Researchers from Helmholtz Munich, the University of Warsaw and the University of Pennsylvania invite AI developers to participate in the AMP Challenge, part of the official NeurIPS 2026 Competition Track. The initiative aims to establish a unified benchmark for AI models that generate antimicrobial peptides and combines computational evaluation with experimental validation. Submissions are open until October 1, 2026.
Artificial intelligence is increasingly being used to design antimicrobial peptides (AMPs), which could contribute to the development of new strategies against bacterial infections. However, the wide range of evaluation methods currently used for generative AI models makes it difficult to compare their performance directly.
The AMP Challenge, organized by collaborators from Helmholtz Munich, the University of Warsaw and the University of Pennsylvania, aims to address this gap. As part of the official NeurIPS 2026 Competition Track, the challenge seeks to establish a comprehensive and standardized benchmark across a broad range of AI models for antimicrobial peptide generation.
Participants are invited to submit antimicrobial peptide sequences generated by their models. The submissions will first be assessed using a carefully designed set of computational metrics covering different aspects of peptide quality.
Based on this evaluation, the top 20 teams will advance to an experimental validation round. Each selected team will nominate 100 of its most promising candidates. From these, 25 sequences per team will be randomly selected for synthesis and experimental testing against 20 bacterial strains under a standardized protocol. If fewer than 20 teams participate, a total of 500 peptides will be selected and distributed proportionally among the participating teams.
A central feature of the challenge is the combination of computational benchmarking and experimental validation. The project is also expected to generate an open-source dataset including minimum inhibitory concentration (MIC) and HC50 measurements, providing a valuable resource for the scientific community.
The organizers further plan to publish a comprehensive benchmarking study comparing the performance of a wide range of generative models. Participants can contribute to this publication as co-authors if they also provide a repository containing their model and its weights.
To register for the challenge, teams are required to submit 50,000 sequences generated by their model. Providing the model repository and weights is only required for participants who wish to be considered for co-authorship of the resulting publication.
The submission deadline is October 1, 2026.
Further information and registration are available at:
https://szczurek-lab.github.io/amp-challenge-website
Questions can be submitted directly to the organizers via:
https://github.com/szczurek-lab/amp-challenge-2027/issues