0
|
1 #!/usr/bin/python
|
|
2
|
|
3 import os.path
|
|
4 import sys
|
|
5 import optparse
|
|
6
|
|
7 import bedClass
|
|
8 from bedClass import *
|
|
9
|
|
10 import vcfClass
|
|
11 from vcfClass import *
|
|
12
|
|
13 import tools
|
|
14 from tools import *
|
|
15
|
|
16 if __name__ == "__main__":
|
|
17 main()
|
|
18
|
|
19 # Intersect two vcf files. It is assumed that the two files are
|
|
20 # sorted by genomic coordinates and the reference sequences are
|
|
21 # in the same order.
|
|
22 def intersectVcf(v1, v2, priority, outputFile):
|
|
23 success1 = v1.getRecord()
|
|
24 success2 = v2.getRecord()
|
|
25 currentReferenceSequence = v1.referenceSequence
|
|
26
|
|
27 # As soon as the end of either file is reached, there can be no
|
|
28 # more intersecting SNPs, so terminate.
|
|
29 while success1 and success2:
|
|
30 if v1.referenceSequence == v2.referenceSequence and v1.referenceSequence == currentReferenceSequence:
|
|
31 if v1.position == v2.position:
|
|
32 writeVcfRecord(priority, v1, v2, outputFile)
|
|
33 success1 = v1.getRecord()
|
|
34 success2 = v2.getRecord()
|
|
35 elif v2.position > v1.position: success1 = v1.parseVcf(v2.referenceSequence, v2.position, False, None)
|
|
36 elif v1.position > v2.position: success2 = v2.parseVcf(v1.referenceSequence, v1.position, False, None)
|
|
37 else:
|
|
38 if v1.referenceSequence == currentReferenceSequence: success1 = v1.parseVcf(v2.referenceSequence, v2.position, False, None)
|
|
39 elif v2.referenceSequence == currentReferenceSequence: success2 = v2.parseVcf(v1.referenceSequence, v1.position, False, None)
|
|
40
|
|
41 # If the last record for a reference sequence is the same for both vcf
|
|
42 # files, they will both have referenceSequences different from the
|
|
43 # current reference sequence. Change the reference sequence to reflect
|
|
44 # this and proceed.
|
|
45 else:
|
|
46 if v1.referenceSequence != v2.referenceSequence:
|
|
47 print >> sys.stderr, "ERROR: Reference sequences for both files are unexpectedly different."
|
|
48 print >> sys.stderr, "Check that both files contain records for the following reference sequences:"
|
|
49 print >> sys.stderr, "\t", v1.referenceSequence, " and ", v2.referenceSequence
|
|
50 exit(1)
|
|
51 currentReferenceSequence = v1.referenceSequence
|
|
52
|
|
53 # Intersect a vcf file and a bed file. It is assumed that the
|
|
54 # two files are sorted by genomic coordinates and the reference
|
|
55 # sequences are in the same order.
|
|
56 def intersectVcfBed(v, b, outputFile):
|
|
57 successb = b.getRecord()
|
|
58 successv = v.getRecord()
|
|
59 currentReferenceSequence = v.referenceSequence
|
|
60
|
|
61 # As soon as the end of the first file is reached, there are no
|
|
62 # more intersections and the program can terminate.
|
|
63 while successv:
|
|
64 if v.referenceSequence == b.referenceSequence:
|
|
65 if v.position < b.start: successv = v.parseVcf(b.referenceSequence, b.start, False, None)
|
|
66 elif v.position > b.end: successb = b.parseBed(v.referenceSequence, v.position)
|
|
67 else:
|
|
68 outputFile.write(v.record)
|
|
69 successv = v.getRecord()
|
|
70 else:
|
|
71 if v.referenceSequence == currentReferenceSequence: successv = v.parseVcf(b.referenceSequence, b.start, False, None)
|
|
72 if b.referenceSequence == currentReferenceSequence: successb = b.parseBed(v.referenceSequence, v.position)
|
|
73 currentReferenceSequence = v.referenceSequence
|
|
74
|
|
75 def main():
|
|
76
|
|
77 # Parse the command line options
|
|
78 usage = "Usage: vcfPytools.py intersect [options]"
|
|
79 parser = optparse.OptionParser(usage = usage)
|
|
80 parser.add_option("-i", "--in",
|
|
81 action="append", type="string",
|
|
82 dest="vcfFiles", help="input vcf files")
|
|
83 parser.add_option("-b", "--bed",
|
|
84 action="store", type="string",
|
|
85 dest="bedFile", help="input bed vcf file")
|
|
86 parser.add_option("-o", "--out",
|
|
87 action="store", type="string",
|
|
88 dest="output", help="output vcf file")
|
|
89 parser.add_option("-f", "--priority-file",
|
|
90 action="store", type="string",
|
|
91 dest="priorityFile", help="output records from this vcf file")
|
|
92
|
|
93 (options, args) = parser.parse_args()
|
|
94
|
|
95 # Check that a single vcf file is given.
