Evogene Completes First-In-Class Foundation Model for Generative Molecule Design with Google Cloud.

Evogene has completed version 1.0 of its AI foundation model for small molecule design, developed with Google Cloud, with 90% precision. The model addresses multiple complex product criteria simultaneously and is expected to enhance Evogene's ChemPass AI capabilities. Version 2.0 is in development, focusing on multi-parameter optimization.

Evogene Ltd. has completed the development of its AI foundation model for small-molecule design, version 1.0, in collaboration with Google Cloud. The model, with a reported 90% precision, addresses multiple complex product criteria simultaneously, enhancing Evogene's ChemPass AI capabilities. This development marks a significant advancement in drug discovery and crop protection, aiming to accelerate R&D by enabling the simultaneous optimization of properties like efficacy, toxicity, and stability.

The AI model, known as ChemPass-GPT, is built on a transformer neural network architecture similar to the GPT models that revolutionized natural language processing. It was trained on an enormous chemical dataset, including roughly 40 billion molecular structures, to learn the "language" of molecules. This extensive training regimen allows the model to generate novel SMILES strings that correspond to chemically valid, drug-like structures [1].

One of the standout features of the ChemPass-GPT model is its ability for multi-objective optimization. Unlike traditional drug discovery methods, which often optimize one property at a time, this AI model can handle multiple objectives simultaneously. It uses advanced machine learning techniques, including multi-task learning and reinforcement learning, to guide the generative model toward satisfying multiple constraints. This capability allows for the design of molecules that meet various criteria, such as potency, lack of toxicity, and good bioavailability, all at once [1].

Version 2.0 of the ChemPass AI is currently in development, focusing on further enhancing multi-parameter optimization. This next-gen model will allow for even more flexible tuning of multiple parameters, including user-defined criteria tailored to specific therapeutic areas or crop requirements. This suggests that future iterations of the model could provide researchers with even greater control over the design process, potentially leading to more efficient and effective molecule generation [1].

The completion of version 1.0 of the ChemPass AI model highlights a wider shift in life-science R&D: the move from laborious trial-and-error workflows to AI-augmented creativity and precision. Unlike human chemists, who tend to stick to known chemical series and iterate slowly, an AI can fathom billions of possibilities and venture into the unexplored 99.9% of chemical space. This opens the door to finding efficacious compounds that don't resemble anything we've seen before, which is crucial for treating diseases with novel chemistry or tackling pests and pathogens that have evolved resistance to existing molecules [1].

References:
[1] https://www.unite.ai/evogene-and-google-cloud-unveil-foundation-model-for-generative-molecule-design-pioneering-a-new-era-in-life-science-ai/