Reproduce “BMFM-DNA: A SNP-aware DNA foundation model to capture variant effects” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Hongyang Li, Sanjoy Dey, Bum Chul Kwon, Michael Danziger, Michal Rosen-Tzvi, Jianying Hu, James Kozloski, Ching-Huei Tsou and 2 more

IBM Research developed BMFM-DNA, a DNA foundation model that incorporates single nucleotide polymorphism (SNP) information during pre-training. This approach consistently outperformed reference-only models on various genomic tasks, demonstrating the value of explicitly encoding natural sequence variations for capturing biological functions.

Key insights

  • Explicitly incorporating natural sequence variations (SNPs) during the pre-training of DNA foundation models enhances their ability to capture biological functions.
  • A novel encoding scheme, mapping biallelic variants to unique single characters, efficiently integrates complex variation information into the model's tokenization.
  • Pre-training, particularly with variation awareness, is critical for learning generalizable genomic representations, significantly outperforming models without pre-training.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.