To find the best way to describe a molecule has always been an unsolved challenge, especially the description required to link with a potential to predict its biological characteristics. The molecule structure is a high dimension graph with underlying biology information. Our AI approach can capture that information for a better prediction and chemical space encoding.
The group theory, the matrix theory, the topology, and other mathematical tools shaped our self-developed molecule representation system along with an advanced deep learning algorithm trained by billions of molecule data. The representation keeps critical information and mapping the molecule to a hyper dimension manifold. During the stage of early drug discovery, the hits to lead(H2L) stage determined future diversity. Our strategy will ensure a high receptor/ligand binding affinity while aiming for the most diverse ligand options. In addition, with precisely ADMET prediction, better PCC(preclinical candidate compounds ) are expected.
Break the traditional framework and cognition by our molecule map (a hidden hyper dimension space), which exceeds the known knowledge with a continuous and smooth encoding.
The representation trained by big data, supports comprehensive predictions of ADMET properties. Our representation carrying underlying information related to but is not limited to molecule structure. We are ready for more sophisticated tasks.
Transfer learning from existed representation model supports task with small sample data. Our prediction is not limited to sample size.
With the unique reinforcement learning add-in, our generative model works under an AI supervisor with powerful strategy guidance.