We have explored billions of chemical molecules and have made advances in deep learning algorithms to build up a rational and explorable hidden space for de nove drug discovery, which ensures a traceable biology character. This model has successfully revealed the underlying relationship between molecule structure and their druggability (bioactivity). For each individual task, the hidden space will help us to identify and mine in a clear direction. With the help of Electrostatic Similarity, Molecular dynamic simulations and Quantum chemistry, we can find better scaffolds more efficiently.
Autoregressive, Generative, Adversarial Network, Reinforcement Learning
Big Data (Molecules, ADMET, Receptors)
Rebuilded chemical space
Boosted screening process(30-60 days)
Custom-made molecule base on receptor
Earlier ADMET prediction
Earlier synthetic accessibility prediction