STAMP: A Species- and Topic-Aware Multimodal Predictor for Cross-Species Antimicrobial Peptide Activity

Published

June 1, 2026

Author(s)

Sarala N. Padi, Kinjal Mondal, David Hoogerheide, Frank N. Heinrich, Mihaela Mihailescu, Jeffery Klauda, Antonio Cardone

Abstract

Antimicrobial resistance poses a major global health challenge, necessitating efficient strategies for discovering potent antimicrobial peptides (AMPs). While recent generative models can produce large numbers of candidate sequences, experimentally validating all generated peptides to find potential AMPs is impractical due to the high costs and time requirements of wet-lab measurements. Consequently, there is a demand for an accurate prediction of peptide efficacy, typically measured as minimum inhibitory concentration (MIC), which is difficult due to the species-specific nature of existing computational models. Here, we present STAMP (Species- and Topic-Aware Multimodal Predictor), a unified machine learning framework for cross-species prediction of antimicrobial peptide activity. STAMP integrates protein language model embeddings with species conditioning and topic-aware representations that capture sequence-level patterns, enabling generalizable prediction across multiple bacterial species within a single model. We evaluate STAMP on three benchmark datasets, including two previously published datasets and a newly curated dataset derived from DBAASP, where duplicates and inconsistencies were systematically addressed. STAMP achieves strong predictive performance across datasets, with a Pearson correlation coefficient (PCC) of 0.837 and an R² of 0.70, outperforming multiple baseline models. Importantly, we further validate our prediction model for the peptides that are experimentally tested for their antimicrobial activity against E.coli. and S.epidermidis, demonstrating real-world applicability. Furthermore, residue-level importance analyses provide insights into sequence determinants governing antimicrobial activity. Together, these results establish STAMP as a scalable framework for MIC prediction and an effective computational tool for accelerating antimicrobial peptide discovery and optimization.

Citation

Nature Communications

Pub Type

Journals

Download Paper

Local Download

Keywords

Large Language Models, Evolutionary Scale Modeling (ESM), Embeddings, MIC prediction, Bio-informatics, Drug design, Antimicrobial peptides, DBAASP

Citation

Padi, S. , Mondal, K. , Hoogerheide, D. , Heinrich, F. , Mihailescu, M. , Klauda, J. and Cardone, A. (2026), STAMP: A Species- and Topic-Aware Multimodal Predictor for Cross-Species Antimicrobial Peptide Activity, Nature Communications, [online], https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=961979 (Accessed June 6, 2026)

Additional citation formats