forked from YPARK/gtex-fqtl
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtwas.py
232 lines (200 loc) · 9.18 KB
/
twas.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
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
r"""TWAS using fqtl models
The major differences between the implementation here and existing
implementations in S-PREDIXCAN (Barbeira et al. 2018) and FUSION (Gusev et
al. 2016, Mancuso et al. 2017) are:
1. We estimate the TWAS statistic based on the following argument for tissue t:
y \sim N(eta a, sigma^2 I), eta = X b
\hat{a} = (eta' eta)^{-1} eta' y
= (b' X' X b)^{-1} b' X' y
= (b' R b)^{-1} b' z, R = X' X / n, z = X' y / n
2. We hard-threshold tissue effect sizes based on PIP, in order to avoid
correlating small posterior mean effect sizes with positive GWAS signals.
2. We use adaptive permutations (Che et al. 2014) to estimate p-values, rather
than making a Normal assumption under the null.
"""
import numpy as np
import pandas as pd
import tabix
import pyplink
import scipy.sparse as ss
import scipy.special as sp
import scipy.stats as st
import sys
_sf = st.norm().sf
def align(twas_data, X, B, pip):
"""Align the TWAS data
Return:
z - GWAS z-scores (p,) [Important: this needs to be broadcastable]
B - fqtl effect size matrix (p, m)
pip - fqtl posterior inclusion probabilities (p, m)
"""
# Align the TWAS data, genotypes, and fqtl effect size matrix
B = B.loc[twas_data['3_y']]
# Hack: chr12 has a dupe in fqtl fitted models "12_48000000_T_C_b37: 2 times"
B = B.loc[~B.index.duplicated(keep='first')]
pip = pip.loc[twas_data['3_y']]
pip = pip.loc[~pip.index.duplicated(keep='first')]
z = twas_data['5_x']
z = z.loc[~z.index.duplicated(keep='first')]
X_ = X.loc[:,~X.columns.duplicated(keep='first')]
X_ = X_.loc[:,twas_data['3_y']]
X_ = X_.loc[:,~X_.columns.duplicated(keep='first')]
R = est_gwas_cov(X_)
return z, R, B, pip
def adaptive_permutation(z, R, B, pip, pip_thresh=0.95, max_success=10, batch_size=1000, max_perm=10_000_000):
"""Return TWAS approximate permutation p-values
Return:
pvals - p-values (m, 1)
"""
B = B[pip > pip_thresh].fillna(0)
n_qtls = (pip > pip_thresh).sum(axis=0)
num_success = 0
num_perm = 0
stats = []
pvals = []
for t in B:
if n_qtls[t] == 0:
stats.append(float(0))
pvals.append(float(1))
continue
stat = (z @ B[t]) / (B[t].T @ R @ B[t])
stats.append(stat)
# Generate permuted TWAS statistics in batches by using sparse matrices to
# represent a batch of permuted non-zero effects
for _ in range(max_perm // batch_size):
perm_stat = []
data = np.repeat(B[t][pip[t] > pip_thresh], batch_size)
row = np.random.choice(B.shape[0], replace=True, size=(n_qtls[t], batch_size)).ravel()
col = np.repeat(np.arange(batch_size), n_qtls[t])
bt = ss.coo_matrix((data, (row, col)), shape=(B.shape[0], batch_size)).tocsr()
num = (z.values.reshape(1, -1) @ bt)
den = (R.values @ bt.power(2)).sum(axis=0, keepdims=True)
perm_stat.append(num / den)
perm_stat = np.vstack(perm_stat)
