Back

Agent-Guided Ranking Policy Improvement for Peptide Drug Candidate Prioritization

Wijaya, E.

2026-04-22 bioinformatics
10.64898/2026.04.19.719536 bioRxiv
Show abstract

Peptide drug programs live or die on triage: picking the handful of candidates worth expensive wet-lab validation from thousands of in silico hits, under competing activity, toxicity, stability, and developability constraints. We asked whether an automated policy-search agent, given a frozen evaluation harness and a scored candidate pool, could learn a better ranking policy than the weighted-sum score a human team would write by hand. Across a public benchmark of 3,554 antimicrobial peptides scored on all four endpoints, the agent-derived policy captures 65% of the best candidates in its top-20 shortlist, compared to 44% for NSGA-II and 61% for both equal-weight scalarization and best-of-1,000 random weight search (Wilcoxon signed-rank p = 0.004 across 10 independent data splits). We report on a public antimicrobial benchmark, not a clinical claim. The code, oracles, and ranking policy are released as a drop-in triage layer: the candidate pool can be swapped with an internal peptide library and in-house assays.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.