|
|
96 if options.vcfFiles == None:
|
|
97 parser.print_help()
|
|
98 print >> sys.stderr, "\nAt least one vcf file (--in, -i) is required for performing intersection."
|
|
99 exit(1)
|
|
100 elif len(options.vcfFiles) > 2:
|
|
101 parser.print_help()
|
|
102 print >> sys.stderr, "\nAt most, two vcf files (--in, -i) can be submitted for performing intersection."
|
|
103 exit(1)
|
|
104 elif len(options.vcfFiles) == 1 and not options.bedFile:
|
|
105 parser.print_help()
|
|
106 print >> sys.stderr, "\nIf only one vcf file (--in, -i) is specified, a bed file is also required for performing intersection."
|
|
107 exit(1)
|
|
108
|
|
109 # Set the output file to stdout if no output file was specified.
|
|
110 outputFile, writeOut = setOutput(options.output) # tools.py
|
|
111
|
|
112 # If intersecting with a bed file, call the bed intersection routine.
|
|
113 if options.bedFile:
|
|
114 v = vcf() # Define vcf object.
|
|
115 b = bed() # Define bed object.
|
|
116
|
|
117 # Open the files.
|
|
118 v.openVcf(options.vcfFiles[0])
|
|
119 b.openBed(options.bedFile)
|
|
120
|
|
121 # Read in the header information.
|
|
122 v.parseHeader(options.vcfFiles[0], writeOut)
|
|
123 taskDescriptor = "##vcfPytools=intersect " + options.vcfFiles[0] + ", " + options.bedFile
|
|
124 writeHeader(outputFile, v, False, taskDescriptor) # tools.py
|
|
125
|
|
126 # Intersect the vcf file with the bed file.
|
|
127 intersectVcfBed(v, b, outputFile)
|
|
128
|
|
129 # Check that the input files had the same list of reference sequences.
|
|
130 # If not, it is possible that there were some problems.
|
|
131 checkReferenceSequenceLists(v.referenceSequenceList, b.referenceSequenceList) # tools.py
|
|
132
|
|
133 # Close the files.
|
|
134 v.closeVcf(options.vcfFiles[0])
|
|
135 b.closeBed(options.bedFile)
|
|
136
|
|
137 else:
|
|
138 priority = setVcfPriority(options.priorityFile, options.vcfFiles)
|
|
139 v1 = vcf() # Define vcf object.
|
|
140 v2 = vcf() # Define vcf object.
|
|
141
|
|
142 # Open the vcf files.
|
|
143 v1.openVcf(options.vcfFiles[0])
|
|
144 v2.openVcf(options.vcfFiles[1])
|
|
145
|
|
146 # Read in the header information.
|
|
147 v1.parseHeader(options.vcfFiles[0], writeOut)
|
|
148 v2.parseHeader(options.vcfFiles[1], writeOut)
|
|
149 if priority == 3:
|
|
150 v3 = vcf() # Generate a new vcf object that will contain the header information of the new file.
|
|
151 mergeHeaders(v1, v2, v3) # tools.py
|
|
152 v1.processInfo = True
|
|
153 v2.processInfo = True
|
|
154 else: checkDataSets(v1, v2)
|
|
155
|
|
156 #print v1.samplesList
|
|
157 #print v2.samplesList
|
|
158
|
|
159 # Check that the header for the two files contain the same samples.
|
|
160 if v1.samplesList != v2.samplesList:
|
|
161 print >> sys.stderr, "vcf files contain different samples (or sample order)."
|
|
162 exit(1)
|
|
163 else:
|
|
164 taskDescriptor = "##vcfPytools=intersect " + v1.filename + ", " + v2.filename
|
|
165 if priority == 3: writeHeader(outputFile, v3, False, taskDescriptor)
|
|
166 elif (priority == 2 and v2.hasHeader) or not v1.hasHeader: writeHeader(outputFile, v2, False, taskDescriptor) # tools.py
|
|
167 else: writeHeader(outputFile, v1, False, taskDescriptor) # tools.py
|
|
168
|
|
169 # Intersect the two vcf files.
|
|
170 intersectVcf(v1, v2, priority, outputFile)
|
|
171
|
|
172 # Check that the input files had the same list of reference sequences.
|
|
173 # If not, it is possible that there were some problems.
|
|
174 checkReferenceSequenceLists(v1.referenceSequenceList, v2.referenceSequenceList) # tools.py
|
|
175
|
|
176 # Close the vcf files.
|
|
177 v1.closeVcf(options.vcfFiles[0])
|
|
178 v2.closeVcf(options.vcfFiles[1])
|
|
179
|
|
180 # End the program.
|
|
181 return 0
|