Mercurial > repos > iuc > cwpair2
comparison cwpair2_util.py @ 0:8600bfe7ed52 draft
planemo upload for repository https://github.com/galaxyproject/tools-iuc/tree/master/tools/cwpair2 commit e96df94dba60050fa28aaf55b5bb095717a5f260
author | iuc |
---|---|
date | Tue, 22 Dec 2015 17:03:46 -0500 |
parents | |
children | abc464ca7260 |
comparison
equal
deleted
inserted
replaced
-1:000000000000 | 0:8600bfe7ed52 |
---|---|
1 import bisect | |
2 import csv | |
3 import os | |
4 import sys | |
5 import traceback | |
6 import matplotlib | |
7 matplotlib.use('Agg') | |
8 from matplotlib import pyplot | |
9 | |
10 # Data outputs | |
11 DETAILS = 'D' | |
12 MATCHED_PAIRS = 'MP' | |
13 ORPHANS = 'O' | |
14 # Data output formats | |
15 GFF_EXT = 'gff' | |
16 TABULAR_EXT = 'tabular' | |
17 # Statistics historgrams output directory. | |
18 HISTOGRAM = 'H' | |
19 # Statistics outputs | |
20 FINAL_PLOTS = 'F' | |
21 PREVIEW_PLOTS = 'P' | |
22 STATS_GRAPH = 'C' | |
23 | |
24 # Graph settings. | |
25 COLORS = 'krg' | |
26 Y_LABEL = 'Peak-pair counts' | |
27 X_LABEL = 'Peak-pair distance (bp)' | |
28 TICK_WIDTH = 3 | |
29 ADJUST = [0.140, 0.9, 0.9, 0.1] | |
30 PLOT_FORMAT = 'pdf' | |
31 pyplot.rc('xtick.major', size=10.00) | |
32 pyplot.rc('ytick.major', size=10.00) | |
33 pyplot.rc('lines', linewidth=4.00) | |
34 pyplot.rc('axes', linewidth=3.00) | |
35 pyplot.rc('font', family='Bitstream Vera Sans', size=32.0) | |
36 | |
37 | |
38 class FrequencyDistribution(object): | |
39 | |
40 def __init__(self, start, end, binsize=10, d=None): | |
41 self.start = start | |
42 self.end = end | |
43 self.dist = d or {} | |
44 self.binsize = binsize | |
45 | |
46 def get_bin(self, x): | |
47 """ | |
48 Returns the bin in which a data point falls | |
49 """ | |
50 return self.start + (x-self.start) // self.binsize * self.binsize + self.binsize/2.0 | |
51 | |
52 def add(self, x): | |
53 x = self.get_bin(x) | |
54 self.dist[x] = self.dist.get(x, 0) + 1 | |
55 | |
56 def graph_series(self): | |
57 x = [] | |
58 y = [] | |
59 for i in range(self.start, self.end, self.binsize): | |
60 center = self.get_bin(i) | |
61 x.append(center) | |
62 y.append(self.dist.get(center, 0)) | |
63 return x, y | |
64 | |
65 def mode(self): | |
66 return max(self.dist.items(), key=lambda data: data[1])[0] | |
67 | |
68 def size(self): | |
69 return sum(self.dist.values()) | |
70 | |
71 | |
72 def stop_err(msg): | |
73 sys.stderr.write(msg) | |
74 sys.exit(1) | |
75 | |
76 | |
77 def distance(peak1, peak2): | |
78 return (peak2[1]+peak2[2])/2 - (peak1[1]+peak1[2])/2 | |
79 | |
80 | |
81 def gff_row(cname, start, end, score, source, type='.', strand='.', phase='.', attrs={}): | |
82 return (cname, source, type, start, end, score, strand, phase, gff_attrs(attrs)) | |
83 | |
84 | |
85 def gff_attrs(d): | |
86 if not d: | |
87 return '.' | |
88 return ';'.join('%s=%s' % item for item in d.items()) | |
89 | |
90 | |
91 def parse_chromosomes(reader): | |
92 # This version of cwpair2 accepts only gff format as input. | |
93 chromosomes = {} | |
94 reader.next() | |
95 for line in reader: | |
96 cname, junk, junk, start, end, value, strand, junk, junk = line | |
97 start = int(start) | |
98 end = int(end) | |
99 value = float(value) | |
100 if cname not in chromosomes: | |
101 chromosomes[cname] = [] | |
102 peaks = chromosomes[cname] | |
103 peaks.append((strand, start, end, value)) | |
104 return chromosomes | |
105 | |
106 | |
107 def perc95(chromosomes): | |
108 """ | |
109 Returns the 95th percentile value of the given chromosomes. | |
110 """ | |
111 values = [] | |
