view cwpair2_util.py @ 5:2e0ddcc726f9 draft

Uploaded

import bisect
import csv
import os
import sys
import traceback
import matplotlib
matplotlib.use('Agg')
from matplotlib import pyplot

DETAILS = 'D'
FINAL_PLOTS = 'F'
ORPHANS = 'O'
PREVIEW_PLOTS = 'P'
SIMPLES = 'S'
STATS_GRAPH = 'C'
GFF_EXT = 'gff'
TABULAR_EXT = 'tabular'

# Graph settings.
COLORS = 'krg'
Y_LABEL = 'Peak-pair counts'
X_LABEL = 'Peak-pair distance (bp)'
TICK_WIDTH = 3
ADJUST = [0.140, 0.9, 0.9, 0.1]
PLOT_FORMAT = 'pdf'
pyplot.rc('xtick.major', size=10.00)
pyplot.rc('ytick.major', size=10.00)
pyplot.rc('lines', linewidth=4.00)
pyplot.rc('axes', linewidth=3.00)
pyplot.rc('font', family='Bitstream Vera Sans', size=32.0)


class FrequencyDistribution(object):

    def __init__(self, start, end, binsize=10, d=None):
        self.start = start
        self.end = end
        self.dist = d or {}
        self.binsize = binsize

    def get_bin(self, x):
        """
        Returns the bin in which a data point falls
        """
        return self.start + (x-self.start) // self.binsize * self.binsize + self.binsize/2.0

    def add(self, x):
        x = self.get_bin(x)
        self.dist[x] = self.dist.get(x, 0) + 1

    def graph_series(self):
        x = []
        y = []
        for i in range(self.start, self.end, self.binsize):
            center = self.get_bin(i)
            x.append(center)
            y.append(self.dist.get(center, 0))
        return x, y

    def mode(self):
        return max(self.dist.items(), key=lambda data: data[1])[0]

    def size(self):
        return sum(self.dist.values())


def stop_err(msg):
    sys.stderr.write(msg)
    sys.exit(1)


def distance(peak1, peak2):
    return (peak2[1]+peak2[2])/2 - (peak1[1]+peak1[2])/2


def gff_row(cname, start, end, score, source, type='.', strand='.', phase='.', attrs={}):
    return (cname, source, type, start, end, score, strand, phase, gff_attrs(attrs))


def gff_attrs(d):
    if not d:
        return '.'
    return ';'.join('%s=%s' % item for item in d.items())


def parse_chromosomes(reader):
    # This version of cwpair2 accepts only gff format as input.
    chromosomes = {}
    reader.next()
    for line in reader:
        cname, junk, junk, start, end, value, strand, junk, junk = line
        start = int(start)
        end = int(end)
        value = float(value)
        if cname not in chromosomes:
            chromosomes[cname] = []
        peaks = chromosomes[cname]
        peaks.append((strand, start, end, value))
    return chromosomes


def perc95(chromosomes):
    """
    Returns the 95th percentile value of the given chromosomes.
    """
    values = []
    for peaks in chromosomes.values():
        for peak in peaks:
            values.append(peak[3])
    values.sort()
    # Get 95% value
    return values[int(len(values)*0.95)]


def filter(chromosomes, threshold=0.05):
    """
    Filters the peaks to those above a threshold. Threshold < 1.0 is interpreted
    as a proportion of the maximum, >=1.0 as an absolute value.
    """
    if threshold < 1:
        p95 = perc95(chromosomes)
        threshold = p95 * threshold
        # Make the threshold a proportion of the
    for cname, peaks in chromosomes.items():
        chromosomes[cname] = [peak for peak in peaks if peak[3] > threshold]


def split_strands(chromosome):
    watson = [peak for peak in chromosome if peak[0] == '+']
    crick = [peak for peak in chromosome if peak[0] == '-']
    return watson, crick


def all_pair_distribution(chromosomes, up_distance, down_distance, binsize):
    dist = FrequencyDistribution(-up_distance, down_distance, binsize=binsize)
    for cname, data in chromosomes.items():
        watson, crick = split_strands(data)
        crick.sort(key=lambda data: float(data[1]))
        keys = make_keys(crick)
        for peak in watson:
            for cpeak in get_window(crick, peak, up_distance, down_distance, keys):
                dist.add(distance(peak, cpeak))
    return dist


def make_keys(crick):
    return [(data[1] + data[2])//2 for data in crick]


