PatchDNA: A Flexible and Biologically-Informed Alternative to Tokenization for DNA

Alice Del Vecchio, Chantriolnt-Andreas Kapourani, Abdullah M Athar, Agnieszka Dobrowolska, Andrew Anighoro, Benjamin Tenmann, Lindsay Edwards, Cristian Regep

Keywords: DNA, DNA language model, gLM, tokenization, genomic sequence representation

TL;DR: Evolutionary conservation–guided “patch” boundaries focus model capacity on the most functionally important regions, yielding smaller models that nonetheless outperform current state-of-the-art benchmarks and, uniquely, permit on-the-fly re-patching

Submission Number: 7