Genesis Therapeutics

Our Platform

GEMS: Genesis Exploration of Molecular Space

Genesis has created the industry's most advanced molecular AI platform – GEMS – which integrates deep learning and molecular simulations for property prediction, and language models for molecular generation.

GEMS allows Genesis to create first- and best-in-class small molecule drugs with extremely high potency and selectivity, to address challenging and previously undruggable targets.

Our technology

Machine Learning

Genesis has developed Dynamic PotentialNet as a proprietary approach to predict potency, selectivity, and ADMET. This technology provides GEMS with unprecedented speed and accuracy in property predictions, including for data-poor and previously undruggable targets.

Dynamic PotentialNet represents the 3D structure and interactions of protein-ligand complexes as spatial graphs. By training on a combination of molecular simulations and experimental data, it learns representations of the physical interactions underlying binding affinity. This enables the model to generalize to novel protein targets and make accurate predictions based on 3D structures alone, without requiring on-target binding affinity data.

Molecular Simulation

Dynamic PotentialNet also integrates our own proprietary molecular simulation platform. This allows GEMS to learn 3D binding dynamics, including effects of protein flexibility and critical water molecules. Integrating deep learning with molecular simulation has been crucial for advancing our drug programs against challenging and previously undruggable targets.

Molecular Generation

In addition to our predictive models, we have developed a massive molecular generation engine. GEMS uses chemically aware ML models that can be directed to create novel molecules and explore specific regions of chemical space for each drug program.

During Hit ID, GEMS generates billions of drug-like and synthetically accessible molecules, enabling a diverse exploration of chemical space. During Hit-to-Lead and Lead Optimization, our chemists direct GEMS to generate novel molecules from a specific starting point, to explore scaffold hops and optimize promising series.

Publications

Our academia-leading research was the initial component of our expanding portfolio of machine learning and molecular simulation technologies

1

Our AI platform expands on the initial discoveries in PotentialNet — field-leading, peer-reviewed methods for molecular property prediction that co-founder Evan Feinberg invented in Stanford's acclaimed Pande lab.

Key components of our platform have been rigorously tested against current state-of-the-art methods in a collaboration between Stanford and a top-five pharma company. We achieved a step-change improvement predicting 20+ different ADMET properties.