PubMed · 2026-05-20
Researchers built an AI called PlantGFM that learned from nearly 11 billion letters of plant DNA across 12 species, then used it to write brand-new plant genes from scratch — and those invented genes actually worked inside a living plant, producing proteins for the first time.
PlantGFM was pre-trained on 10.84 billion nucleotides from 12 plant species and matched or exceeded specialized gene-prediction software after fine-tuning on 10 annotated plant genomes.
All 7 AI-designed gene candidates showed transcriptional activity (RNA production) when introduced into Nicotiana benthamiana plants.
2 of the 7 AI-invented genes achieved stable protein expression — the first demonstration of full DNA-to-RNA-to-protein activity from large-language-model-generated sequences in any plant.