We develop state-of-the-art deep-learning architectures to predict important molecular properties, including on-target binding, toxicity, and ADMET for accelerated lead discovery and optimization.

We regularly enumerate and screen millions of compounds in our virtual libraries for hit identification. We combine the expertise of medicinal chemists and the predictive power of machine learning models, to triage hits that are both potent and have ideal drug-like properties.

We employ simulation techniques, including relative binding free-energy calculations, non-equilibrium targeted molecular dynamics, docking, conformational expansion, and quantum mechanical methods to understand the energetics of bound ligands to their protein targets, to design superior kinase inhibitors. We then combine these representations with statistical ML to improve performance and scalability.

We invest heavily in building rich representations of our data. We use a diverse range of featurizations for small molecules and proteins, from string-based SMILES methods to custom QM-based representations.

We focus on generating selective molecules for our intended targets, while also being synthesizable, stable, and tissue-penetrant, amongst other molecular criteria. To do so, we use computational techniques, such as genetic algorithms and continuous latent models, to efficiently traverse and perform multi-parameter optimization of molecules across vast swaths of chemical space.