2021/04/09 by Garrett A. Stevenson, Derek C. Jones, Stevenson, Garrett A. +61
Biochemistry, Genetics and Molecular Biology · Computer Science · #Biomolecules (q-bio.BM) #Computational Drug Discovery Methods #FOS: Biological sciences #FOS: Computer and information sciences #Machine Learning (cs.LG) #Protein Structure and Dynamics #vaccines and immunoinformatics approaches
paper · pdf · doi:10.48550/arxiv.2104.04547
openalex publication_date 2021/04/09 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Structure-based Deep Fusion models were recently shown to outperform several\nphysics- and machine learning-based protein-ligand binding affinity prediction\nmethods. As part of a multi-institutional COVID-19 pandemic response, over 500\nmillion small molecules were computationally screened against four protein\nstructures from the novel coronavirus (SARS-CoV-2), which causes COVID-19.\nThree enhancements to Deep Fusion were made in order to evaluate more than 5\nbillion docked poses on SARS-CoV-2 protein targets. First, the Deep Fusion\nconcept was refined by formulating the architecture as one, coherently\nbackpropagated model (Coherent Fusion) to improve binding-affinity prediction\naccuracy. Secondly, the model was trained using a distributed, genetic\nhyper-parameter optimization. Finally, a scalable, high-throughput screening\ncapability was developed to maximize the number of ligands evaluated and\nexpedite the path to experimental evaluation. In this work, we present both the\nmethods developed for machine learning-based high-throughput screening and\nresults from using our computational pipeline to find SARS-CoV-2 inhibitors.\n