GAIA introduces a novel "token-free" genomic language model based on the Byte Latent Transformer (BLT) architecture to address the global threat of antimicrobial resistance (AMR) in ESKAPE pathogens. By processing raw genomic sequences at the byte level, the system eliminates the biases of traditional k-mer tokenisation, allowing for single-nucleotide resolution in predicting resistance phenotypes and Minimum Inhibitory Concentration (MIC) values .
The core of the project is an Explainable AI (XAI) engine that identifies predictive genetic loci, cross-referencing them with global databases (CARD, POINT, ResFinder) to isolate novel resistance markers. Validated patterns are flagged for experimental confirmation by the team of biological partners at the University of Catania. The project delivers a containerised, HPC-scalable workflow for genomic discovery on EuroHPC JU infrastructure.
Principal Investigator, Company and Country
Marco Chessari, Teleconsys SpA, Italy