-
Notifications
You must be signed in to change notification settings - Fork 3
/
regression_fit_track.py
150 lines (121 loc) · 6.38 KB
/
regression_fit_track.py
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
"""
This script is illustrates the regression fit error.
This script calculates the regression fit error for each transcript, for log(data)-log(norm_data) or for log(data)
if no norm data is provided.
data = mean of all input bigwig files which doesn't contain norm in their names
norm_data = mean of all input bigwig files which contain norm in their names
"""
import numpy as np
import pandas as pd
from bx.bbi.bigwig_file import BigWigFile
from utils import read_known_genes
WINDOW_SIZE = 50
LOG_SCALE = True
def generate_bigwig_all(all_bw, exp, chrom_sizes):
from sklearn.linear_model import LinearRegression
def win_arr(arr):
try:
arr = np.nan_to_num(arr)
except:
print(arr)
raise
return np.mean(
arr[:(arr.shape[0] // WINDOW_SIZE) * WINDOW_SIZE].reshape(arr.shape[0] // WINDOW_SIZE, WINDOW_SIZE), -1)
transcripts = read_known_genes('hg19')
# skip transcript in special chromosomes
transcripts = transcripts.loc[~transcripts.chrom.isin([x for x in transcripts.chrom.unique() if '_' in x])]
transcripts = transcripts.loc[transcripts.cdsStart < transcripts.cdsEnd]
chrom_sizes = chrom_sizes.loc[chrom_sizes.chrom.isin(transcripts.chrom.unique())]
# create sum bigwig
exp_dict = dict()
norm_dict = dict()
for chrom, chrom_size in chrom_sizes.itertuples(False):
print(chrom)
# to avoid memory overflow
chrom_exp_data = np.concatenate(
[win_arr(np.nansum([v.get_as_array(chrom, i - 100000, i) for v in all_bw[exp]], 0)) for i in
range(100000, chrom_size, 100000)] + [
win_arr(np.nansum(
[v.get_as_array(chrom, chrom_size - chrom_size % 100000, chrom_size) for v in all_bw[exp]], 0))])
chrom_exp_data = np.log2(chrom_exp_data + 1)
chrom_norm_data = np.concatenate(
[win_arr(np.nansum([v.get_as_array(chrom, i - 100000, i) for v in all_bw['norm']], 0)) for i in
range(100000, chrom_size, 100000)] + [
win_arr(np.nansum(
[v.get_as_array(chrom, chrom_size - chrom_size % 100000, chrom_size) for v in all_bw['norm']], 0))])
chrom_norm_data = np.log2(chrom_norm_data + 1)
norm_dict[chrom] = chrom_norm_data
exp_dict[chrom] = chrom_exp_data
# compute z-score
# skip transcripts with no cds
all_transcripts_padded = transcripts[
['name', 'chrom', 'strand', 'txStart', 'txEnd', 'cdsStart', 'cdsEnd', 'exonStarts', 'exonEnds']]
all_transcripts_padded.cdsStart += all_transcripts_padded.txStart
all_transcripts_padded.cdsEnd += all_transcripts_padded.txStart
all_transcripts_padded.exonStarts += all_transcripts_padded.txStart
all_transcripts_padded.exonEnds += all_transcripts_padded.txStart
all_transcripts_padded[['exonStarts', 'exonEnds', 'txStart', 'txEnd', 'cdsStart', 'cdsEnd']] /= WINDOW_SIZE
all_transcripts_padded.txStart = all_transcripts_padded.txStart.astype(int)
all_transcripts_padded.txEnd = all_transcripts_padded.txEnd.astype(int)
print('%s zscore' % exp)
fit_err_score_dict = dict()
count_dict = dict()
for chrom in exp_dict.keys():
fit_err_score_dict[chrom] = np.zeros_like(exp_dict[chrom])
count_dict[chrom] = np.zeros_like(exp_dict[chrom])
all_transcripts_padded = all_transcripts_padded.loc[all_transcripts_padded.chrom.isin(exp_dict.keys())]
print('going over transcripts')
for chrom, cdsStart, cdsEnd, txStart, txEnd, strand, exonSt, exonEn in all_transcripts_padded[
['chrom', 'cdsStart', 'cdsEnd', 'txStart', 'txEnd', 'strand',
'exonStarts', 'exonEnds']].itertuples(False):
# ex_windows_pos = np.arange(txStart, txEnd)
exon_wins = np.concatenate([np.arange(st, en, dtype=int) for st, en in zip(exonSt, exonEn)])
cds_pos = [np.arange(st, en, dtype=int) for st, en in
zip(exonSt, exonEn) if en < cdsEnd and st > cdsStart]
if len(cds_pos) == 0:
continue
lin_model = LinearRegression().fit(np.array([norm_dict[chrom][exon_wins]]).T, exp_dict[chrom][exon_wins])
# hinge diff
tx_score = exp_dict[chrom][exon_wins] - np.maximum(lin_model.predict(np.array([norm_dict[chrom][exon_wins]]).T),
0)
fit_err_score_dict[chrom][exon_wins] += tx_score
count_dict[chrom][exon_wins] += 1
print('end transcripts')
# write to bedgraph
all_z_score = []
for chrom, chrom_vals in fit_err_score_dict.items():
chrom_vals /= count_dict[chrom]
chrom_vals = np.nan_to_num(chrom_vals)
chrom_vals = pd.DataFrame(np.array([chrom_vals[chrom_vals != 0], np.where(chrom_vals != 0)[0]]).T)
chrom_vals.columns = ['score', 'chromStart']
chrom_vals['chromStart'] *= WINDOW_SIZE
chrom_vals['chromEnd'] = chrom_vals.chromStart + WINDOW_SIZE
chrom_vals['chrom'] = chrom
all_z_score.append(chrom_vals)
all_z_score = pd.concat(all_z_score)
all_z_score.chromStart = all_z_score.chromStart.astype(int)
all_z_score.chromEnd = all_z_score.chromEnd.astype(int)
all_z_score[['chrom', 'chromStart', 'chromEnd', 'score']].to_csv(
'%s_fit.bedgraph' % (exp),
index=False, sep=' ', header=False)
print('create %s_fit.bedgraph' % (exp))
def main():
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('name', help='input type')
parser.add_argument('cromsize_path', help='Chromosome sizes file')
parser.add_argument('-norm_type', help='substring appears in the norm files')
parser.add_argument('file', help='bigwig input', nargs='+')
args = parser.parse_args()
print('Input files:\n' + '\n'.join([f for f in args.file if (args.norm_type is None) or not (args.norm_type in f)]))
all_bw_by_type = [BigWigFile(file=open(f, 'rb')) for f in args.file if
(args.norm_type is None) or not (args.norm_type in f)]
all_bw = {args.name: all_bw_by_type}
if args.norm_type is not None:
print('Norm files:\n' + '\n'.join([f for f in args.file if args.norm_type in f]))
norm_bw = [BigWigFile(file=open(f, 'rb')) for f in args.file if args.norm_type in f]
all_bw['norm'] = norm_bw
chrom_sizes = pd.read_csv(args.cromsize_path, sep='\t', names=['chrom', 'size'])
generate_bigwig_all(all_bw, args.name, chrom_sizes)
if __name__ == '__main__':
main()