Can Plant Genomic Foundation Models Write Entirely New DNA?

Imagine trying to write a bestselling novel using only sentences you have cut out of old magazines. For decades, crop breeders have done something similar, tweaking existing plant DNA to make crops hardier or more productive.

But natural breeding limits us to what already exists in nature. To fight climate change, we need to design entirely new biological instructions. This is where plant genomic foundation models enter the field.

How Plant Genomic Foundation Models Write DNA

Researchers built PlantGFM, an AI trained on 10.84 billion letters of genetic code from 12 different plant species. Unlike tools that only edit existing genes, this AI can read long stretches of DNA and write entirely new, functional genes from scratch.

The team tested seven AI-designed genes in tobacco plants (Nicotiana benthamiana). The study measured that:

• All seven synthetic genes successfully transcribed into RNA.
• Two candidates achieved stable protein expression.
• The model matched or exceeded specialised gene prediction tools.

This represents the first demonstration of DNA-to-RNA-to-protein expression of AI-generated sequences in plants.

The Future of Synthetic Crops

This success suggests that AI can generate viable, synthetic plant genes that actually work inside living organisms. This could allow researchers to organise plant defences against pests without relying on existing natural traits.

In the future, this approach could help scientists design bespoke crops tailored to survive harsh, changing climates. Instead of just editing the genetic dictionary, we are beginning to write entirely new biological chapters.