112 for peaks in chromosomes.values(): | |
113 for peak in peaks: | |
114 values.append(peak[3]) | |
115 values.sort() | |
116 # Get 95% value | |
117 return values[int(len(values)*0.95)] | |
118 | |
119 | |
120 def filter(chromosomes, threshold=0.05): | |
121 """ | |
122 Filters the peaks to those above a threshold. Threshold < 1.0 is interpreted | |
123 as a proportion of the maximum, >=1.0 as an absolute value. | |
124 """ | |
125 if threshold < 1: | |
126 p95 = perc95(chromosomes) | |
127 threshold = p95 * threshold | |
128 # Make the threshold a proportion of the | |
129 for cname, peaks in chromosomes.items(): | |
130 chromosomes[cname] = [peak for peak in peaks if peak[3] > threshold] | |
131 | |
132 | |
133 def split_strands(chromosome): | |
134 watson = [peak for peak in chromosome if peak[0] == '+'] | |
135 crick = [peak for peak in chromosome if peak[0] == '-'] | |
136 return watson, crick | |
137 | |
138 | |
139 def all_pair_distribution(chromosomes, up_distance, down_distance, binsize): | |
140 dist = FrequencyDistribution(-up_distance, down_distance, binsize=binsize) | |
141 for cname, data in chromosomes.items(): | |
142 watson, crick = split_strands(data) | |
143 crick.sort(key=lambda data: float(data[1])) | |
144 keys = make_keys(crick) | |
145 for peak in watson: | |
146 for cpeak in get_window(crick, peak, up_distance, down_distance, keys): | |
147 dist.add(distance(peak, cpeak)) | |
148 return dist | |
149 | |
150 | |
151 def make_keys(crick): | |
152 return [(data[1] + data[2])//2 for data in crick] | |
153 | |
154 | |
155 def get_window(crick, peak, up_distance, down_distance, keys=None): | |
156 """ | |
157 Returns a window of all crick peaks within a distance of a watson peak. | |
158 crick strand MUST be sorted by distance | |
159 """ | |
160 strand, start, end, value = peak | |
161 midpoint = (start + end) // 2 | |
162 lower = midpoint - up_distance | |
163 upper = midpoint + down_distance | |
164 keys = keys or make_keys(crick) | |
165 start_index = bisect.bisect_left(keys, lower) | |
166 end_index = bisect.bisect_right(keys, upper) | |
167 return [cpeak for cpeak in crick[start_index:end_index]] | |
168 | |
169 | |
170 def match_largest(window, peak): | |
171 if not window: | |
172 return None | |
173 return max(window, key=lambda cpeak: cpeak[3]) | |
174 | |
175 | |
176 def match_closest(window, peak): | |
177 if not window: | |
178 return None | |
179 | |
180 def key(cpeak): | |
181 d = distance(peak, cpeak) | |
182 # Search negative distances last | |
183 if d < 0: | |
184 # And then prefer less negative distances | |
185 d = 10000 - d | |
186 return d | |
187 return min(window, key=key) | |
188 | |
189 | |
190 def match_mode(window, peak, mode): | |
191 if not window: | |
192 return None | |
193 return min(window, key=lambda cpeak: abs(distance(peak, cpeak)-mode)) | |
194 | |
195 METHODS = {'mode': match_mode, 'closest': match_closest, 'largest': match_largest} | |
196 | |
197 | |
198 def frequency_plot(freqs, fname, labels=[], title=''): | |
199 pyplot.clf() | |
200 pyplot.figure(figsize=(10, 10)) | |
201 for i, freq in enumerate(freqs): | |
202 x, y = freq.graph_series() | |
203 pyplot.plot(x, y, '%s-' % COLORS[i]) | |
204 if len(freqs) > 1: | |
205 pyplot.legend(labels) | |
206 pyplot.xlim(freq.start, freq.end) | |
207 pyplot.ylim(ymin=0) | |
208 pyplot.ylabel(Y_LABEL) | |
209 pyplot.xlabel(X_LABEL) | |
210 pyplot.subplots_adjust(left=ADJUST[0], right=ADJUST[1], top=ADJUST[2], bottom=ADJUST[3]) | |
211 # Get the current axes | |
212 ax = pyplot.gca() | |
