Machine Learning Optimization Approach to Design Multi-Epitope Marburg Vaccine Construct

Shreyansh Suyash¹, Wajihul Hasan Khan², Priyasha Maitra¹, Vinod Jangid¹, Parveen Punia³ and Avinash Mishra^1*

¹Growdea Technologies Pvt. Ltd., Gurugram., Haryana India.

²All India Institute of Medical Sciences, New Delhi India

³Pt. Neki Ram Sharma Government College, Rohtak Haryana India.

Corresponding Author E-mail:avish2k@gmail.com

ABSTRACT: The Marburg virus (MARV) causes severe hemorrhagic fevers with life-threatening symptoms. A study aimed to design a multi-epitope vaccine (MEV) using immunoinformatic approaches for treatment for MARV infection. A comprehensive screening procedure was used to identify immunogenic protein sequences within seven crucial proteins from MARV that could trigger T-cell and B-cell responses. A computational analysis of these epitopes showed a non-allergenic nature and significant antigenicity, validating the structural parameters. The final construct of virus-like particle (VLP) was used for mutation using machine-learning model. A machine learning model, DeepPurpose framework was developed and trained to screen out the best vaccine construct/VLP sequence among all the generated sequences. Best variant VLP had the predicted IC50 of 0.021 nM with the receptor TLR4. Model structures of the native and mutant VLP with prediction confidence scores of 96.2% and 88.5% were selected for molecular docking and molecular dynamic simulation to assess stability. RMSD of native construct ranged from 1.75 to 2 nm, while variant had 1.5 to 1.75 nm which was lower than the wild type, suggesting more stable conformation. The VLPs when bound with the toll-like receptor-4 (TLR4), plays a role in innate immunity. Designed VLP-TLR4 complex showed high stability post MD simulation of 500 ns and had strong average binding free energy (ΔG) of -520.13 (kcal/mol). The vaccine's stability helps it trigger a tailored immune response, making it an attractive candidate for viral neutralization strategies. The study showed a computational pipeline for designing and validating MARV multi-epitope vaccines using physics and machine learning. Additionally, the variant VLP exhibited favourable properties, suggesting its potential suitability for experimental validation, which could provide valuable insights. Nonetheless, the present study relies on in silico methodologies instead of in vivo or in vitro investigations, which is a limitation. This approach has promising applicability in the design of novel peptide vaccines against the MARV.

KEYWORDS: Immunoinformatic; Marburg virus; Molecular Docking; Molecular Dynamics Simulation; Vaccine Design; Virus-like particle (VLP)

[ HTML Full Text]