Innovative AI Framework Predicts Viral Evolutionary Fitness of SARS-CoV-2 Variants

Viral mutations and rapid evolution present ongoing challenges in managing infectious diseases such as COVID-19. During the pandemic, new SARS-CoV-2 variants emerged with mutations that increased transmissibility, complicating containment efforts. Understanding a virus's "fitness"—its ability to spread within populations—is crucial for effective surveillance and response strategies.

Researchers from The University of Tokyo, led by Associate Professor Jumpei Ito and including Dr. Adam Strange and Professor Kei Sato, have developed a groundbreaking AI-based framework called CoVFit. This tool predicts the fitness of SARS-CoV-2 variants using only their spike protein sequences, integrating molecular data with large-scale epidemiological observations. The model aims to explain why certain variants succeed and spread widely while others decline.

CoVFit analyzes mutations, especially in the spike protein, which play a key role in immune escape and infectivity. By combining mutation patterns with epidemiological trends, the system predicts a variant’s potential to expand. Developed through innovative processes that consider both molecular biology and population dynamics, CoVFit can generate a fitness score for any given variant.

The team trained and tested CoVFit using extensive datasets, enabling high-accuracy predictions of how single amino acid substitutions affect viral fitness. Notably, when applied to the Omicron BA.2.86 lineage, CoVFit accurately forecasted mutations at specific spike protein positions that later appeared in globally spreading descendant lineages. This demonstrates its capacity to anticipate future viral evolution.

Dr. Ito emphasizes that CoVFit can identify high-risk variants early—sometimes even when only a single sequence is available—thus providing a valuable tool for proactive public health measures. The framework also systematically generates hypothetical mutants to identify mutations likely to emerge in future strains.

Overall, CoVFit marks a significant advance in viral surveillance, combining molecular insights with epidemiological data via AI. It offers a flexible and timely approach to understanding and predicting viral evolution, which is critical for pandemic preparedness and response worldwide.