Research
Northwestern × LBNL JGI
GenomeOcean MoE
Research on Mixture of Experts architectures for genomic foundation models. Accepted as an oral presentation at ISMB 2026; paper in preparation.
Context
Mixture of Experts has become a standard way to scale NLP foundation models with sub-linear compute growth — more total parameters without a proportional increase in per-token cost. This project investigates whether that scaling direction transfers to genomic foundation models. The work builds on GenomeOcean, an open-source prokaryotic genomic foundation model, and is a collaboration with Zhong Wang’s group at Lawrence Berkeley National Laboratory’s Joint Genome Institute.
Status
This work was accepted as an oral presentation at ISMB 2026 (Intelligent Systems for Molecular Biology, the ISCB’s flagship conference), held in July. A full paper is in preparation. The abstract, methodology, and results will be posted on this page once the paper is public; until then, the specifics are held back to respect academic priority.