Mercurial > repos > iuc > decontaminator
comparison predict.py @ 0:b856d3d95413 draft default tip
planemo upload for repository https://github.com/galaxyproject/tools-iuc/tree/master/tools/decontaminator commit 3f8e87001f3dfe7d005d0765aeaa930225c93b72
author | iuc |
---|---|
date | Mon, 09 Jan 2023 13:27:09 +0000 |
parents | |
children |
comparison
equal
deleted
inserted
replaced
-1:000000000000 | 0:b856d3d95413 |
---|---|
1 #!/usr/bin/env python | |
2 # -*- coding: utf-8 -*- | |
3 # Credits: Grigorii Sukhorukov, Macha Nikolski | |
4 import argparse | |
5 import os | |
6 from pathlib import Path | |
7 | |
8 import numpy as np | |
9 import pandas as pd | |
10 from Bio import SeqIO | |
11 from models import model_10 | |
12 from utils import preprocess as pp | |
13 | |
14 os.environ["CUDA_VISIBLE_DEVICES"] = "" | |
15 os.environ["TF_XLA_FLAGS"] = "--tf_xla_cpu_global_jit" | |
16 # loglevel : | |
17 # 0 all printed | |
18 # 1 I not printed | |
19 # 2 I and W not printed | |
20 # 3 nothing printed | |
21 os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3' | |
22 | |
23 | |
24 def predict_nn(ds_path, nn_weights_path, length, batch_size=256): | |
25 """ | |
26 Breaks down contigs into fragments | |
27 and uses pretrained neural networks to give predictions for fragments | |
28 """ | |
29 try: | |
30 seqs_ = list(SeqIO.parse(ds_path, "fasta")) | |
31 except FileNotFoundError: | |
32 raise Exception("test dataset was not found. Change ds variable") | |
33 | |
34 out_table = { | |
35 "id": [], | |
36 "length": [], | |
37 "fragment": [], | |
38 "pred_vir": [], | |
39 "pred_other": [], | |
40 } | |
41 if not seqs_: | |
42 raise ValueError("All sequences were smaller than length of the model") | |
43 test_fragments = [] | |
44 test_fragments_rc = [] | |
45 for seq in seqs_: | |
46 fragments_, fragments_rc, _ = \ | |
47 pp.fragmenting( | |
48 [seq], | |
49 length, | |
50 max_gap=0.8, | |
51 sl_wind_step=int(length / 2) | |
52 ) | |
53 test_fragments.extend(fragments_) | |
54 test_fragments_rc.extend(fragments_rc) | |
55 for j in range(len(fragments_)): | |
56 out_table["id"].append(seq.id) | |
57 out_table["length"].append(len(seq.seq)) | |
58 out_table["fragment"].append(j) | |
59 test_encoded = pp.one_hot_encode(test_fragments) | |
60 test_encoded_rc = pp.one_hot_encode(test_fragments_rc) | |
61 model = model_10.model(length) | |
62 model.load_weights(Path(nn_weights_path, f"model_{length}.h5")) | |
63 prediction = model.predict([test_encoded, test_encoded_rc], batch_size) | |
64 out_table['pred_vir'].extend(list(prediction[..., 1])) | |
65 out_table['pred_other'].extend(list(prediction[..., 0])) | |
66 print('Exporting predictions to csv file') | |
67 df = pd.DataFrame(out_table) | |
68 df['NN_decision'] = np.where(df['pred_vir'] > df['pred_other'], 'virus', 'other') | |
69 return df | |
70 | |
71 | |
72 def predict_test(ds_path, length): | |
73 """ | |
74 Breaks down contigs into fragments | |
75 and gives 1 as prediction to all fragments | |
76 use only for testing! | |
77 """ | |
78 try: | |
79 seqs_ = list(SeqIO.parse(ds_path, "fasta")) | |
80 except FileNotFoundError: | |
81 raise Exception("test dataset was not found. Change ds variable") | |
82 | |
83 out_table = { | |
84 "id": [], | |
85 "length": [], | |
86 "fragment": [], | |
87 } | |
88 if not seqs_: | |
89 raise ValueError("All sequences were smaller than length of the model") | |
90 for seq in seqs_: | |
91 fragments_, fragments_rc, _ = \ | |
92 pp.fragmenting( | |
93 [seq], | |
94 length, | |
95 max_gap=0.8, | |
96 sl_wind_step=int(length / 2) | |
97 ) | |
98 for j in range(len(fragments_)): | |
99 out_table["id"].append(seq.id) | |
100 out_table["length"].append(len(seq.seq)) | |
101 out_table["fragment"].append(j) | |
102 print('Exporting predictions to tsv file') | |
103 df = pd.DataFrame(out_table) | |
104 df['pred_vir'] = 1 | |
105 df['pred_other'] = 0 | |
106 df['NN_decision'] = 'virus' | |
107 return df | |
108 | |
109 | |
110 def predict_contigs(df): | |
111 """ | |
112 Based on predictions of predict_rf for fragments | |
113 gives a final prediction for the whole contig | |
114 """ | |
115 df = ( | |
116 df.groupby( | |
117 ["id", | |
118 "length", | |
119 'NN_decision'], | |
120 sort=False | |
121 ).size().unstack(fill_value=0) | |
122 ) | |
123 df = df.reset_index() | |
124 df = df.reindex( | |
125 ['length', 'id', 'virus', 'other', ], | |
126 axis=1 | |
127 ).fillna(value=0) | |
