Living Models pairs Gemma 4 with BOTANIC-1 to accelerate plant DNA decoding

Living Models has developed a pipeline combining Gemma 4 with BOTANIC-1, a foundation model trained on plant genomes, to accelerate crop breeding research. By pairing generalist reasoning with specialized genomic intelligence, the system compresses traditional genetic analysis that previously took years of field trials into hours of computation.

Breaking Genetic Deadlocks

Traditional methods like Bulk Segregant Analysis often leave researchers with thousands of candidate mutations that co-segregate identically, making it impossible to statistically distinguish the causal driver from passengers. In a test involving melon genetics, Gemma 4 acting as an autonomous bioinformatician using standard tools could only narrow candidates to a tie between the true mutation and an unrelated gene.

When equipped with BOTANIC-1, which evaluates variants based on deep evolutionary conservation, the system successfully identified the validated CmEIN3 mutation as the top candidate out of 2,494 options. This approach raised Recall@1 from 0.15 with expert guidance to 0.90, with no fabricated variants across 20 runs.

Benchmark Performance and Architecture

The team evaluated BOTANIC-1 against classical tools and leading biological models like PlantCAD2 and Evo2 across more than 500 published causal plant mutations. BOTANIC-1 placed the true causal mutation in the top 0.1% of candidates in 15.9% of cases, compared to 4.6% for the best non-GLM baseline. It ranked the causal driver within the top 1.0% in 48.8% of cases, versus 33.9% for classical pipelines.

The architecture relies on a separation of concerns: Gemma 4 E4B handles cognitive orchestration and tool execution, while BOTANIC-1 provides the biophysical signal needed to break linkage disequilibrium. Running Gemma 4 E4B on a single NVIDIA GPU allows agricultural labs to execute these pipelines on-premises, ensuring data sovereignty.