Research / Projects
Foundation models · Microbiome
DNABERT-MB
A microbiome-pretrained DNABERT-2 that predicts antimicrobial resistance, combining learned sequence embeddings with DBGWAS-derived genomic features.
What it does
Antimicrobial resistance is a growing threat, and predicting it from bacterial genomes can guide treatment and surveillance. DNABERT-MB adapts DNABERT-2 to microbial genomes and pairs its embeddings with features from DBGWAS, a genome-wide association method for bacteria.
Result
The approach placed second in the CAMDA 2025 challenge at ISMB/ECCB 2025.
Papers
Team
Jack Vaska (lead), Pratik Dutta, Max Chao, Rekha Sathian, Zhihan Zhou, Han Liu, Ramana V. Davuluri