Predicting the Response to Therapy with Deep Neural Networks using Virus Sequence
share
Published in NIPS 2016 Workshop on Computational Biology The use of anti-retroviral treatment is the gold standard for HIV-1 therapy. How- ever, treatment failures can occur and patients are still at risk of developing drug resistance. The high mutation rates of the virus and its complex interaction with the host genome are challenging problems in computational biology to develop new models for drug discovery. Deep Learning (DL) has shown breakthrough performances in a wide spectrum of applications including biological sequence modeling. DL has the ability to model patterns and dynamic behaviors from se- quences. We propose a deep neural network, DeepHIV, that combines Convolution Networks and Recurrent Networks to learn a shared representation from the two virus sequences, Protease (PR) and Reverse Transcriptase (RT). We compared DeepHIV with Random Forests trained on manually curated features from known mutations and epitopes. The preliminary results on a publicly available dataset of 1000 patients showed that DeepHIV has a good performance compared with Random Forests, the winner method in Kaggle challenge. We demonstrated, with examples, how DeepHIV can be used to identify regions in the virus sequences with high risk of causing drug resistance.
READ FULL TEXT