Deciphering the Code of Viral-Host Adaptation Through Maximum-Entropy Nucleotide Bias Models.

Area of Science:

• Virology
• Bioinformatics
• Computational Biology

Background:

• Viral evolution is significantly influenced by host interactions.
• Understanding host-virus dependencies is crucial for predicting viral spread and adaptation.
• Existing methods for viral classification and host prediction can be computationally intensive.

Purpose of the Study:

• To introduce and validate Maximum Entropy Nucleotide Bias (MENB) models for quantitatively characterizing host-virus evolutionary dependence.
• To develop a rapid, interpretable, and robust method for classifying viral families and hosts.
• To explore the potential of MENB models in tracking viral adaptation and designing novel viral sequences.

Main Methods:

• MENB models were developed using single, di-, and tri-nucleotide usage from viral genomic sequences.
• The models were trained to classify viral families (four ssRNA families) and hosts (three hosts).
• Performance was benchmarked against deep neural network methods.

Main Results:

• MENB models successfully decode host and viral family "fingerprints" in nucleotide motif usage.
• The approach accurately predicts viral family and host from genomic sequences, outperforming deep learning methods in speed and interpretability.
• MENB demonstrated strong generalization to new viral families and host taxa, identified intermediate hosts for Influenza A and Human Coronavirus, and detected genomic recombination events.

Conclusions:

• MENB models provide a powerful and efficient tool for understanding host-virus co-evolution.
• The models offer insights into selective pressures driving viral adaptation and can guide the design of viral sequences.
• MENB represents a significant advancement in viral genomics analysis, offering both predictive and explanatory capabilities.