213 for l in ax.get_xticklines() + ax.get_yticklines(): | |
214 l.set_markeredgewidth(TICK_WIDTH) | |
215 pyplot.savefig(fname) | |
216 | |
217 | |
218 def create_directories(): | |
219 # Output histograms in pdf. | |
220 os.mkdir(HISTOGRAM) | |
221 os.mkdir('data_%s' % DETAILS) | |
222 os.mkdir('data_%s' % ORPHANS) | |
223 os.mkdir('data_%s' % MATCHED_PAIRS) | |
224 | |
225 | |
226 def process_file(dataset_path, galaxy_hid, method, threshold, up_distance, | |
227 down_distance, binsize, output_files): | |
228 if method == 'all': | |
229 match_methods = METHODS.keys() | |
230 else: | |
231 match_methods = [method] | |
232 statistics = [] | |
233 for match_method in match_methods: | |
234 stats = perform_process(dataset_path, | |
235 galaxy_hid, | |
236 match_method, | |
237 threshold, | |
238 up_distance, | |
239 down_distance, | |
240 binsize, | |
241 output_files) | |
242 statistics.append(stats) | |
243 if output_files == 'all' and method == 'all': | |
244 frequency_plot([s['dist'] for s in statistics], | |
245 statistics[0]['graph_path'], | |
246 labels=METHODS.keys()) | |
247 return statistics | |
248 | |
249 | |
250 def perform_process(dataset_path, galaxy_hid, method, threshold, up_distance, | |
251 down_distance, binsize, output_files): | |
252 output_details = output_files in ["all", "matched_pair_orphan_detail"] | |
253 output_plots = output_files in ["all"] | |
254 output_orphans = output_files in ["all", "matched_pair_orphan", "matched_pair_orphan_detail"] | |
255 # Keep track of statistics for the output file | |
256 statistics = {} | |
257 input = csv.reader(open(dataset_path, 'rt'), delimiter='\t') | |
258 fpath, fname = os.path.split(dataset_path) | |
259 statistics['fname'] = '%s: data %s' % (method, str(galaxy_hid)) | |
260 statistics['dir'] = fpath | |
261 if threshold >= 1: | |
262 filter_string = 'fa%d' % threshold | |
263 else: | |
264 filter_string = 'f%d' % (threshold * 100) | |
265 fname = '%s_%su%dd%d_on_data_%s' % (method, filter_string, up_distance, down_distance, galaxy_hid) | |
266 | |
267 def make_histogram_path(output_type, fname): | |
268 return os.path.join(HISTOGRAM, 'histogram_%s_%s.%s' % (output_type, fname, PLOT_FORMAT)) | |
269 | |
270 def make_path(output_type, extension, fname): | |
271 # Returns the full path for an output. | |
272 return os.path.join(output_type, '%s_%s.%s' % (output_type, fname, extension)) | |
273 | |
274 def td_writer(output_type, extension, fname): | |
275 # Returns a tab-delimited writer for a specified output. | |
276 output_file_path = make_path(output_type, extension, fname) | |
277 return csv.writer(open(output_file_path, 'wt'), delimiter='\t') | |
278 | |
279 try: | |
280 chromosomes = parse_chromosomes(input) | |
281 except Exception: | |
282 stop_err('Unable to parse file "%s".\n%s' % (dataset_path, traceback.format_exc())) | |
283 if output_details: | |
284 # Details | |
285 detailed_output = td_writer('data_%s' % DETAILS, TABULAR_EXT, fname) | |
286 detailed_output.writerow(('chrom', 'start', 'end', 'value', 'strand') * 2 + ('midpoint', 'c-w reads sum', 'c-w distance (bp)')) | |
287 if output_plots: | |
288 # Final Plot | |
289 final_plot_path = make_histogram_path(FINAL_PLOTS, fname) | |
290 if output_orphans: | |
291 # Orphans | |
292 orphan_output = td_writer('data_%s' % ORPHANS, TABULAR_EXT, fname) | |
293 orphan_output.writerow(('chrom', 'strand', 'start', 'end', 'value')) | |
294 if output_plots: | |
295 # Preview Plot | |
296 preview_plot_path = make_histogram_path(PREVIEW_PLOTS, fname) | |
297 # Matched Pairs. | |
