High-Throughput Virtual Screening of Small Molecule Inhibitors for SARS-CoV-2 Protein Targets with Deep Fusion Models

Structure-based Deep Fusion models were recently shown to outperform several physics- and machine learning-based protein-ligand binding affinity prediction methods. As part of a multi-institutional COVID-19 pandemic response, over 500 million small molecules were computationally screened against four protein structures from the novel coronavirus (SARS-CoV-2), which causes COVID-19. Three enhancements to Deep Fusion were made in order to evaluate more than 5 billion docked poses on SARS-CoV-2 protein targets. First, the Deep Fusion concept was refined by formulating the architecture as one, coherently backpropagated model (Coherent Fusion) to improve binding-affinity prediction accuracy. Secondly, the model was trained using a distributed, genetic hyper-parameter optimization. Finally, a scalable, high-throughput screening capability was developed to maximize the number of ligands evaluated and expedite the path to experimental evaluation. In this work, we present both the methods developed for machine learning-based high-throughput screening and results from using our computational pipeline to find SARS-CoV-2 inhibitors.

1 Introduction

The COVID-19 disease caused by the severe acute respiratory syndrome coronavirus (SARS-CoV-2) is responsible for the most recent, severe pandemic in modern human history [33]. At the onset of the COVID-19 pandemic, a worldwide effort began to identify and provide target proteins for vaccine and drug development to neutralize the virus. Two distinct proteins were rapidly solved; the trimeric spike protein (spike), which binds to human ACE2 to enter human cells [3] and the main protease (M^pro), which plays a pivotal role in viral gene expression and replication [59]. In response to the pandemic, we participated in a large-scale multi-institutional effort to virtually screen, experimentally test, and optimize therapeutic leads targeting the spike and M^pro SARS-CoV-2 protein targets. Two different binding sites from the spike protein (denoted spike1, spike2) and two different conformations of the M^pro active site (denoted protease1, protease2) were used in the high-throughput screening calculations.