Drug Resistance Predictions Based on a Directed Flag Transformer

Area of Science:

• Virology
• Drug Discovery
• Computational Biology

Background:

• The SARS-CoV-2 virus continues to evolve, posing a significant public health threat.
• Emerging resistance to antiviral drugs like PAXLOVID (nirmatrelvir) is a growing concern.
• The viral main protease (Mpro) is a key target for nirmatrelvir, and mutations can confer resistance.

Purpose of the Study:

• To develop a computational framework for predicting the impact of SARS-CoV-2 Mpro mutations on nirmatrelvir binding.
• To identify and assess drug-resistant SARS-CoV-2 variants.
• To establish a system for real-time viral surveillance and guide the development of next-generation therapeutics.

Main Methods:

• Development of CAPTURE (direCted flAg laPlacian Transformer for drUg Resistance prEdictions), a tool integrating mutation analysis and resistance prediction.
• Utilizing DFFormer-seq, an ensemble model combining a Directed Flag Transformer and sequence embeddings for resistance prediction.
• Analysis of Mpro mutations from May to December 2022 to track resistance evolution.

Main Results:

• Observed increasing frequencies of Mpro mutations near the nirmatrelvir binding site, suggesting accelerated evolution due to PAXLOVID use.
• Identified several potential drug-resistant mutations, including experimentally validated H172Y and F140L, plus five others awaiting confirmation.
• CAPTURE demonstrated 57% recall and 71% precision in predicting drug-resistant mutations on Mpro mutant data.

Conclusions:

• The study presents CAPTURE as a robust framework for predicting antiviral drug resistance in SARS-CoV-2.
• Findings highlight the need for continuous viral surveillance to monitor the emergence of drug-resistant strains.
• The developed methodology can inform the design of future antiviral therapies effective against evolving viral variants.