![]() |
ИСТИНА |
Войти в систему Регистрация |
ИПМех РАН |
||
In this study, we utilized the results of HT-SELEX in vitro experiments, which provided millions of 40-bp sequences that bind to the specific transcription factor. To predict the transcription factor binding to oligonucleotides, we employed our recent LegNet neural network architecture, which achieved 1st place in the DREAM-2022 contest. The model was enhanced to incorporate additional SELEX-specific information for training, including read counts and cycle numbers. LegNet neural network was used as a binary classifier of bound and non-bound sequences for the specific TF. The difference in obtained scores for a pair of sequences with SNPs was used to predict the significance of the variants on TF binding. Our deep learning model demonstrated superior performance compared to the current state-of-the-art gkmSVM model when trained on the same dataset.