Soilytix has released LOAM, a new family of genomic AI models trained on long-read environmental genomes, as the company looks to turn the largely unexplored biological diversity in soil and other environments into a source of new agricultural candidates.

 

The company also released a preprint evaluating LOAM’s biological performance.

 

“Billions of years of biological R&D have occurred in the soil beneath our feet. The challenge is knowing where to look and what is worth testing,” said Tim Rajakumar, Chief Scientific Officer at Soilytix. “We use field evidence, deep sequencing and models such as LOAM to narrow that search.”

 

Most genomic AI models are trained primarily on established reference collections, leaving vast amounts of environmental microbial diversity poorly characterised.

 

Soilytix trained LOAM on 15,640 microbial genomes reconstructed from long-read sequencing of soil, sediment and water, representing approximately 67.5 billion DNA bases.

 

The company evaluated the models across gene essentiality, enzyme function and genetic variant effects. On an enzyme-function benchmark covering experimentally annotated genes across 128 functional classes, LOAM-624M ranked first among all models tested — including models roughly 10 times larger.

 

Across the broader evaluation, LOAM delivered leading performance among similarly sized models while remaining competitive with substantially larger systems.

 

The findings point to a potential advantage in using long-read environmental genomes as training data: they can expose genomic AI models to biology that is poorly represented in conventional reference collections.

 

Soilytix is combining LOAM with biological signals gathered directly from agricultural fields.

 

One example is disease-suppressive soil, where a pathogen may be present but disease repeatedly fails to develop. Such environments could contain microbial communities with biological mechanisms capable of suppressing pathogens.

 

The company has assembled a permissioned dataset spanning thousands of agricultural field samples, linking microbial communities with pathogen occurrence and disease outcomes. Its laboratory can then generate deep long-read genomic data from promising environments, while LOAM and established bioinformatics approaches are used to prioritise candidate genes and proteins for experimental testing.

 

The goal is to shrink an enormous biological search space into a focused list of candidates that can be tested in the laboratory.

 

“The value for a crop-protection partner isn't access to another AI model,” said Bruno Steinkraus, Founder and CEO of Soilytix. “It is being able to start with a defined pathogen or product objective and work towards biological candidates that can actually be tested.”