1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
|
import argparse
import csv
import os
from collections import namedtuple
from keras.models import load_model
from dl.SegmentGenotypingClassesFunctions import get_properties_json, BatchNumpyFileSequence
import numpy as np
vcf_header = ("##fileformat=VCFv4.1\n"
"##GobyPython={}\n"
"##modelPath={}\n"
"##modelPrefix={}\n"
"##datasetPath={}\n"
"##datasetPrefix={}\n"
"##indelsTrimmed={}\n"
"##FORMAT=<ID=GT,Number=1,Type=String,Description=\"Genotype\">\n"
"##FORMAT=<ID=MC,Number=1,Type=String,Description=\"Model Calls.\">\n"
"##FORMAT=<ID=P,Number=1,Type=Float,Description=\"Model proability.\">\n")
VcfOutputWriter = namedtuple("VcfOutputWriter", ["vcf_writer", "bed_writer"])
VcfOutputLine = namedtuple("VcfLine", ["vcf_ref", "vcf_alts", "vcf_gt", "vcf_mc", "vcf_model_probability",
"vcf_max_len"])
class VcfLine:
# TODO: Avoid duplication b/w __init__ and clear- call __init__ from clear, call clear from __init__, other way
def __init__(self):
self.is_indel = False
self.last_base_location = 0
self.last_gap_location = 0
self.vcf_location = None
self.vcf_ref_bases = []
self.vcf_predictions = []
self.vcf_probabilities = []
self.vcf_chromosome = None
def clear(self):
self.is_indel = False
self.last_base_location = 0
self.last_gap_location = 0
self.vcf_location = None
self.vcf_ref_bases = []
self.vcf_predictions = []
self.vcf_probabilities = []
self.vcf_chromosome = None
def add_base(self, segment_ref_base, segment_prediction_base, segment_probability_base, segment_location_base,
segment_chromosome):
if self.vcf_location is None:
self.vcf_location = segment_location_base
if self.vcf_chromosome is None:
self.vcf_chromosome = segment_chromosome
else:
if self.vcf_chromosome != segment_chromosome:
raise ValueError("VCF lines should have same chromosome")
if "-" in segment_prediction_base:
self.is_indel = True
self.last_gap_location = segment_location_base
self.last_base_location = segment_location_base
self.vcf_ref_bases.append(segment_ref_base)
self.vcf_predictions.append(segment_prediction_base)
self.vcf_probabilities.append(segment_probability_base)
def is_empty(self):
return self.vcf_location is None
def need_to_flush(self, segment_next_location):
if self.is_empty():
return False
if not self.is_indel:
return self.last_base_location != segment_next_location
else:
if self.last_base_location == self.last_gap_location:
return False
else:
return self.last_base_location != segment_next_location
def _get_basename(path):
return os.path.splitext(os.path.basename(path))[0]
def _write_vcf_files(model, properties_json, test_data, **vcf_output_writers):
vcf_line = VcfLine()
for data_idx in range(len(test_data)):
if data_idx % 20 == 0:
print("Evaluating batch {} of {}...".format(data_idx, len(test_data)))
(batch_input_dict, batch_label_dict), batch_ref, batch_location, batch_chromosome = test_data[data_idx]
batch_input = batch_input_dict["model_input"]
batch_label = batch_label_dict["main_output"]
batch_predictions = model.predict_on_batch(batch_input)
for segment_in_batch_idx in range(batch_predictions.shape[0]):
segment_chromosome = batch_chromosome[segment_in_batch_idx][0]
segment_label_categorical = batch_label[segment_in_batch_idx]
