Combining machine learning and iterative experiments to keep pace with emerging viral variants of concern

Area of Science:

• Virology
• Genomics
• Computational Biology

Background:

• Predicting viral mutations is crucial for pandemic preparedness, but current models falter with evolving strains.
• Early identification of variants of concern (VOCs) guides updates for vaccines, diagnostics, and therapeutics.

Purpose of the Study:

• To develop a scalable framework integrating machine learning and high-throughput experimentation to anticipate and evaluate emerging SARS-CoV-2 receptor-binding domain (RBD) variants.
• To improve the accuracy and generalizability of predictive models for viral mutations.

Main Methods:

• Integrated random forest and neural network models with targeted high-throughput experimentation.
• Trained predictive models on public datasets for ACE2 binding, RBD expression, and antibody escape.
• Refined models using experimental data from over 200 SARS-CoV-2 variants (wild-type and Omicron).
• Employed an indirect transfer learning approach and encoded complementarity-determining regions (CDRs) for model generalizability.

Main Results:

• Achieved high accuracy in predicting antibody binding, with correlation coefficients up to 0.79.
• Demonstrated model generalizability across diverse antibody types, including heavy-chain-only antibodies (HCAbs).
• The framework enables rapid assessment and prioritization of emerging viral variants.

Conclusions:

• The dynamic framework supports a proactive, data-driven response to evolving viral threats.
• Facilitates timely prioritization of therapeutic strategies against emerging variants.
• Enhances pandemic preparedness through accurate prediction of viral evolution.