-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathutils.py
209 lines (196 loc) · 8.51 KB
/
utils.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
import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split,KFold
from tqdm import tqdm
import copy
from collections import defaultdict
import Levenshtein
from tcrpeg.TCRpeg import TCRpeg
from model import TEINet
import torch
def split(path,record_dic,fold=5,sep=','):
'''
Split the data to k-fold
@path: path to the file
@record_dic: the directory to store the processed k fold data
@sep: seperator for the file type (csv:','; tsv:'\t')
'''
data = pd.read_csv(path,sep=sep)
kf = KFold(n_splits=fold,shuffle=True,random_state=42)
i= 1
for train_idx, test_idx in kf.split(data):
df_train,df_test = data.iloc[train_idx].reset_index(drop=True),data.iloc[test_idx].reset_index(drop=True)
df_train.to_csv('{}/train_{}_positive.csv'.format(record_dic,i),index=False)
df_test.to_csv('{}/test_{}_positive.csv'.format(record_dic,i),index=False)
i += 1
def epitope_sample_1fold(positive_file,record_file,sample_num=1,fre=True):
'''
Negative sampling. For each TCR, sample a epitope as its negative.
The output file contains both positive (input) and negative pairs
@positive_file: path to the file that records the positive pairs
@record_file: path to the output file
@sample_num: for each TCR, the number of sampled epitopes
@fre: epitope sampled randomly (False) or based on their frequency distribution (True)
'''
data = pd.read_csv(positive_file)
cdrs,epitopes = data['CDR3.beta'].values, data['Epitope'].values
cdrs_new,epitopes_new = [],[]
cdrs_original,epitopes_original = [],[]
t2e = defaultdict(set)
for i in range(len(cdrs)):
if len(epitopes[i]) <= 15:
t2e[cdrs[i]].add(epitopes[i])
for i in tqdm(range(len(cdrs))):
tcr,e = cdrs[i],epitopes[i]
cdrs_original.append(tcr)
epitopes_original.append(e)
to_remove = t2e[tcr]
e2f = defaultdict(int)
sum_ = 0
for ee in epitopes: #
if ee not in to_remove:
e2f[ee] += 1
sum_ += 1
if not fre:
sample = np.random.choice(np.array(list(e2f.keys())),sample_num,replace=False)
else :
sample = np.random.choice(np.array(list(e2f.keys())),sample_num,replace=False,p = [e2f[x]/sum_ for x in e2f.keys()])
epitopes_new.extend(sample)
cdrs_new.extend([tcr] * sample_num)
processed_data = pd.DataFrame(columns=['CDR3.beta','Epitope','Label'])
processed_data['CDR3.beta'] = cdrs_original + cdrs_new
processed_data['Epitope'] = epitopes_original + epitopes_new
processed_data['Label'] = [1] * len(cdrs_original) + [0]*len(cdrs_new)
processed_data.to_csv(record_file,index=False)
def tcr_sample_1fold(positive_file,record_file,sample_num=1,reference_file='None'):
'''
Negative sampling. For each epitope, sample a tcr as its negative.
The output file contains both positive (input) and negative pairs
@positive_file: path to the file that records the positive pairs
@record_file: path to the output file
@sample_num: for each epitope, the number of sampled tcrs
@reference_file: Path to reference TCR; If specified, will sample TCR from the reference data
'''
data = pd.read_csv(positive_file)
cdrs,epitopes = data['CDR3.beta'].values, data['Epitope'].values
cdrs_new,epitopes_new = [],[]
cdrs_original,epitopes_original = [],[]
reference=False
if reference_file != 'None':
refer_tcr = pd.read_csv(reference_file)['CDR3.beta'].values
reference=True
# epitope_reference = set(epitopes)
e2t = defaultdict(set)
for i in range(len(cdrs)):
if len(epitopes[i]) <= 15:
e2t[epitopes[i]].add(cdrs[i])
for i in tqdm(range(len(cdrs))):
tcr,e = cdrs[i],epitopes[i]
cdrs_original.append(tcr)
epitopes_original.append(e)
to_remove = e2t[e]
to_sample = set()
for t in cdrs:
if t not in to_remove:
to_sample.add(t)
if reference:
sample = np.random.choice(refer_tcr,sample_num,replace=False)
else :
sample = np.random.choice(np.array(list(to_sample)),sample_num,replace=False)
epitopes_new.extend([e] * sample_num)
cdrs_new.extend(sample)
processed_data = pd.DataFrame(columns=['CDR3.beta','Epitope','Label'])
processed_data['CDR3.beta'] = cdrs_original + cdrs_new
processed_data['Epitope'] = epitopes_original + epitopes_new
processed_data['Label'] = [1] * len(cdrs_original) + [0]*len(cdrs_new)
processed_data.to_csv(record_file,index=False)
def compute_metric(pos_data,predictions,k=3):
'''
Compute Precision, Recall, and NDCG
@pos_data: dict, key is [tcr]->[e1,e2,...]