def get_window(crick, peak, up_distance, down_distance, keys=None):
    """
    Returns a window of all crick peaks within a distance of a watson peak.
    crick strand MUST be sorted by distance
    """
    strand, start, end, value = peak
    midpoint = (start + end) // 2
    lower = midpoint - up_distance
    upper = midpoint + down_distance
    keys = keys or make_keys(crick)
    start_index = bisect.bisect_left(keys, lower)
    end_index = bisect.bisect_right(keys, upper)
    return [cpeak for cpeak in crick[start_index:end_index]]


def match_largest(window, peak):
    if not window:
        return None
    return max(window, key=lambda cpeak: cpeak[3])


def match_closest(window, peak):
    if not window:
        return None

    def key(cpeak):
        d = distance(peak, cpeak)
        # Search negative distances last
        if d < 0:
            # And then prefer less negative distances
            d = 10000 - d
        return d
    return min(window, key=key)


def match_mode(window, peak, mode):
    if not window:
        return None
    return min(window, key=lambda cpeak: abs(distance(peak, cpeak)-mode))

METHODS = {'mode': match_mode, 'closest': match_closest, 'largest': match_largest}


def frequency_plot(freqs, fname, labels=[], title=''):
    pyplot.clf()
    pyplot.figure(figsize=(10, 10))
    for i, freq in enumerate(freqs):
        x, y = freq.graph_series()
        pyplot.plot(x, y, '%s-' % COLORS[i])
    if len(freqs) > 1:
        pyplot.legend(labels)
    pyplot.xlim(freq.start, freq.end)
    pyplot.ylim(ymin=0)
    pyplot.ylabel(Y_LABEL)
    pyplot.xlabel(X_LABEL)
    pyplot.subplots_adjust(left=ADJUST[0], right=ADJUST[1], top=ADJUST[2], bottom=ADJUST[3])
    # Get the current axes
    ax = pyplot.gca()
    for l in ax.get_xticklines() + ax.get_yticklines():
        l.set_markeredgewidth(TICK_WIDTH)
    pyplot.savefig(fname)


def create_directories(method):
    if method == 'all':
        match_methods = METHODS.keys()
    else:
        match_methods = [method]
    for match_method in match_methods:
        os.mkdir('%s_%s' % (match_method, DETAILS))
        os.mkdir('%s_%s' % (match_method, FINAL_PLOTS))
        os.mkdir('%s_%s' % (match_method, ORPHANS))
        os.mkdir('%s_%s' % (match_method, PREVIEW_PLOTS))
        os.mkdir('%s_%s' % (match_method, SIMPLES))
        os.mkdir('%s_%s' % (match_method, STATS_GRAPH))


def process_file(dataset_path, galaxy_hid, method, threshold, up_distance,
                 down_distance, binsize, output_files, sort_score):
    if method == 'all':
        match_methods = METHODS.keys()
    else:
        match_methods = [method]
    statistics = []
    for match_method in match_methods:
        stats = perform_process(dataset_path,
                                galaxy_hid,
                                match_method,
                                threshold,
                                up_distance,
                                down_distance,
                                binsize,
                                output_files,
                                sort_score)
        statistics.append(stats)
    if output_files == 'all' and method == 'all':
        frequency_plot([s['dist'] for s in statistics],
                       statistics[0]['graph_path'],
                       labels=METHODS.keys())
    return statistics


def perform_process(dataset_path, galaxy_hid, method, threshold, up_distance,
                    down_distance, binsize, output_files, sort_score):
    output_details = output_files in ["all", "simple_orphan_detail"]
    output_plots = output_files in ["all"]
    output_orphans = output_files in ["all", "simple_orphan", "simple_orphan_detail"]
    # Keep track of statistics for the output file
    statistics = {}
    input = csv.reader(open(dataset_path, 'rt'), delimiter='\t')
    fpath, fname = os.path.split(dataset_path)
    statistics['fname'] = '%s: data %s' % (method, str(galaxy_hid))
    statistics['dir'] = fpath
    if threshold >= 1:
        filter_string = 'fa%d' % threshold
    else:
        filter_string = 'f%d' % (threshold * 100)
    fname = 'data_%s_%su%dd%db%d' % (galaxy_hid, filter_string, up_distance, down_distance, binsize)

    def make_path(output_type, extension=TABULAR_EXT):
        # Returns the full path for a certain output.
        return os.path.join(output_type, '%s_%s.%s' % (output_type, fname, extension))

    def td_writer(output_type, extension=TABULAR_EXT):
        # Returns a tab-delimited writer for a specified output.
        output_file_path = make_path(output_type, extension)
        return csv.writer(open(output_file_path, 'wt'), delimiter='\t')