segment_prediction_categorical = batch_predictions[segment_in_batch_idx]
segment_ref_with_padding = batch_ref[segment_in_batch_idx]
segment_label_with_padding = np.argmax(segment_label_categorical, axis=1)
segment_prediction_with_padding = np.argmax(segment_prediction_categorical, axis=1)
segment_model_probabilities_with_padding = np.max(segment_prediction_categorical, axis=1)
# Only use positions where label != 0, as label 0 reserved for padding
segment_label_non_padding_positions = segment_label_with_padding != 0
segment_prediction = np.extract(segment_label_non_padding_positions, segment_prediction_with_padding)
segment_model_probabilities = np.extract(segment_label_non_padding_positions,
segment_model_probabilities_with_padding)
segment_ref = np.extract(segment_label_non_padding_positions, segment_ref_with_padding)
segment_true_genotype_prediction = [
properties_json["genotype.segment.label_plus_one.{}".format(label)]
for label in segment_prediction
]
segment_locations = np.extract(segment_label_non_padding_positions,
batch_location[segment_in_batch_idx])
for base_idx in range(len(segment_ref)):
base_location = segment_locations[base_idx]
base_ref = segment_ref[base_idx]
base_prediction = segment_true_genotype_prediction[base_idx]
base_probability = segment_model_probabilities[base_idx]
if vcf_line.need_to_flush(base_location):
_write_vcf_line(vcf_line=vcf_line,
dataset_field=properties_json["batch_prefix"],
**vcf_output_writers)
vcf_line.clear()
vcf_line.add_base(segment_ref_base=base_ref,
segment_prediction_base=base_prediction,
segment_probability_base=base_probability,
segment_location_base=base_location,
segment_chromosome=segment_chromosome)
if not vcf_line.is_empty():
_write_vcf_line(vcf_line=vcf_line,
dataset_field=properties_json["batch_prefix"],
**vcf_output_writers)
vcf_line.clear()
def _generate_vcf_output_line(vcf_ref, vcf_predicted_alleles, vcf_line, trim_indels):
if trim_indels:
vcf_ref, vcf_predicted_alleles = _trim_indels(vcf_ref, vcf_predicted_alleles)
vcf_alts = list(set(filter(lambda x: x != vcf_ref, vcf_predicted_alleles)))
vcf_max_len = max(map(len, vcf_alts)) if vcf_alts else 0
vcf_max_len = max(vcf_max_len, len(vcf_ref))
vcf_possible_alleles = [vcf_ref] + vcf_alts
vcf_unique_predicted_alleles = []
for allele in vcf_predicted_alleles:
if allele not in vcf_unique_predicted_alleles:
vcf_unique_predicted_alleles.append(allele)
vcf_gt = [vcf_possible_alleles.index(allele) for allele in vcf_unique_predicted_alleles]
vcf_mc = [vcf_possible_alleles[allele_idx] for allele_idx in vcf_gt]
vcf_model_probability = np.mean(vcf_line.vcf_probabilities)
return VcfOutputLine(vcf_ref=vcf_ref, vcf_alts=vcf_alts, vcf_gt=vcf_gt, vcf_mc=vcf_mc,
vcf_model_probability=vcf_model_probability, vcf_max_len=vcf_max_len)
def _invalid_entry(formatted_ref, formatted_alts, gt):
invalid_ref = formatted_ref == "." and 0 in gt
invalid_alts = formatted_alts == "." and (1 in gt or 2 in gt)
return invalid_ref or invalid_alts
def _generate_vcf_entries(vcf_output_line, vcf_line, dataset_field):
formatted_ref = _format_alleles(vcf_output_line.vcf_ref)
formatted_alts = _format_alleles(*vcf_output_line.vcf_alts)
