Drug Discovery using Deep Learning
Jayesh Sharma, Shyam Mourya, Rounak Rai, Rushikesh Nikam · Zenodo (CERN European Organization for Nuclear Research) · 2021
The SARS-CoV-2 infection has killed over 3.9 million people, indicating there is an urgent need for effective treatment. This, however, cannot be accomplished with present drug development or application systems, as it takes several years for newly discovered drugs to reach the market. In this project we have tried to identify commercially available anti-viral drugs, synthetic molecules that could potentially disrupt SARS-CoV-2's viral components. Our aim was to bind our molecule with the main enzyme of SARS-CoV-2 slowing the virus's replication process enabling our body to fight against the virus. We first took a large number of molecules and fed them to an RNN-LSTM. The molecules which would be fed would be in a format similar to a string. The RNN would then identify the patterns and rules from these molecules using them to generate molecules which are currently not in existence but could be synthesized later in the future. Later we combined the new molecules and the pre-existing molecules, forming a new set. A diverse subset is then selected on which molecular docking was then performed with the SARS-CoV-2 virus's main protease and potential inhibitors were identified.