@predictions: dict, key is [tcr] -> [(e1,scores1),(e2,scores2), .... ]
'''
recalls,precisions = [],[]
ndcgs = []
for tcr in predictions.keys():
pres, trues = predictions[tcr], set(pos_data[tcr])
pres.sort(key=lambda x: -x[1])
pres = [p[0] for p in pres]
count = 0
dcp = 0
for i,p in enumerate(pres[:k]):
if p in trues:
count += 1
dcp += 1 / np.log2(i+1 + 1)
idcp = sum([1/np.log2(i+1+1) for i in range(min(len(trues),k))])
ndcg = dcp / idcp if idcp > 0 else 0
ndcgs.append(ndcg)
precisions.append(count / k)
recalls.append(count / len(trues))
print('Precision, Recall, NDCG at top {} are:'.format(k))
print(str(np.mean(precisions)) + ', ',str(np.mean(recalls)) + ', '+str(np.mean(ndcgs)))
return np.mean(precisions), np.mean(recalls), np.mean(ndcgs)
def levenstein_filter(path_ref,path_to_filter, threshold,record_path=None):
'''
Filter TCRs based on levenstein distance
@path_ref: path to the reference data (namely the training set)
@path_to_filter: path to the data that needs filtering (namely the test set)
@threshold: filter threshold (0 to 1)
@record_path: if specified, will store the filtered data
'''
cdrs_ref = pd.read_csv(path_ref)['CDR3.beta'].values
data_filter = pd.read_csv(path_to_filter)
cdrs_to_filter = data_filter['CDR3.beta'].values
to_filter = []
for i in tqdm(range(len(cdrs_to_filter))):
for cdr_ref in cdrs_ref:
if Levenshtein.ratio(cdr_ref, cdrs_to_filter[i]) >= threshold:
to_filter.append(i)
break
data = data_filter.drop(to_filter)
print('The original dataset has {} pairs'.format(len(cdrs_to_filter)))
print('After filtering there are {} pairs'.format(len(data)))
if record_path is not None:
data.to_csv(record_path,index=False)
def aa_scan(seq,scan_aa='A',scan_aa_alternate='G',offset = 0):
seq = list(seq)
return_seq = []
for i in range(offset,len(seq)-offset):
s_ = copy.copy(seq)
s_[i] = scan_aa
if s_ != seq:
return_seq.append(''.join(s_))
else :
s_[i] = scan_aa_alternate
return_seq.append(''.join(s_))
return return_seq
def contact_index(dis,axis = 0,threshold = 5,offset=1):
#original is 5 threshold
#start from 0
assert axis in [0,1]
c_index = []
for i in range(offset,dis.shape[axis]-offset):
compare = dis[i] if axis == 0 else dis[:,i]
if sum(compare <= threshold) > 0:
c_index.append(i)
return c_index
def mean_dis(dis,axis = 0,threshold = 5,offset=1):
#original is 5 threshold
#start from 0
assert axis in [0,1]
mean_distance = []
for i in range(offset,dis.shape[axis]-offset):
compare = dis[i] if axis == 0 else dis[:,i]
# if sum(compare <= threshold) > 0:
# c_index.append(i)
mean_distance.append(np.mean(compare))
return mean_distance
def load_teinet(path,emb_path1='encoders/aa_emb_tcr.txt',emb_path2 = 'encoders/aa_emb_tcr.txt',device='cuda:0',normalize=True,weight_decay = 0):
model_tcr = TCRpeg(hidden_size=768,num_layers = 3,load_data=False,embedding_path=emb_path1,device=device)
model_tcr.create_model()
model_epi = TCRpeg(hidden_size=768,num_layers = 3,load_data=False,embedding_path=emb_path2,device=device)
model_epi.create_model()
model = TEINet(en_tcr=model_tcr,en_epi = model_epi,cat_size=768*2,dropout=0.1,normalize=True,weight_decay = 0,device=device)
model.load_state_dict(torch.load(path))
model = model.to(device)
return model