0
|
1 #!/usr/bin/env python
|
|
2
|
|
3 """
|
|
4 count.py -- Take SAM files and output a table of counts with column
|
|
5 names that are the filenames, and rowname that are the reference
|
|
6 names.
|
|
7
|
|
8 Author: Vince Buffalo
|
|
9 Email: vsbuffaloAAAAA@gmail.com (with poly-A tail removed)
|
|
10 """
|
|
11
|
|
12 VERSION = 0.91
|
|
13
|
|
14 import sys
|
|
15 import csv
|
|
16 from os import path
|
|
17 try:
|
|
18 import pysam
|
|
19 except ImportError:
|
|
20 sys.exit("pysam not installed; please install it\n")
|
|
21
|
|
22 from optparse import OptionParser
|
|
23
|
|
24 def SAM_file_to_counts(filename, bam=False, extra=False, use_all_references=True):
|
|
25 """
|
|
26 Take SAM filename, and create a hash of mapped and unmapped reads;
|
|
27 keys are reference sequences, values are the counts of occurences.
|
|
28
|
|
29 Also, a hash of qualities (either 0 or >0) of mapped reads
|
|
30 is output, which is handy for diagnostics.
|
|
31 """
|
|
32 counts = dict()
|
|
33 unique = dict()
|
|
34 nonunique = dict()
|
|
35 mode = 'r'
|
|
36 if bam:
|
|
37 mode = 'rb'
|
|
38 sf = pysam.Samfile(filename, mode)
|
|
39
|
|
40 if use_all_references:
|
|
41 # Make dictionary of all entries in header
|
|
42 for sn in sf.header['SQ']:
|
|
43 if extra:
|
|
44 unique[sn['SN']] = 0
|
|
45 nonunique[sn['SN']] = 0
|
|
46 counts[sn['SN']] = 0
|
|
47
|
|
48 for read in sf:
|
|
49 if not read.is_unmapped:
|
|
50 id_name = sf.getrname(read.rname) if read.rname != -1 else 0
|
|
51
|
|
52 if not use_all_references and not counts.get(id_name, False):
|
|
53 ## Only make keys based on aligning reads, make empty hash
|
|
54 if extra:
|
|
55 unique[id_name] = 0
|
|
56 nonunique[id_name] = 0
|
|
57 ## initiate entry; even if not mapped, record 0 count
|
|
58 counts[id_name] = counts.get(id_name, 0)
|
|
59
|
|
60
|
|
61 counts[id_name] = counts.get(id_name, 0) + 1
|
|
62
|
|
63 if extra:
|
|
64 if read.mapq == 0:
|
|
65 nonunique[id_name] = nonunique[id_name] + 1
|
|
66 else:
|
|
67 unique[id_name] = unique[id_name] + 1
|
|
68
|
|
69 if extra:
|
|
70 return {'counts':counts, 'unique':unique, 'nonunique':nonunique}
|
|
71
|
|
72 return {'counts':counts}
|
|
73
|
|
74 def collapsed_nested_count_dict(counts_dict, all_ids, order=None):
|
|
75 """
|
|
76 Takes a nested dictionary `counts_dict` and `all_ids`, which is
|
|
77 built with the `table_dict`. All files (first keys) in
|
|
78 `counts_dict` are made into columns with order specified by
|
|
79 `order`.
|
|
80
|
|
81 Output is a dictionary with keys that are the id's (genes or
|
|
82 transcripts), with values that are ordered counts. A header will
|
|
83 be created on the first row from the ordered columns (extracted
|
|
84 from filenames).
|
|
85 """
|
|
86 if order is None:
|
|
87 col_order = counts_dict.keys()
|
|
88 else:
|
|
89 col_order = order
|
|
90
|
|
91 collapsed_dict = dict()
|
|
92 for i, filename in enumerate(col_order):
|
|
93 for id_name in all_ids:
|
|
94 if not collapsed_dict.get(id_name, False):
|
|
95 collapsed_dict[id_name] = list()
|
|
96
|
|
97 # get counts and append
|
|
98 c = counts_dict[filename].get(id_name, 0)
|
|
99 collapsed_dict[id_name].append(c)
|
|
100 return {'table':collapsed_dict, 'header':col_order}
|
|
101
|
|
102
|
|
103 def counts_to_file(table_dict, outfilename, delimiter=',', labels=''):
|
|
104 """
|
|
105 A function for its side-effect of writing `table_dict` (which
|
|
106 contains a table and header), to `outfilename` with the specified
|
|
107 `delimiter`.
