Atom Bioworks
commited on
Create utils.py
Browse files
utils.py
ADDED
|
@@ -0,0 +1,199 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import numpy as np
|
| 2 |
+
import random
|
| 3 |
+
import math
|
| 4 |
+
|
| 5 |
+
from sklearn.metrics import *
|
| 6 |
+
|
| 7 |
+
import torch
|
| 8 |
+
import torch.nn as nn
|
| 9 |
+
import torch.nn.functional as F
|
| 10 |
+
from torch.utils.data import Dataset
|
| 11 |
+
import pickle
|
| 12 |
+
|
| 13 |
+
|
| 14 |
+
def word2idx(word, words):
|
| 15 |
+
if word in words.keys():
|
| 16 |
+
return int(words[word])
|
| 17 |
+
|
| 18 |
+
return 0
|
| 19 |
+
|
| 20 |
+
def pad_seq(dataset, max_len):
|
| 21 |
+
output = []
|
| 22 |
+
for seq in dataset:
|
| 23 |
+
pad = np.zeros(max_len)
|
| 24 |
+
pad[:len(seq)] = seq
|
| 25 |
+
output.append(pad)
|
| 26 |
+
|
| 27 |
+
return np.array(output)
|
| 28 |
+
|
| 29 |
+
def str2bool(seq):
|
| 30 |
+
out = []
|
| 31 |
+
for s in seq:
|
| 32 |
+
if s == "positive":
|
| 33 |
+
out.append(1)
|
| 34 |
+
elif s == "negative":
|
| 35 |
+
out.append(0)
|
| 36 |
+
|
| 37 |
+
return np.array(out)
|
| 38 |
+
|
| 39 |
+
class API_Dataset(Dataset):
|
| 40 |
+
def __init__(self, apta, esm_prot, y, apta_attn_mask, prot_attn_mask):
|
| 41 |
+
super(Dataset, self).__init__()
|
| 42 |
+
|
| 43 |
+
self.apta = np.array(apta, dtype=np.int64)
|
| 44 |
+
self.esm_prot = np.array(esm_prot, dtype=np.int64)
|
| 45 |
+
self.y = np.array(y, dtype=np.int64)
|
| 46 |
+
self.apta_attn_mask = np.array(apta_attn_mask)
|
| 47 |
+
self.prot_attn_mask = np.array(prot_attn_mask)
|
| 48 |
+
self.len = len(self.apta)
|
| 49 |
+
|
| 50 |
+
def __len__(self):
|
| 51 |
+
return self.len
|
| 52 |
+
|
| 53 |
+
def __getitem__(self, index):
|
| 54 |
+
return torch.tensor(self.apta[index], dtype=torch.int64), torch.tensor(self.esm_prot[index], dtype=torch.int64), torch.tensor(self.y[index], dtype=torch.int64), torch.tensor(self.apta_attn_mask[index], dtype=torch.int64), torch.tensor(self.prot_attn_mask[index], dtype=torch.int64)
|
| 55 |
+
|
| 56 |
+
def find_opt_threshold(target, pred):
|
| 57 |
+
result = 0
|
| 58 |
+
best = 0
|
| 59 |
+
|
| 60 |
+
for i in range(0, 1000):
|
| 61 |
+
pred_threshold = np.where(pred > i/1000, 1, 0)
|
| 62 |
+
now = f1_score(target, pred_threshold)
|
| 63 |
+
if now > best:
|
| 64 |
+
result = i/1000
|
| 65 |
+
best = now
|
| 66 |
+
|
| 67 |
+
return result
|
| 68 |
+
|
| 69 |
+
def argument_seqset(seqset):
|
| 70 |
+
arg_seqset = []
|
| 71 |
+
for s, ss in seqset:
|
| 72 |
+
arg_seqset.append([s, ss])
|
| 73 |
+
|
| 74 |
+
arg_seqset.append([s[::-1], ss[::-1]])
|
| 75 |
+
|
| 76 |
+
return arg_seqset
|
| 77 |
+
|
| 78 |
+
def augment_apis(apta, prot, ys):
|
| 79 |
+
aug_apta = []
|
| 80 |
+
aug_prot = []
|
| 81 |
+
aug_y = []
|
| 82 |
+
for a, p, y in zip(apta, prot, ys):
|
| 83 |
+
aug_apta.append(a)
|
| 84 |
+
aug_prot.append(p)
|
| 85 |
+
aug_y.append(y)
|
| 86 |
+
|
| 87 |
+
aug_apta.append(a[::-1])
|
| 88 |
+
aug_prot.append(p)
|
| 89 |
+
aug_y.append(y)
|
| 90 |
+
|
| 91 |
+
return np.array(aug_apta), np.array(aug_prot), np.array(aug_y)
|
| 92 |
+
|
| 93 |
+
|
| 94 |
+
|
| 95 |
+
def load_data_source(filepath):
|
| 96 |
+
with open(filepath,"rb") as fr:
|
| 97 |
+
dataset = pickle.load(fr)
|
