Living Models paired Gemma 4 with BOTANIC-1 to decode plant DNA. BOTANIC-1 is a genomic model trained on 320 plant species. The combo compresses years of crop genetics into hours. Gemma 4 acts as the reasoning layer. BOTANIC-1 evaluates evolutionary constraints in plant genomes. The system identified a causal melon gene mutation as the top candidate. Classical tools left researchers with tied mutations. BOTANIC-1 broke the tie using evolutionary signals. The approach ranked causal mutations in the top 0.1% in 15.9% of cases. It ranked them in the top 1% in 48.8% of cases. The pipeline runs on a single NVIDIA L4 GPU. Proprietary genomes remain sovereign on-premises.
What changed
Traditional crop genetics relied on years of field trials and greenhouse screening to pinpoint causal mutations.
What it unlocks
Compressing years of crop genetics into hours of computation to identify high-confidence genetic drivers.
- 320 plant species in BOTANIC-1 training
- 47,492 to 3,061 candidate mutations narrowed
- 2,494 single nucleotide mutations scored
- 15.9% top 0.1% causal mutation placement
What you need to act on it
- Gemma 4 E4B model
- BOTANIC-1 genomic model
- NVIDIA L4 GPU or equivalent hardware
Sources