| Popis: |
Computational Chemistry offers accurate physics-based simulation methods to assist in drug discovery and protein modeling problems. The associated computational costs grow significantly with a number of atoms. Machine learning algorithms solve this problem by accelerating computations while preserving accuracy. Recently, a deep neural network has obtained nearly exact solutions of the electronic Schrödinger equation for small molecules. Another recent breakthrough is the prediction of three-dimensional structures of proteins by neural network-based methods, Alphafold and RoseTTafold. In my PhD, I worked on three different domains of computational chemistry where we integrated machine learning to improve as well as accelerate conventional techniques; 1) Designed an efficient workflow for drug discovery by integrating a hybrid neural-network (pseudo-quantum) and classical forcefields based molecular dynamics simulations against COVID-19 main protease (MPRO), 2) Developed a machine learning- enabled pipeline for large-scale virtual drug screening using clustering and deep learning in combination with physics-based approaches against RPN11 - a drug target for breast cancer, and 3) Implemented a generative deep network to guide conformational sampling of intrinsically disordered proteins; an efficient protocol to accelerate molecular dynamics simulations. Together, these PhD projects present a new perspective on how machine learning can be used in the rapid and accurate screening of new drugs and accelerating molecular dynamics simulations for mining proteins conformational landscape. With the advances in the machine learning field, this work will have many other possible applications to help accelerate molecular dynamics and drug-discovery platforms. |
