Evogene launches generative AI foundation model for molecule design

Computational biology firm Evogene has announced the completion of a generative AI foundation model designed to transform small molecule discovery for the pharmaceutical and agriculture sectors. Developed in collaboration with Google Cloud, the new model expands Evogene’s ChemPass AI platform and aims to simultaneously address multiple complex product criteria—an approach the company claims sharply improves the efficiency and precision of molecule design.

The foundation model is built on a dataset of approximately 38 billion molecular structures and trained using Google Cloud’s high-performance computing infrastructure. It delivers around 90% design precision for novel molecules—far surpassing the approximate 29% accuracy seen in traditional GPT-based models, according to Evogene’s internal assessments. The company says this capability will enable its partners to develop more potent, patentable compounds while reducing development time and failure rates.

Evogene positions the model as a response to long-standing inefficiencies in traditional drug and ag-chemical discovery pipelines, which typically assess product requirements in sequence and tend to stay within well-trodden chemical spaces. By enabling the simultaneous optimization of parameters such as efficacy, safety, and patentability, the AI-driven approach offers a more expansive and commercially viable search for new compounds.

Ofer Haviv, CEO of Evogene, called the foundation model a “major milestone,” emphasizing its ability to generate proprietary molecules and unlock new intellectual property frontiers. “This is key to overcoming long-standing challenges in life-science R&D,” he said, citing both pharmaceutical and sustainable agriculture applications.

Google Cloud Israel’s Managing Director, Boaz Maoz, noted the strategic significance of the collaboration. “Evogene’s progress with ChemPass AI highlights the strength of pairing advanced AI infrastructure with deep scientific insight,” he said.

Development is already underway for version 2.0 of the model, which will offer enhanced capabilities for multi-parameter optimization in both therapeutic and agricultural use cases. The update will incorporate predefined, domain-specific constraints to generate more targeted and commercially viable outputs.

Evogene continues to seek partnerships with companies aiming to integrate ChemPass AI into early-stage discovery pipelines across pharma and agriculture. The platform is part of Evogene’s broader suite of tech-engines, which also includes MicroBoost AI for microbiome-based therapeutics and GeneRator AI for genomic applications.