BOTANIC-1: a series of long-context plant genomic foundation models in the agentic era

Botanic-1, a series of long-context plant genomic foundation models, marks a significant advancement in agentic plant research. These models aim to accelerate the development of climate-resilient crops by directly reasoning over plant genomic sequences and pinpointing trait-associated regions or loci. Current challenges include anticipating the impact of DNA base changes (variants) and understanding regulatory mechanisms, which remain active areas of research.

Self-supervised training on unannotated genomic data allows genomic language models (gLMs) like Botanic-1 to learn DNA syntax and grammar that surpasses current annotations. This capability complements standard bioinformatics analyses, which rely on rules established by decades of genomics research. Botanic-1 is designed to operate on sequences ranging from hundreds of base pairs to 128 kilobases.

The Model Factory, which powers Botanic-1, outperforms all generalist and plant-specific gLMs, as well as specialized baselines, on one of the largest sets of plant genomics evaluation tasks reported. Remarkably, these models achieve this superior performance at a much smaller computational budget than concurrent models. Mechanistic interpretability analysis identifies features associated with biologically meaningful sequence properties, such as coding region boundaries and splice site motifs.

This demonstrates that Botanic-1 is not just a benchmark performer but also a source of valuable biological insight.

To fully harness the potential of these models, Botanic-1 is integrated as a specialized genomic layer callable by a generalist large language model (LLM) agent. This hybrid system could significantly accelerate plant biology research. The Botanic1 models, available as Botanic1-S, Botanic1-M, and Botanic1-L, are released for research use at https://huggingface.co/collections/living-models/botanic1-6a97f4e3c33f3d109a75057d.