invalid_entry = _invalid_entry(formatted_ref, formatted_alts, vcf_output_line.vcf_gt)
vcf_entry = {
"CHROM": vcf_line.vcf_chromosome,
"POS": vcf_line.vcf_location + 1,
"ID": ".",
"REF": vcf_output_line.vcf_ref,
"ALT": formatted_alts,
"QUAL": ".",
"FILTER": ".",
"INFO": ".",
"FORMAT": "GT:MC:P",
dataset_field: "{}:{}:{}".format("/".join(map(str, vcf_output_line.vcf_gt)),
"/".join(vcf_output_line.vcf_mc),
vcf_output_line.vcf_model_probability),
}
bed_entry = {
"chrom": vcf_line.vcf_chromosome,
"start": vcf_line.vcf_location,
"end": vcf_line.vcf_location + vcf_output_line.vcf_max_len,
}
return vcf_entry, bed_entry, invalid_entry
def _write_vcf_line(vcf_line, dataset_field, regular_vcf_output_writer, original_vcf_output_writer=None,
error_vcf_output_writer=None):
vcf_ref = "".join(vcf_line.vcf_ref_bases)
vcf_predicted_alleles = ["".join(bases) for bases in list(zip(*map(list, vcf_line.vcf_predictions)))]
regular_vcf_output_line = _generate_vcf_output_line(vcf_ref, vcf_predicted_alleles, vcf_line, trim_indels=True)
regular_entry = _generate_vcf_entries(regular_vcf_output_line, vcf_line, dataset_field)
regular_vcf_entry, regular_bed_entry, invalid_entry = regular_entry
if not invalid_entry:
regular_vcf_output_writer.vcf_writer.writerow(regular_vcf_entry)
regular_vcf_output_writer.bed_writer.writerow(regular_bed_entry)
if original_vcf_output_writer is not None or error_vcf_output_writer is not None:
original_vcf_output_line = _generate_vcf_output_line(vcf_ref, vcf_predicted_alleles, vcf_line,
trim_indels=False)
original_entry = _generate_vcf_entries(original_vcf_output_line, vcf_line, dataset_field)
original_vcf_entry, original_bed_entry, _ = original_entry
if original_vcf_output_writer is not None:
if not invalid_entry:
original_vcf_output_writer.vcf_writer.writerow(original_vcf_entry)
original_vcf_output_writer.bed_writer.writerow(original_bed_entry)
else:
if error_vcf_output_writer is not None:
error_vcf_output_writer.vcf_writer.writerow(original_vcf_entry)
error_vcf_output_writer.bed_writer.writerow(original_bed_entry)
def _trim_indels(ref, predicted_alleles):
"""
Properly format indels for inclusion in VCF files
1. trim all alleles to index of last dash any allele,
IE: from: GTAC to: G--C,G-AC -> from: GTA to: G--,G-A
2. delete dashes
IE: from: GTA to: G--,G-A -> from: GTA to: G,GA
Based on FormatIndelVCF from VariationAnalysis project
:param ref: reference allele
:param predicted_alleles: list of predicted alleles
:return: ref, alts
"""
last_suffix_index = len(ref)
for suffix_index in range(len(ref) - 1, 0, -1):
ref_pred = ref[suffix_index]
found_last_suffix_index = False
for predicted_allele in predicted_alleles:
if predicted_allele[suffix_index] != ref_pred:
found_last_suffix_index = True
break
if found_last_suffix_index:
last_suffix_index = suffix_index + 1
break
ref = ref[:last_suffix_index].replace("-", "")
predicted_alleles = [predicted_allele[:last_suffix_index].replace("-", "")
for predicted_allele in predicted_alleles]
return ref, predicted_alleles
def _format_alleles(*alleles):
# Only select non-empty alts
valid_alleles = list(filter(lambda allele: allele, alleles))
if len(valid_alleles) > 0:
return ",".join(valid_alleles)
else:
return "."
