Soilytix releases genomic AI models trained on environmental DNA

Soilytix released its LOAM genomic AI models and an accompanying preprint, reporting that LOAM-624M ranked first among evaluated models on an enzyme-function benchmark, including models roughly ten times larger. The benchmark used experimentally annotated genes across 128 functional classes. The models were trained on 15,640 microbial genomes reconstructed from long-read sequencing of soil, sediment and water, representing approximately 67.5 billion DNA bases. Soilytix combines the models with agricultural field evidence and laboratory sequencing to prioritize candidate genes and proteins for crop-protection testing. The company is opening discovery pilots for 2027. The reported results concern model evaluation and candidate selection, rather than demonstrated crop-protection products.