Artificial intelligence for antiviral drug discovery in low resourced settings: A perspective

Abstract

Current antiviral drug discovery efforts face many challenges, including development of new drugs during an outbreak and coping with drug resistance due to rapidly accumulating viral mutations. Emerging artificial intelligence and machine learning (AI/ML) methods can accelerate anti-infective drug discovery and have the potential to reduce overall development costs in Low and Middle-Income Countries (LMIC), which in turn may help to develop new and/or accessible therapies against communicable diseases within these countries. While the marketplace currently offers a plethora of data-driven AI/ML tools, most to date have been developed within the context of non-communicable diseases like cancer, and several barriers have limited the translation of existing tools to the discovery of drugs against infectious diseases. Here, we provide a perspective on the benefits, limitations, and pitfalls of AI/ML tools in the discovery of novel therapeutics with a focus on antivirals. We also discuss available and emerging data sharing models including intellectual property-preserving AI/ML. In addition, we review available data sources and platforms and provide examples for low-cost and accessible screening methods and other virus-based bioassays suitable for implementation of AI/ML-based programs in LMICs. Finally, we introduce an emerging AI/ML-based Center in Cameroon (Central Africa) which is currently developing methods and tools to promote local, independent drug discovery and represents a model that could be replicated among LMIC globally.

Small Molecule Drugs Targeting Viral Polymerases

“…Consequently, this presents a significant threat that requires extensive research efforts to investigate diverse strategies aimed at designing and developing next-generation small molecule inhibitors with heightened potency against drug-resistant viral strains. This involves, but is not limited to, the use of high-throughput screening technologies and advanced computational modelling/artificial intelligence methods to identify novel chemical scaffolds and refining existing inhibitors for improved resistance profiles [118,[179][180][181][182]. Strategies may include, for example, targeting conserved regions of the enzyme that are less prone to mutation or designing molecules with increased structural flexibility (as demonstrated by second-generation HIV NNRTIs compared to their first-generation counterparts), the latter being able to dynamically adjust their binding modes in response to mutations, thereby preserving their efficacy against viral strains that have acquired resistance.…”

First fully-automated AI/ML virtual screening cascade implemented at a drug discovery centre in Africa

“…We hope that the work presented here serves as a proof-of-concept for the potential of AI/ML tools to support drug discovery efforts in LMICs. Incipient centres in West Africa (Amewu et al 2022) and Central Africa (Namba-Nzanguim et al 2022) may benefit from similar implementations. More globally, ZairaChem offers a competitive, free and constantly-updated software solution to model small-molecule bioactivity data, where no strong data science skills are required to run the tool.…”

5-chloro-3-(2-(2,4-dinitrophenyl) hydrazono)indolin-2-one: synthesis, characterization, biochemical and computational screening against SARS-CoV-2

“…Ligand-based methods often do not require a knowledge of the structural features of a receptor site (Ferreira et al 2015 ; Vazquez et al 2020 ). These include quantitative structure–activity relationships (QSAR), ligand-based pharmacophore querying methods, and most recently, artificial intelligence/machine learning (AI/ML) models (Selvaraj et al 2022 ; Subramanian et al 2022 ; Namba-Nzanguim et al 2022 ; Turon et al 2023 ). Among these methods, the most cite is molecular docking and scoring techniques have been proven to be efficient ways of identifying active compounds from an electronic library of compounds by a what is commonly called virtual screening (Morris & Lim-Wilby 2008 ; Chen 2015 ; Lohning et al 2017 ).…”