Uncovering Epistatic Interactions in SARS-CoV-2 Evolution Through Hidden Markov Models.

Area of Science:

• Virology and Evolutionary Biology
• Computational Biology and Bioinformatics
• Genomic Surveillance

Background:

• Epistatic interactions, where mutations collectively affect viral fitness, are crucial for understanding pathogen evolution but are often analyzed using static methods.
• Previous approaches lack the ability to capture the dynamic, temporal nature of these complex genetic relationships within evolving viral populations like SARS-CoV-2.
• Predicting pathogen evolution requires methods that can account for how mutations interact over time.

Purpose of the Study:

• To develop and validate a hidden Markov model (HMM) framework for capturing the temporal dynamics of epistatic relationships in SARS-CoV-2.
• To address the limitations of static, network-based approaches by introducing a time-aware analysis of mutation interactions.
• To provide computational tools for early detection of viral variants and understanding their evolutionary trajectories.

Main Methods:

• A hidden Markov model (HMM) was developed to model single amino acid variant pairs as a two-state system (linked/unlinked).
• Emission probabilities were derived from linkage disequilibrium theory, and transition probabilities were optimized using the Baum-Welch algorithm.
• Permutation-based validation with temporal order noise reduction was employed to ensure biological signal accuracy, applied to over 2 million SARS-CoV-2 spike protein sequences.

Main Results:

• The HMM identified three classes of epistatic dynamics: permanent (0.3%), transient (0.3%), and oscillating (0.7%) linkages.
• Analysis of the Alpha variant revealed significantly higher epistatic linkage (78%) compared to general spike protein pairs (1.3%), with many oscillating patterns.
• The framework detected known and novel epistatic interactions, identified early Alpha variant networks, and revealed distinct dynamic patterns for Delta and Omicron variants.

Conclusions:

• The temporal HMM framework effectively captures the dynamic nature of epistatic interactions in SARS-CoV-2, outperforming static methods.
• This approach provides valuable insights into variant evolution, identifying specific linkage patterns (e.g., oscillating) indicative of frequency-dependent selection.
• The computational tools enable real-time genomic surveillance, facilitating early variant detection and a deeper understanding of viral evolution.