298 matched_pairs_output = td_writer('data_%s' % MATCHED_PAIRS, GFF_EXT, fname) | |
299 statistics['stats_path'] = 'statistics.%s' % TABULAR_EXT | |
300 if output_plots: | |
301 statistics['graph_path'] = make_histogram_path(STATS_GRAPH, fname) | |
302 statistics['perc95'] = perc95(chromosomes) | |
303 if threshold > 0: | |
304 # Apply filter | |
305 filter(chromosomes, threshold) | |
306 if method == 'mode': | |
307 freq = all_pair_distribution(chromosomes, up_distance, down_distance, binsize) | |
308 mode = freq.mode() | |
309 statistics['preview_mode'] = mode | |
310 if output_plots: | |
311 frequency_plot([freq], preview_plot_path, title='Preview frequency plot') | |
312 else: | |
313 statistics['preview_mode'] = 'NA' | |
314 dist = FrequencyDistribution(-up_distance, down_distance, binsize=binsize) | |
315 orphans = 0 | |
316 # x will be used to archive the summary dataset | |
317 x = [] | |
318 for cname, chromosome in chromosomes.items(): | |
319 # Each peak is (strand, start, end, value) | |
320 watson, crick = split_strands(chromosome) | |
321 # Sort by value of each peak | |
322 watson.sort(key=lambda data: -float(data[3])) | |
323 # Sort by position to facilitate binary search | |
324 crick.sort(key=lambda data: float(data[1])) | |
325 keys = make_keys(crick) | |
326 for peak in watson: | |
327 window = get_window(crick, peak, up_distance, down_distance, keys) | |
328 if method == 'mode': | |
329 match = match_mode(window, peak, mode) | |
330 else: | |
331 match = METHODS[method](window, peak) | |
332 if match: | |
333 midpoint = (match[1] + match[2] + peak[1] + peak[2]) // 4 | |
334 d = distance(peak, match) | |
335 dist.add(d) | |
336 # Simple output in gff format. | |
337 x.append(gff_row(cname, | |
338 source='cwpair', | |
339 start=midpoint, | |
340 end=midpoint + 1, | |
341 score=peak[3] + match[3], | |
342 attrs={'cw_distance': d})) | |
343 if output_details: | |
344 detailed_output.writerow((cname, | |
345 peak[1], | |
346 peak[2], | |
347 peak[3], | |
348 '+', | |
349 cname, | |
350 match[1], | |
351 match[2], | |
352 match[3], '-', | |
353 midpoint, | |
354 peak[3]+match[3], | |
355 d)) | |
356 i = bisect.bisect_left(keys, (match[1]+match[2])/2) | |
357 del crick[i] | |
358 del keys[i] | |
359 else: | |
360 if output_orphans: | |
361 orphan_output.writerow((cname, peak[0], peak[1], peak[2], peak[3])) | |
362 # Keep track of orphans for statistics. | |
363 orphans += 1 | |
364 # Remaining crick peaks are orphans | |
365 if output_orphans: | |
366 for cpeak in crick: | |
367 orphan_output.writerow((cname, cpeak[0], cpeak[1], cpeak[2], cpeak[3])) | |
368 # Keep track of orphans for statistics. | |
369 orphans += len(crick) | |
370 # Sort output descending by score. | |
371 x.sort(key=lambda data: float(data[5]), reverse=True) | |
372 # Writing a summary to gff format file | |
373 for row in x: | |
374 row_tmp = list(row) | |
375 # Dataset in tuple cannot be modified in Python, so row will | |
376 # be converted to list format to add 'chr'. | |
377 if row_tmp[0] == "999": | |
378 row_tmp[0] = 'chrM' | |
379 elif row_tmp[0] == "998": | |
380 row_tmp[0] = 'chrY' | |
381 elif row_tmp[0] == "997": | |
382 row_tmp[0] = 'chrX' | |
383 else: | |
384 row_tmp[0] = row_tmp[0] | |
385 # Print row_tmp. | |
386 matched_pairs_output.writerow(row_tmp) | |
387 statistics['paired'] = dist.size() * 2 | |
388 statistics['orphans'] = orphans | |
389 statistics['final_mode'] = dist.mode() | |
390 if output_plots: | |
391 frequency_plot([dist], final_plot_path, title='Frequency distribution') | |
392 statistics['dist'] = dist | |
393 return statistics |