num_success += (np.square(perm_stat) >= np.square(stat)).sum()
num_perm += batch_size
if num_success >= max_success:
pvals.append(num_success / num_perm)
break
else:
pvals.append((num_success + 1) / (max_perm + 1))
# Important: after filtering low PIP factors/loadings, B is not guaranteed to
# have all 44 tissues
stats = pd.Series(stats, index=B.columns)
pvals = pd.Series(pvals, index=B.columns)
return stats, pvals
def load_fqtl_model(prefix, pip_thresh=0.95):
"""Return fqtl metadata, effect sizes, and PIPs
Return:
snp_annot - DataFrame (chromosome, ID, dummy, position, ref, alt)
B - fqtl effect sizes (p, m)
pip - fqtl posterior inclusion probabilities (p, m)
"""
snp_annot = pd.read_csv(f'{prefix}/fqtl.snp.info.gz', sep='\t', header=None)
snp_lodds = pd.read_csv(f'{prefix}/fqtl.snp.lodds.txt.gz', sep='\t', header=None)
snp_lodds.index = snp_annot[3]
tis_annot = pd.read_csv(f'{prefix}/fqtl.tis.info.gz', sep='\t', header=None)
tis_lodds = pd.read_csv(f'{prefix}/fqtl.tis.lodds.txt.gz', sep='\t', header=None)
tis_lodds.index = tis_annot[1].apply(lambda x: x.rsplit('_', maxsplit=1)[0])
keep = (sp.expit(snp_lodds) > pip_thresh).any(axis=0)
L = pd.read_csv(f'{prefix}/fqtl.snp.theta.txt.gz', sep='\t', header=None)
L.index = snp_annot[3]
F = pd.read_csv(f'{prefix}/fqtl.tis.theta.txt.gz', sep='\t', header=None)
F.index = tis_annot[1].apply(lambda x: x[:-9])
B = L.loc[:,keep].dot(F.loc[:,keep].T)
pip = pd.DataFrame(sp.expit(snp_lodds.loc[:,keep]) @ sp.expit(tis_lodds.loc[:,keep]).T,
index=B.index, columns=B.columns)
assert np.isfinite(pip.values).all()
return snp_annot, B, pip
def load_sqtl_model(prefix, pip_thresh=0.95):
"""Return sqtl metadata, effect sizes, and PIPs
Return:
snp_annot - DataFrame (chromosome, ID, dummy, position, ref, alt)
B - fqtl effect sizes (p, m)
pip - fqtl posterior inclusion probabilities (p, m)
"""
tis_annot = pd.read_csv(f'{prefix}/sqtl.tis.info.gz', sep='\t', header=None, index_col=0)
snp_annot = pd.read_csv(f'{prefix}/sqtl.snp.info.gz', sep='\t', header=None)
snp_lodds = pd.read_csv(f'{prefix}/sqtl.effect.lodds.txt.gz', sep='\t', header=None)
pip = sp.expit(snp_lodds)
pip.index = snp_annot[3]
pip.columns = tis_annot[1].apply(lambda x: x.rsplit('_', maxsplit=1)[0])
B = pd.read_csv(f'{prefix}/sqtl.effect.theta.txt.gz', sep='\t', header=None)
B.index = snp_annot[3]
B.columns = tis_annot[1].apply(lambda x: x.rsplit('_', maxsplit=1)[0])
return snp_annot, B, pip
def load_geno(prefix):
"""Return DataFrame containing genotypes"""
with pyplink.PyPlink(prefix) as f:
fam = f.get_fam()
bim = f.get_bim()
x = np.ma.masked_equal(np.array([row for _, row in f.iter_geno()]).T, -1).astype(float)
x -= x.mean()
x /= x.std()
x = pd.DataFrame(x, index=fam['fid'], columns=bim['pos'])
return x
def est_gwas_cov(x, sv_thresh=1e-4):
u, d, vt = np.linalg.svd(x, full_matrices=False)
d[d < sv_thresh] = 0
return pd.DataFrame(vt.T.dot(np.diag(np.square(d))).dot(vt) / x.shape[0], index=x.columns, columns=x.columns)
if __name__ == '__main__':
f = tabix.open(sys.argv[1])
