Deep Learning-Assisted Design of de novo Protein Binders Targeting Hepatitis C Virus E2 Protein

Abstract

Hepatitis C virus is a grievous disease with an increased mortality rate worldwide. Chemical-based medications possess deleterious side effects and are considered inefficient in combating viral infections. Advanced therapeutic strategies are being examined with increased specificity against viral proteins such as designing highly regular proteins facilitating the development of highly effective inhibitors. Here, we present for the first time, the use of deep learning-based large language protein model ProtGPT2 with a unique strategy to design novel therapeutic binders that have the potential to mimic host receptors and inhibit the viral protein, especially the HCV envelope glycoprotein E2 for clinically relevant genotype 1a and 1b. We generated five de novo proteins for each host receptor that mimic the human receptors, based on the interacting residues which were identified by the tools of the PDBSum database in the docked host-E2 complexes generated with the ClusPro web server. The Root Mean Square Deviation score revealed that each de novo designed binder exhibited high similarity with the human receptors indicating a successful generation. Furthermore, multiple interactions were observed between these de novo designed proteins and E2 protein, emphasizing the potential of these de novo designed proteins as significant inhibitors. A comparative analysis of molecular docking between human interacting partners and de novo designed proteins revealed that de novo proteins, such as CD81-D1 and CLDN-D4, are the most effective inhibitors having the lowest binding energy when interacting with the most conserved regions of the E2 protein. These generated proteins may inhibit the interaction of E2 with CD81 and CLDN host receptors.