Upload app.py
Browse files
app.py
ADDED
@@ -0,0 +1,94 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import numpy as np
|
2 |
+
from transformers import AutoTokenizer, AutoModelForSequenceClassification
|
3 |
+
import pandas as pd
|
4 |
+
from transformers import set_seed
|
5 |
+
import torch
|
6 |
+
import torch.nn as nn
|
7 |
+
from collections import OrderedDict
|
8 |
+
import warnings
|
9 |
+
import gradio as gr
|
10 |
+
|
11 |
+
warnings.filterwarnings('ignore')
|
12 |
+
set_seed(4)
|
13 |
+
device = "cuda:0"
|
14 |
+
model_checkpoint = "facebook/esm2_t30_150M_UR50D"
|
15 |
+
|
16 |
+
class MyModel(nn.Module):
|
17 |
+
def __init__(self):
|
18 |
+
super().__init__()
|
19 |
+
self.bert = AutoModelForSequenceClassification.from_pretrained(model_checkpoint,num_labels=320)
|
20 |
+
self.bn1 = nn.BatchNorm1d(256)
|
21 |
+
self.bn2 = nn.BatchNorm1d(128)
|
22 |
+
self.bn3 = nn.BatchNorm1d(64)
|
23 |
+
self.relu = nn.ReLU()
|
24 |
+
self.fc1 = nn.Linear(320,256)
|
25 |
+
self.fc2 = nn.Linear(256,128)
|
26 |
+
self.fc3 = nn.Linear(128,64)
|
27 |
+
self.output_layer = nn.Linear(64,2)
|
28 |
+
self.dropout = nn.Dropout(0)
|
29 |
+
|
30 |
+
def forward(self,x):
|
31 |
+
with torch.no_grad():
|
32 |
+
bert_output = self.bert(input_ids=x['input_ids'].to(device),attention_mask=x['attention_mask'].to(device))
|
33 |
+
output_feature = self.dropout(bert_output["logits"])
|
34 |
+
output_feature = self.relu(self.bn1(self.fc1(output_feature)))
|
35 |
+
output_feature = self.relu(self.bn2(self.fc2(output_feature)))
|
36 |
+
output_feature = self.relu(self.bn3(self.fc3(output_feature)))
|
37 |
+
output_feature = self.output_layer(output_feature)
|
38 |
+
return torch.softmax(output_feature,dim=1)
|
39 |
+
|
40 |
+
def Kmers_funct(seq,num):
|
41 |
+
for i in range(len(seq)):
|
42 |
+
a = seq[i]
|
43 |
+
l = []
|
44 |
+
for index in range(len(a)):
|
45 |
+
t = a[index:index + num]
|
46 |
+
if (len(t)) == num:
|
47 |
+
l.append(t)
|
48 |
+
return l
|
49 |
+
|
50 |
+
def ACE(file):
|
51 |
+
model = MyModel()
|
52 |
+
model.load_state_dict(torch.load("best_model.pth"))
|
53 |
+
model = model.to(device)
|
54 |
+
model.eval()
|
55 |
+
tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)
|
56 |
+
test_sequences = file
|
57 |
+
max_len = 30
|
58 |
+
test_data = tokenizer(test_sequences, max_length=max_len, padding="max_length",truncation=True, return_tensors='pt')
|
59 |
+
out_probability = []
|
60 |
+
with torch.no_grad():
|
61 |
+
predict = model(test_data)
|
62 |
+
out_probability.extend(np.max(np.array(predict.cpu()),axis=1).tolist())
|
63 |
+
test_argmax = np.argmax(predict.cpu(), axis=1).tolist()
|
64 |
+
id2str = {0:"non-ACE", 1:"ACE"}
|
65 |
+
return id2str[test_argmax[0]], out_probability[0]
|
66 |
+
|
67 |
+
def main(file):
|
68 |
+
test_seq = file
|
69 |
+
all = []
|
70 |
+
seq_all = []
|
71 |
+
output_all = []
|
72 |
+
probability_all = []
|
73 |
+
for j in range(2, 11):
|
74 |
+
X = Kmers_funct([test_seq], j)
|
75 |
+
all.extend(X)
|
76 |
+
for seq in all:
|
77 |
+
output, probability = ACE(str(seq))
|
78 |
+
seq_all.append(seq)
|
79 |
+
output_all.append(output)
|
80 |
+
probability_all.append(probability)
|
81 |
+
|
82 |
+
summary = OrderedDict()
|
83 |
+
summary['Seq'] = seq_all
|
84 |
+
summary['Class'] = output_all
|
85 |
+
summary['Probability'] = probability_all
|
86 |
+
summary_df = pd.DataFrame(summary)
|
87 |
+
summary_df.to_csv('output.csv', index=False)
|
88 |
+
return 'outputs.csv'
|
89 |
+
|
90 |
+
|
91 |
+
iface = gr.Interface(fn=main,
|
92 |
+
inputs="text",
|
93 |
+
outputs= "file")
|
94 |
+
iface.launch()
|