genes = pd.read_csv(sys.argv[2], sep='\t', header=None, index_col=0)
fqtl_stats = dict()
fqtl_pvals = dict()
sqtl_stats = dict()
sqtl_pvals = dict()
for k, row in genes.iterrows():
try:
snp_annot, B, pip = load_fqtl_model(f'/broad/compbio/ypp/gtex/analysis/fqtl-gtex/result/fqtl-std/{k}/50/')
except FileNotFoundError:
print(f'warning: skipping empty model ({k}: {row[1]})')
continue
try:
gwas_z = (pd.DataFrame(f.query(f'chr{row[2]}', int(row[3] - 1e6), int(row[3] + 1e6)))
.astype(dict(enumerate([str, int, int, str, str, float, float]))))
except tabix.TabixError:
print(f'warning: skipping empty cis-region in GWAS file {sys.argv[1]} ({k}: {row[1]})')
continue
try:
twas_data = gwas_z.merge(snp_annot, left_on=2, right_on=3).set_index(2)
strand_flip = dict(zip('ACGT', 'TGCA'))
temp = twas_data[['3_x', '4_x']].applymap(lambda x: strand_flip[x])
strand_flipped = np.logical_or(
np.logical_and(temp['3_x'] == twas_data['4_y'], temp['4_x'] == twas_data['5_y']),
np.logical_and(temp['3_x'] == twas_data['5_y'], temp['4_x'] == twas_data['4_y']))
twas_data.loc[strand_flipped, ['3_x', '4_x']] = temp
allele_flipped = np.logical_and(twas_data['3_x'] == twas_data['5_y'], twas_data['4_x'] == twas_data['4_y'])
twas_data.loc[allele_flipped, '3_x'], twas_data.loc[allele_flipped, '4_x'] = \
twas_data.loc[allele_flipped, '4_x'], twas_data.loc[allele_flipped, '3_x']
twas_data.loc[allele_flipped, '5_x'] *= -1
aligned = np.logical_and(twas_data['3_x'] == twas_data['4_y'], twas_data['4_x'] == twas_data['5_y'])
twas_data = twas_data.loc[aligned]
except:
# Hack: genes like ENSG00000215784.4 have garbage in snp_annot
print(f'warning: skipping corrupted model ({k}: {row[1]})')
continue
if twas_data.empty:
print(f'warning: no SNPs left ({k}: {row[1]})')
continue
elif twas_data.shape[0] < 0.5 * snp_annot.shape[0]:
print(f'warning: more than 50% SNPs lost ({k}: {row[1]})')
print(f'estimating sTWAS statistics ({k}: {row[1]})')
X = load_geno(f'/broad/hptmp/aksarkar/geno/{k}')
z, R, B, pip = align(twas_data, X, B, pip)
stat, pval = adaptive_permutation(z, R, B, pip)
fqtl_stats[row[1]] = stat
fqtl_pvals[row[1]] = pval
try:
snp_annot, B, pip = load_sqtl_model(f'/broad/compbio/ypp/gtex/analysis/fqtl-gtex/result/sqtl/{row[2]}/{row.name}/')
except FileNotFoundError:
print(f'warning: skipping empty sqtl model ({k}: {row[1]})')
continue
z, R, B, pip = align(twas_data, X, B, pip)
stat, pval = adaptive_permutation(z, R, B, pip)
sqtl_stats[row[1]] = stat
sqtl_pvals[row[1]] = pval
pd.DataFrame.from_dict(sqtl_stats, orient='index').to_csv(f'{sys.argv[3]}.sqtl.stat.txt.gz', sep='\t', compression='gzip')
pd.DataFrame.from_dict(sqtl_pvals, orient='index').to_csv(f'{sys.argv[3]}.sqtl.pval.txt.gz', sep='\t', compression='gzip')
pd.DataFrame.from_dict(fqtl_stats, orient='index').to_csv(f'{sys.argv[3]}.fqtl.stat.txt.gz', sep='\t', compression='gzip')
pd.DataFrame.from_dict(fqtl_pvals, orient='index').to_csv(f'{sys.argv[3]}.fqtl.pval.txt.gz', sep='\t', compression='gzip')