def main(args):
properties_json_to_use = get_properties_json(args.testing)
model_to_use = load_model(args.model)
max_base_count = model_to_use.input_shape[1]
test_data_to_use = BatchNumpyFileSequence(args.testing, max_base_count, properties_json_to_use, array_type='vcf')
vcf_fields = ["CHROM", "POS", "ID", "REF", "ALT", "QUAL", "FILTER", "INFO", "FORMAT",
properties_json_to_use["batch_prefix"]]
bed_fields = ["chrom", "start", "end"]
prefix_dir = os.path.dirname(args.prefix)
if prefix_dir:
os.makedirs(prefix_dir, exist_ok=True)
prefix = os.path.splitext(os.path.basename(args.prefix))[0]
prefix = os.path.join(prefix_dir, prefix)
regular_vcf_file = open("{}.vcf".format(prefix), "w")
regular_bed_file = open("{}.bed".format(prefix), "w")
regular_vcf_file.write(vcf_header.format(args.version, args.model, _get_basename(args.model), args.testing,
properties_json_to_use["batch_prefix"], True))
regular_vcf_file.write("#{}\n".format("\t".join(vcf_fields)))
regular_vcf_writer = csv.DictWriter(regular_vcf_file, fieldnames=vcf_fields, delimiter="\t", lineterminator="\n")
regular_bed_writer = csv.DictWriter(regular_bed_file, fieldnames=bed_fields, delimiter="\t", lineterminator="\n")
regular_vcf_output_writer = VcfOutputWriter(vcf_writer=regular_vcf_writer, bed_writer=regular_bed_writer)
original_vcf_file = None
original_bed_file = None
original_vcf_output_writer = None
if args.generate_original_vcf:
original_vcf_file = open("{}_original.vcf".format(prefix), "w")
original_bed_file = open("{}_original.bed".format(prefix), "w")
original_vcf_file.write(vcf_header.format(args.version, args.model, _get_basename(args.model), args.testing,
properties_json_to_use["batch_prefix"], False))
original_vcf_file.write("#{}\n".format("\t".join(vcf_fields)))
original_vcf_writer = csv.DictWriter(original_vcf_file, fieldnames=vcf_fields, delimiter="\t",
lineterminator="\n")
original_bed_writer = csv.DictWriter(original_bed_file, fieldnames=bed_fields, delimiter="\t",
lineterminator="\n")
original_vcf_output_writer = VcfOutputWriter(vcf_writer=original_vcf_writer, bed_writer=original_bed_writer)
error_vcf_file = None
error_bed_file = None
error_vcf_output_writer = None
if args.generate_error_vcf:
error_vcf_file = open("{}_error.vcf".format(prefix), "w")
error_bed_file = open("{}_error.bed".format(prefix), "w")
error_vcf_file.write(vcf_header.format(args.version, args.model, _get_basename(args.model), args.testing,
properties_json_to_use["batch_prefix"], False))
error_vcf_file.write("#{}\n".format("\t".join(vcf_fields)))
error_vcf_writer = csv.DictWriter(error_vcf_file, fieldnames=vcf_fields, delimiter="\t",
lineterminator="\n")
error_bed_writer = csv.DictWriter(error_bed_file, fieldnames=bed_fields, delimiter="\t",
lineterminator="\n")
error_vcf_output_writer = VcfOutputWriter(vcf_writer=error_vcf_writer, bed_writer=error_bed_writer)
_write_vcf_files(model_to_use, properties_json_to_use, test_data_to_use,
regular_vcf_output_writer=regular_vcf_output_writer,
original_vcf_output_writer=original_vcf_output_writer,
error_vcf_output_writer=error_vcf_output_writer)
regular_vcf_file.close()
regular_bed_file.close()
if args.generate_original_vcf:
original_vcf_file.close()
original_bed_file.close()
if args.generate_error_vcf:
error_vcf_file.close()
error_bed_file.close()
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("-m", "--model", type=str, required=True, help="Path to model to evaluate.")
parser.add_argument("-t", "--testing", type=str, required=True,
help="Path to test set directory that was preprocessed via GenerateDatasetsFromSSI.")
parser.add_argument("-p", "--prefix", type=str, required=True,
help="Prefix for generated VCF and BED files.")
parser.add_argument("--version", type=str, help="Version of goby being used", default="1.4.1-SNAPSHOT")
parser.add_argument("--generate-original-vcf", action="store_true", dest="generate_original_vcf",
help="If present, generate separate file at <prefix>_original.{vcf|bed} representing the "
"original calls made by the model, before any reformatting to handle indels")
parser.add_argument("--generate-error-vcf", action="store_true", dest="generate_error_vcf",
help="If present, generate seprate file at <prefix>_error.{vcf|bed} with any calls that are "
"malformed for the VCF specification.")
parser_args = parser.parse_args()
main(parser_args)
|