128 df['decision'] = np.where(df['virus'] >= df['other'], 'virus', 'other') | |
129 df = df.sort_values(by='length', ascending=False) | |
130 df = df.loc[:, ['length', 'id', 'virus', 'other', 'decision']] | |
131 df = df.rename( | |
132 columns={ | |
133 'virus': '# viral fragments', | |
134 'other': '# other fragments', | |
135 } | |
136 ) | |
137 df['# viral / # total'] = (df['# viral fragments'] / (df['# viral fragments'] + df['# other fragments'])).round(3) | |
138 df['# viral / # total * length'] = df['# viral / # total'] * df['length'] | |
139 df = df.sort_values(by='# viral / # total * length', ascending=False) | |
140 return df | |
141 | |
142 | |
143 def predict(test_ds, weights, out_path, return_viral=True): | |
144 """filters out contaminant contigs from the fasta file. | |
145 | |
146 test_ds: path to the input file with | |
147 contigs in fasta format (str or list of str) | |
148 weights: path to the folder containing weights | |
149 for NN and RF modules trained on 500 and 1000 fragment lengths (str) | |
150 out_path: path to the folder to store predictions (str) | |
151 return_viral: whether to return contigs annotated as | |
152 viral in separate fasta file (True/False) | |
153 """ | |
154 | |
155 test_ds = test_ds | |
156 if isinstance(test_ds, list): | |
157 pass | |
158 elif isinstance(test_ds, str): | |
159 test_ds = [test_ds] | |
160 else: | |
161 raise ValueError('test_ds was incorrectly assigned in the config file') | |
162 | |
163 assert Path(test_ds[0]).exists(), f'{test_ds[0]} does not exist' | |
164 # assert Path(weights).exists(), f'{weights} does not exist' | |
165 limit = 0 | |
166 Path(out_path).mkdir(parents=True, exist_ok=True) | |
167 | |
168 # parameter to activate test function. Only for debugging on github | |
169 # test is launched when the weights directory is empty | |
170 use_test_f = not Path(weights, 'model_1000.h5').exists() | |
171 for ts in test_ds: | |
172 dfs_fr = [] | |
173 dfs_cont = [] | |
174 for l_ in 500, 1000: | |
175 print(f'starting prediction for {Path(ts).name} ' | |
176 f'for fragment length {l_}') | |
177 if use_test_f: | |
178 df = predict_test(ds_path=ts, length=l_, ) | |
179 else: | |
180 df = predict_nn( | |
181 ds_path=ts, | |
182 nn_weights_path=weights, | |
183 length=l_, | |
184 ) | |
185 df = df.round(3) | |
186 dfs_fr.append(df) | |
187 df = predict_contigs(df) | |
188 dfs_cont.append(df) | |
189 print('prediction finished') | |
190 df_500 = dfs_fr[0][(dfs_fr[0]['length'] >= limit) & (dfs_fr[0]['length'] < 1500)] | |
191 df_1000 = dfs_fr[1][(dfs_fr[1]['length'] >= 1500)] | |
192 df = pd.concat([df_1000, df_500], ignore_index=True) | |
193 pred_fr = Path(out_path, "predicted_fragments.tsv") | |
194 df.to_csv(pred_fr, sep='\t') | |
195 | |
196 df_500 = dfs_cont[0][(dfs_cont[0]['length'] | |
197 >= limit) & (dfs_cont[0]['length'] < 1500)] | |
198 df_1000 = dfs_cont[1][(dfs_cont[1]['length'] | |
199 >= 1500)] | |
200 df = pd.concat([df_1000, df_500], ignore_index=True) | |
201 pred_contigs = Path(out_path, "predicted.tsv") | |
202 df.to_csv(pred_contigs, sep='\t') | |
203 | |
204 if return_viral: | |
205 viral_ids = list(df[df["decision"] == "virus"]["id"]) | |
206 seqs_ = list(SeqIO.parse(ts, "fasta")) | |
207 viral_seqs = [s_ for s_ in seqs_ if s_.id in viral_ids] | |
208 SeqIO.write( | |
209 viral_seqs, | |
210 Path( | |
211 out_path, | |
212 "viral.fasta"), 'fasta') | |
213 | |
214 | |
215 if __name__ == '__main__': | |
216 parser = argparse.ArgumentParser() | |
217 parser.add_argument("--test_ds", help="path to the input " | |
218 "file with contigs " | |
219 "in fasta format " | |
220 "(str or list of str)") | |
221 parser.add_argument("--weights", help="path to the folder containing " | |
222 "weights for NN and RF modules " | |
223 "trained on 500 and 1000 " | |
224 "fragment lengths (str)") | |
225 parser.add_argument("--out_path", help="path to the folder to store " | |
226 "predictions (str)") | |
227 parser.add_argument("--return_viral", help="whether to return " | |
228 "contigs annotated " | |
229 "as viral in separate " | |
230 "fasta file (True/False)") | |
231 | |
232 args = parser.parse_args() | |
233 if args.test_ds: | |
234 test_ds = args.test_ds | |
235 if args.weights: | |
236 weights = args.weights | |
237 if args.out_path: | |
238 out_path = args.out_path | |
239 if args.return_viral: | |
240 return_viral = args.return_viral | |
241 | |
242 predict(test_ds, weights, out_path, return_viral) |