pval in peak_data:
if peak_end - peak_start > max_width:
trimmed_peaks.append((peak_start, peak_end, pval))
continue
if verbose: progress.next()
continuous = genome[chrom.name][peak_start:peak_end, trackname]
trim_peak_start, trim_peak_end = max_contiguous_signal(continuous)
# update coords with new start & end - set end first
peak_end = peak_start + trim_peak_end
peak_start = peak_start + trim_peak_start
trimmed_peaks.append((peak_start, peak_end, pval))
# clean up progress bar
if verbose: progress.end()
return trimmed_peaks
def identify_local_peaks(chromosome, track_index, peak_width, lambda_base,
log_pval_thresh, local_lambda, shuffle_data, verbose):
''' identify peaks within a chromsome.
returns: list of (start, end, pvalue) tuples'''
local_peaks = []
# pdb.set_trace()
if verbose:
progress = ProgressBar(num_supercontigs(chromosome),
label=">> calling peaks: ")
for supercontig, continuous in chromosome.itercontinuous():
if verbose: progress.next()
# load data into memory
track_continuous = continuous[:, track_index]
# shuffle the data if requested
if shuffle_data: shuffle(track_continuous)
# generate window ranges i.e. peak widths
peak_range = xrange(0, len(track_continuous), int(peak_width))
for peak_start in peak_range:
# set the peak end
peak_end = int(peak_start + peak_width)
peak_data = track_continuous[peak_start:peak_end]
# skip the window if it is empty
if isnan(nansum(peak_data)): continue
# calculate local lambdas and choose max
if local_lambda:
lambda_values = calc_local_lambdas(peak_width,
track_continuous,
peak_start, peak_end)
lambda_values.append(lambda_base)
lambda_value = max(lambda_values)
else:
lambda_value = lambda_base
# initial pvalue for the region
log_pvalue = calc_log_pvalue(peak_data, lambda_value,
verbose)
# check that the pvalue is defined & that it meets the
# threshold
if not log_pvalue or log_pvalue < log_pval_thresh: continue
# calculate chromosomal coords relative to current
# contig coords
chrom_start = supercontig.start + peak_start
chrom_end = supercontig.start + peak_end
# save current peak
fields = (chrom_start, chrom_end, log_pvalue)
local_peaks.append(fields)
# reset lambda value to base
lambda_value = lambda_base
if verbose: progress.end()
return local_peaks
