Recent advances in genomic foundation models have opened new opportunities for representing bacterial genomes through contextualized protein embeddings. Building upon Bacformer [1], a transformer-based genomic language model, this project investigates its application to pangenome analysis.
The objective of this work is to adapt and evaluate Bacformer for the analysis of bacterial pangenomic data. The study was conducted in two stages: first on a single bacterial strain to validate the adapted pipeline, and subsequently on a complete pangenome to assess scalability and identify challenges associated with larger datasets.
Dataset: Acinetobacter baumannii AYE (~3,770 genes, 3,639 gene families)
Objective: Validate the adapted pipeline and ensure correct protein embedding generation.
Dataset: Acinetobacter baylyi pangenome (21 genomes from PanGBank)
Objective: Extend the approach to pangenome-scale data and explore challenges related to scaling the model to larger genomic datasets.
[1] Wiatrak M et al. bioRxiv. 2025. doi:10.1101/2025.07.20.665723.