| 98 |
+
dataset_train = np.array(dataset[dataset["dataset"]=="training dataset"])
|
| 99 |
+
dataset_test = np.array(dataset[dataset["dataset"]=="test dataset"])
|
| 100 |
+
dataset_bench = np.array(dataset[dataset['dataset']=='benchmark dataset'])
|
| 101 |
+
|
| 102 |
+
return dataset_train, dataset_test, dataset_bench
|
| 103 |
+
|
| 104 |
+
|
| 105 |
+
def get_dataset(filepath, prot_max_len, n_prot_vocabs, prot_words):
|
| 106 |
+
dataset_train, dataset_test, dataset_bench = load_data_source(filepath)
|
| 107 |
+
|
| 108 |
+
|
| 109 |
+
arg_apta, arg_prot, arg_y = augment_apis(dataset_train[:, 0], dataset_train[:, 1], dataset_train[:, 2])
|
| 110 |
+
datasets_train = [rna2vec(arg_apta), tokenize_sequences(arg_prot, prot_max_len, n_prot_vocabs, prot_words), str2bool(arg_y)]
|
| 111 |
+
datasets_test = [rna2vec(dataset_test[:, 0]), tokenize_sequences(dataset_test[:, 1], prot_max_len, n_prot_vocabs, prot_words), str2bool(dataset_test[:, 2])]
|
| 112 |
+
datasets_bench = [rna2vec(dataset_bench[:, 0]), tokenize_sequences(dataset_bench[:, 1], prot_max_len, n_prot_vocabs, prot_words), str2bool(dataset_bench[:, 2])]
|
| 113 |
+
|
| 114 |
+
return datasets_train, datasets_test, datasets_bench
|
| 115 |
+
|
| 116 |
+
|
| 117 |
+
def get_esm_dataset(filepath, batch_converter, alphabet):
|
| 118 |
+
dataset_train, dataset_test, dataset_bench = load_data_source(filepath)
|
| 119 |
+
|
| 120 |
+
# arg_apta, arg_prot, arg_y = augment_apis(dataset_train[:, 0], dataset_train[:, 1], dataset_train[:, 2])
|
| 121 |
+
# arg_prot is a np.array of strings (4640,) -> convert this to np.array of size (2x4640) where first row is a label
|
| 122 |
+
|
| 123 |
+
arg_apta, arg_prot, arg_y = dataset_train[:, 0], dataset_train[:, 1], dataset_train[:, 2]
|
| 124 |
+
arg_apta, arg_prot, arg_y = augment_apis(arg_apta, arg_prot, arg_y)
|
| 125 |
+
|
| 126 |
+
train_inputs = [(i, j) for i, j in zip(arg_y, arg_prot)]
|
| 127 |
+
_, _, prot_tokens = batch_converter(train_inputs)
|
| 128 |
+
datasets_train = [rna2vec(arg_apta), prot_tokens, str2bool(arg_y)]
|
| 129 |
+
|
| 130 |
+
test_inputs = [(i, j) for i, j in enumerate(dataset_test[:, 1])]
|
| 131 |
+
_, _, test_prot_tokens = batch_converter(test_inputs)
|
| 132 |
+
datasets_test = [rna2vec(dataset_test[:, 0]), test_prot_tokens, str2bool(dataset_test[:, 2])]
|
| 133 |
+
|
| 134 |
+
bench_inputs = [(i, j) for i, j in enumerate(dataset_bench[:, 1])]
|
| 135 |
+
_, _, bench_prot_tokens = batch_converter(bench_inputs)
|
| 136 |
+
# truncating
|
| 137 |
+
bench_prot_tokenized = bench_prot_tokens[:, :1678]
|
| 138 |
+
# padding
|
| 139 |
+
prot_ex = torch.ones((bench_prot_tokenized.shape[0], 1678), dtype=torch.int64)*alphabet.padding_idx
|
| 140 |
+
prot_ex[:, :bench_prot_tokenized.shape[1]] = bench_prot_tokenized
|
| 141 |
+
datasets_bench = [rna2vec(dataset_bench[:, 0]), prot_ex, str2bool(dataset_bench[:, 2])]
|
| 142 |
+
|
| 143 |
+
return datasets_train, datasets_test, datasets_bench
|
| 144 |
+
|
| 145 |
+
def get_nt_esm_dataset(filepath, nt_tokenizer, batch_converter, alphabet):
|
| 146 |
+
dataset_train, dataset_test, dataset_bench = load_data_source(filepath)
|
| 147 |
+
|
| 148 |
+
arg_apta, arg_prot, arg_y = augment_apis(dataset_train[:, 0], dataset_train[:, 1], dataset_train[:, 2])
