The Role of AI in Predicting Viral Evolution
Advancements in artificial intelligence (AI) are enabling scientists to predict how viruses, such as SARS-CoV-2 and influenza, might evolve by analyzing their genetic sequences. While predicting viral evolution in its entirety remains a challenge, AI tools are making significant strides in this field, offering hope for better pandemic preparedness through enhanced vaccine and antiviral development.
RNA viruses, including SARS-CoV-2, evolve rapidly by accumulating mutations. Some of these mutations allow viruses to evade host immunity and spread more effectively. By anticipating these changes, researchers can potentially design vaccines and antiviral treatments before new variants emerge.
Current Capabilities and Limitations of AI Tools
AI models currently excel at predicting which single mutations are likely to be most advantageous for a virus in the short term. However, forecasting the interplay of multiple mutations or long-term evolutionary pathways remains an elusive goal.
“This is an exciting and highly valuable area of research,” says Brian Hie, a computational biologist at Stanford University. Hie was among the pioneers in applying large language models to study viral mutations. Despite these advancements, predicting viral evolution involves complex challenges.
Traditional laboratory experiments to identify viral variants with enhanced properties have been labor-intensive and time-consuming. For example, Yunlong Cao’s team at Peking University has conducted detailed experiments to understand how individual mutations enable viruses to escape detection by antibodies. These experiments provide valuable insights but cannot fully explain the dynamics of viral evolution.
The Impact of Protein-Structure Prediction Tools
The introduction of AI-driven protein-structure prediction tools, such as DeepMind’s AlphaFold and Meta’s ESM-2 and ESMFold, has injected fresh momentum into the field. According to David Robertson, a virologist at the University of Glasgow, these tools have expanded the possibilities for predicting viral behavior.
AI models rely on vast datasets for training. The global sequencing efforts during the COVID-19 pandemic have resulted in nearly 17 million SARS-CoV-2 sequences, providing a rich resource for model development.
Key Models Advancing the Field
One prominent model, EVEscape, developed by Debora Marks and her team at Harvard Medical School, has engineered 83 potential versions of the SARS-CoV-2 spike protein. These “spike avatars” evade antibodies from individuals vaccinated against or infected with current variants. This innovation could aid in evaluating the effectiveness of future COVID-19 vaccines.
Similarly, Jumpei Ito’s team at the University of Tokyo developed CoVFit, an AI model that predicts the relative fitness of SARS-CoV-2 variants. Trained on over 13,000 spike-protein variants, CoVFit integrates experimental data on immune evasion. The model successfully forecasted the rise of variants, including XBB5 and JN.1, long before they dominated global transmission.
In March 2024, Ito’s group identified three key amino acid changes that could enhance the fitness of the JN.1 variant. These mutations subsequently appeared in rapidly spreading variants, showcasing the model’s predictive power.
Future Potential and Challenges
AI’s ability to predict viral evolution marks a paradigm shift in public health strategies. By leveraging such tools, scientists could stay ahead of viral evolution, mitigating the impact of future pandemics. However, challenges remain, such as improving the prediction of complex mutation combinations and addressing biases in training data.
Beyond SARS-CoV-2, these AI advancements have broader implications for other pathogens, including influenza and emerging viruses. Integrating real-time data with AI models could further enhance their accuracy and utility.
Conclusion
The synergy between AI and virology has opened new avenues for understanding and countering viral evolution. While the ultimate goal of predicting all future mutations remains distant, the progress made so far highlights AI’s transformative potential in global health. Continued investment in AI research and collaboration between computational biologists, virologists, and immunologists will be key to realizing this vision.