    try:
        chromosomes = parse_chromosomes(input)
    except Exception:
        stop_err('Unable to parse file "%s".\n%s' % (dataset_path, traceback.format_exc()))
    if output_details:
        # Details
        detailed_output = td_writer('%s_%s' % (method, DETAILS), extension=TABULAR_EXT)
        detailed_output.writerow(('chrom', 'start', 'end', 'value', 'strand') * 2 + ('midpoint', 'c-w reads sum', 'c-w distance (bp)'))
    if output_plots:
        # Final Plot
        final_plot_path = make_path('%s_%s' % (method, FINAL_PLOTS), PLOT_FORMAT)
    if output_orphans:
        # Orphans
        orphan_output = td_writer('%s_%s' % (method, ORPHANS), extension=TABULAR_EXT)
        orphan_output.writerow(('chrom', 'strand', 'start', 'end', 'value'))
    if output_plots:
        # Preview Plot
        preview_plot_path = make_path('%s_%s' % (method, PREVIEW_PLOTS), PLOT_FORMAT)
    # Simple
    simple_output = td_writer('%s_%s' % (method, SIMPLES), extension=GFF_EXT)
    statistics['stats_path'] = 'statistics.%s' % TABULAR_EXT
    if output_plots:
        statistics['graph_path'] = make_path('%s_%s' % (method, STATS_GRAPH), PLOT_FORMAT)
    statistics['perc95'] = perc95(chromosomes)
    if threshold > 0:
        # Apply filter
        filter(chromosomes, threshold)
    if method == 'mode':
        freq = all_pair_distribution(chromosomes, up_distance, down_distance, binsize)
        mode = freq.mode()
        statistics['preview_mode'] = mode
        if output_plots:
            frequency_plot([freq], preview_plot_path, title='Preview frequency plot')
    else:
        statistics['preview_mode'] = 'NA'
    dist = FrequencyDistribution(-up_distance, down_distance, binsize=binsize)
    orphans = 0
    # x will be used to archive the summary dataset
    x = []
    for cname, chromosome in chromosomes.items():
        # Each peak is (strand, start, end, value)
        watson, crick = split_strands(chromosome)
        # Sort by value of each peak
        watson.sort(key=lambda data: -float(data[3]))
        # Sort by position to facilitate binary search
        crick.sort(key=lambda data: float(data[1]))
        keys = make_keys(crick)
        for peak in watson:
            window = get_window(crick, peak, up_distance, down_distance, keys)
            if method == 'mode':
                match = match_mode(window, peak, mode)
            else:
                match = METHODS[method](window, peak)
            if match:
                midpoint = (match[1] + match[2] + peak[1] + peak[2]) // 4
                d = distance(peak, match)
                dist.add(d)
                # Simple output in gff format.
                x.append(gff_row(cname,
                                 source='cwpair',
                                 start=midpoint,
                                 end=midpoint + 1,
                                 score=peak[3] + match[3],
                                 attrs={'cw_distance': d}))
                if output_details:
                    detailed_output.writerow((cname,
                                              peak[1],
                                              peak[2],
                                              peak[3],
                                              '+',
                                              cname,
                                              match[1],
                                              match[2],
                                              match[3], '-',
                                              midpoint,
                                              peak[3]+match[3],
                                              d))
                i = bisect.bisect_left(keys, (match[1]+match[2])/2)
                del crick[i]
                del keys[i]
            else:
                if output_orphans:
                    orphan_output.writerow((cname, peak[0], peak[1], peak[2], peak[3]))
                # Keep track of orphans for statistics.
                orphans += 1
        # Remaining crick peaks are orphans
        if output_orphans:
            for cpeak in crick:
                orphan_output.writerow((cname, cpeak[0], cpeak[1], cpeak[2], cpeak[3]))
        # Keep track of orphans for statistics.
        orphans += len(crick)
    # Sort output by score if specified.
    if sort_score == "desc":
        x.sort(key=lambda data: float(data[5]), reverse=True)
    elif sort_score == "asc":
        x.sort(key=lambda data: float(data[5]))
    # Writing a summary to gff format file
    for row in x:
        row_tmp = list(row)
        # Dataset in tuple cannot be modified in Python, so row will
        # be converted to list format to add 'chr'.
        if row_tmp[0] == "999":
            row_tmp[0] = 'chrM'
        elif row_tmp[0] == "998":
            row_tmp[0] = 'chrY'
        elif row_tmp[0] == "997":
            row_tmp[0] = 'chrX'
        else:
            row_tmp[0] = row_tmp[0]
        # Print row_tmp.
        simple_output.writerow(row_tmp)
    statistics['paired'] = dist.size() * 2
    statistics['orphans'] = orphans
    statistics['final_mode'] = dist.mode()
    if output_plots:
        frequency_plot([dist], final_plot_path, title='Frequency distribution')
    statistics['dist'] = dist
    return statistics