|
| 149 |
+
# arg_prot is a np.array of strings (4640,) -> convert this to np.array of size (2x4640) where first row is a label
|
| 150 |
+
max_length = 275#nt_tokenizer.model_max_length
|
| 151 |
+
|
| 152 |
+
train_inputs = [(i, j) for i, j in zip(arg_y, arg_prot)]
|
| 153 |
+
_, _, prot_tokens = batch_converter(train_inputs)
|
| 154 |
+
apta_toks = nt_tokenizer.batch_encode_plus(arg_apta, return_tensors='pt', padding='max_length', max_length=max_length)['input_ids']
|
| 155 |
+
apta_attention_mask = apta_toks != nt_tokenizer.pad_token_id
|
| 156 |
+
prot_attention_mask = prot_tokens != alphabet.padding_idx
|
| 157 |
+
# datasets_train = [apta_toks, prot_tokens, str2bool(arg_y)]
|
| 158 |
+
datasets_train = [apta_toks, prot_tokens, str2bool(arg_y), apta_attention_mask, prot_attention_mask]
|
| 159 |
+
|
| 160 |
+
test_inputs = [(i, j) for i, j in enumerate(dataset_test[:, 1])]
|
| 161 |
+
_, _, test_prot_tokens = batch_converter(test_inputs)
|
| 162 |
+
prot_ex = torch.ones((test_prot_tokens.shape[0], 1680), dtype=torch.int64)*alphabet.padding_idx
|
| 163 |
+
prot_ex[:, :test_prot_tokens.shape[1]] = test_prot_tokens
|
| 164 |
+
apta_toks = nt_tokenizer.batch_encode_plus(dataset_test[:, 0], return_tensors='pt', padding='max_length', max_length=max_length)['input_ids']
|
| 165 |
+
apta_attention_mask = apta_toks != nt_tokenizer.pad_token_id
|
| 166 |
+
prot_attention_mask = prot_ex != alphabet.padding_idx
|
| 167 |
+
datasets_test = [apta_toks, prot_ex, str2bool(dataset_test[:, 2]), apta_attention_mask, prot_attention_mask]
|
| 168 |
+
|
| 169 |
+
bench_inputs = [(i, j) for i, j in enumerate(dataset_bench[:, 1])]
|
| 170 |
+
_, _, bench_prot_tokens = batch_converter(bench_inputs)
|
| 171 |
+
# padding
|
| 172 |
+
prot_ex = torch.ones((bench_prot_tokens.shape[0], 1680), dtype=torch.int64)*alphabet.padding_idx
|
| 173 |
+
prot_ex[:, :bench_prot_tokens.shape[1]] = bench_prot_tokens
|
| 174 |
+
apta_toks = nt_tokenizer.batch_encode_plus(dataset_bench[:, 0], return_tensors='pt', padding='max_length', max_length=max_length)['input_ids']
|
| 175 |
+
apta_attention_mask = apta_toks != nt_tokenizer.pad_token_id
|
| 176 |
+
prot_attention_mask = prot_ex != alphabet.padding_idx
|
| 177 |
+
datasets_bench = [apta_toks, prot_ex, str2bool(dataset_bench[:, 2]), apta_attention_mask, prot_attention_mask]
|
| 178 |
+
|
| 179 |
+
return datasets_train, datasets_test, datasets_bench
|
| 180 |
+
|
| 181 |
+
def get_scores(target, pred):
|
| 182 |
+
threshold = find_opt_threshold(target, pred)
|
| 183 |
+
pred_threshold = np.where(pred > threshold, 1, 0)
|
| 184 |
+
acc = accuracy_score(target, pred_threshold)
|
| 185 |
+
roc_auc = roc_auc_score(target, pred)
|
| 186 |
+
mcc = matthews_corrcoef(target, pred_threshold)
|
| 187 |
+
f1 = f1_score(target, pred_threshold)
|
| 188 |
+
pr_auc = average_precision_score(target, pred)
|
| 189 |
+
cls_report = classification_report(target, pred_threshold)
|
| 190 |
+
scores = {
|
| 191 |
+
'threshold': threshold,
|
| 192 |
+
'acc': acc,
|
| 193 |
+
'roc_auc': roc_auc,
|
| 194 |
+
'mcc': mcc,
|
| 195 |
+
'f1': f1,
|
| 196 |
+
'pr_auc': pr_auc,
|
| 197 |
+
'cls_report': cls_report
|
| 198 |
+
}
|
| 199 |
+
return scores
|