PROPHET: Phylogenetically Robust Antiviral Peptide Design Against Heterogeneous Evolutionary Trajectories

Kimberly Liang, Navya Nori, Elizabeth H Mahood, Yinuo Zhang, Pranam Chatterjee

Keywords: phylogenetic robustness, antiviral design, evolutionary escape, discrete flow matching, therapeutics

TL;DR: PROPHET designs antiviral peptides that remain effective against plausible future viral escape variants inferred from phylogenetic evolution.

Abstract: Antiviral therapeutics often lose efficacy as viral proteins accumulate escape mutations, yet existing peptide binder design methods optimize against a single observed target sequence and do not account for future evolutionary variation. We introduce a method for **P**hylogenetically **R**obust antiviral **P**eptide design against **H**eterogeneous **E**volutionary **T**rajectories, termed **PROPHET**, a framework for evolutionary robustness-aware peptide generation. PROPHET estimates a phylogenetic energy landscape from FastTree phylogenies using site-specific mutation rates, substitution preferences, and direct coupling analysis of coevolving residues, defining a structured distribution over evolutionarily plausible future variants. We then leverage multi-objective discrete flow matching to guide against this distribution using a CVaR robustness objective computed over Gibbs-sampled variants from the phylogenetic energy landscape. The resulting peptides jointly optimize wild-type affinity *and* robustness across plausible escape trajectories implied by observed viral evolution. On HIV-1 protease, PROPHET achieves substantially stronger mean and worst-case binding across held-out escape variants than methods optimized against the wild-type alone, uniformly weighted observed sequences, or randomly generated variants, while preserving high affinity to the current target. Overall, our results demonstrate that the phylogenetic history of a virus can directly guide the generative design of antiviral peptide therapeutics robust to future evolutionary escape.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 113