DeepDuSite

Method for dU sites prediction

Pretrain

Using BERT-pytorch for pretrained, pretrained sequence data could download from UCSC. The please transfer coodinate to DNA sequence, and then segment them to the corpus of one line with two 400bp adjacent DNA sequences.

Train

Using bert_ensemble_train.py for training.

python bert_ensemble_train.py \
--train_corpus train_label.csv \
--validation_corpus valid_label.csv \
--bert_path bert.model.ep9 \
--saved_path ./

Test

Using bert_ensemble_test.py for test.

python bert_ensemble_test.py \
--test_corpus test_label.csv \
--model_path bert.ensemble.ep1