|
|
108 """
|
|
109 writer = csv.writer(open(outfilename, 'w'), delimiter=delimiter)
|
|
110 table = table_dict['table']
|
|
111 if labels:
|
|
112 header = labels.split(',')
|
|
113 else:
|
|
114 header = table_dict['header']
|
|
115
|
|
116 header_row = True
|
|
117 for id_name, fields in table.items():
|
|
118 if header_row:
|
|
119 row = ['id'] + header
|
|
120 writer.writerow(row)
|
|
121 header_row = False
|
|
122
|
|
123 if id_name == 0:
|
|
124 continue
|
|
125 row = [id_name]
|
|
126 row.extend(fields)
|
|
127 writer.writerow(row)
|
|
128
|
|
129 if __name__ == '__main__':
|
|
130 parser = OptionParser()
|
|
131 parser.add_option("-d", "--delimiter", dest="delimiter",
|
|
132 help="the delimiter (default: tab)", default='\t')
|
|
133 parser.add_option("-o", "--out-file", dest="out_file",
|
|
134 help="output filename (default: counts.txt)",
|
|
135 default='counts.txt', action="store", type="string")
|
|
136 parser.add_option("-u", "--extra-output", dest="extra_out",
|
|
137 help="output extra information on non-unique and unique mappers (default: False)",
|
|
138 default=False, action="store_true")
|
|
139 parser.add_option("-b", "--bam", dest="bam",
|
|
140 help="all input files are BAM (default: False)",
|
|
141 default=False, action="store_true")
|
|
142 parser.add_option("-r", "--use-all-references", dest="use_all_references",
|
|
143 help="Use all the references from the SAM header (default: True)",
|
|
144 default=True, action="store_false")
|
|
145 parser.add_option("-f", "--extra-out-files", dest="extra_out_files",
|
|
146 help="comma-delimited filenames of unique and non-unique output "
|
|
147 "(default: unique.txt,nonunique.txt)",
|
|
148 default='unique.txt,nonunique.txt', action="store", type="string")
|
|
149 parser.add_option("-v", "--verbose", dest="verbose",
|
|
150 help="enable verbose output")
|
|
151 parser.add_option("-l", "--columns-labels", dest="col_labels", help="comma-delimited label names for samples", action="store", type="string")
|
|
152
|
|
153 (options, args) = parser.parse_args()
|
|
154
|
|
155 if len(args) < 1:
|
|
156 parser.error("one or more SAM files as arguments required")
|
|
157
|
|
158 file_counts = dict()
|
|
159 file_unique_counts = dict()
|
|
160 file_nonunique_counts = dict()
|
|
161 all_ids = list()
|
|
162 files = [path.basename(f) for f in args]
|
|
163
|
|
164 if options.col_labels and len(files) != len(options.col_labels.split(',')):
|
|
165 parser.error("Number of sample names does not equal number of files")
|
|
166
|
|
167 if len(set(files)) != len(set(args)):
|
|
168 parser.error("file args must have unique base names (i.e. no foo/bar joo/bar)")
|
|
169
|
|
170 ## do a pre-run check that all files exist
|
|
171 for full_filename in args:
|
|
172 if not path.exists(full_filename):
|
|
173 parser.error("file '%s' does not exist" % full_filename)
|
|
174
|
|
175 for full_filename in args:
|
|
176 filename = path.basename(full_filename)
|
|
177 ## read in SAM file, extract counts, and unpack counts
|
|
178 tmp = SAM_file_to_counts(full_filename, bam=options.bam, extra=options.extra_out,
|
|
179 use_all_references=options.use_all_references)
|
|
180
|
|
181 if options.extra_out:
|
|
182 counts, unique, nonunique = tmp['counts'], tmp['unique'], tmp['nonunique']
|
|
183 else:
|
|
184 counts = tmp['counts']
|
|
185
|
|
186 ## save counts, and unique/non-unique counts
|
|
187 file_counts[filename] = counts
|
|
188
|
|
189 if options.extra_out:
|
|
190 file_unique_counts[filename] = unique
|
|
191 file_nonunique_counts[filename] = nonunique
|
|
192
|
|
193 ## add all ids encountered in this in this file
|
|
194 all_ids.extend(file_counts[filename].keys())
|
|
195
|
|
196 ## Uniquify all_ids, and then take the nested file_counts
|
|
197 ## dictionary, collapse, and write to file.
|
|
198 all_ids = set(all_ids)
|
|
199 table_dict = collapsed_nested_count_dict(file_counts, all_ids, order=files)
|
|
200 counts_to_file(table_dict, options.out_file, delimiter=options.delimiter, labels=options.col_labels)
|
|
201
|
|
202 if options.extra_out:
|
|
203 unique_fn, nonunique_fn = options.extra_out_files.split(',')
|
|
204 unique_table_dict = collapsed_nested_count_dict(file_unique_counts, all_ids, order=files)
|
|
205 nonunique_table_dict = collapsed_nested_count_dict(file_nonunique_counts, all_ids, order=files)
|
|
206
|
|
207 counts_to_file(unique_table_dict, unique_fn, delimiter=options.delimiter)
|
|
208 counts_to_file(nonunique_table_dict, nonunique_fn, delimiter=options.delimiter)
|
|
209
|