AI Empowers Drug Discovery

Network based

Achieve substantial performance improvement over other state-of-the-art methods for drug–target interaction prediction

NeoDTI: a new nonlinear end-to-end learning model, which integrates diverse information from heterogeneous network data and automatically learns topology-preserving representations of drugs and targets to facilitate DTI prediction

Achieve substantial performance improvement over other state-of-the-art DTI prediction methods

Robust against a wide range of hyperparameters and ready to integrate more DTI information such as binding affinity

Sequence based

Sequence Model description: a novel general and scalable computational framework that combines effective feature embedding with powerful deep learning methods to accurately predict compound-protein interactions (CPIs) at a large scale, with superior predictive performance compared to baseline methods, capable to predict structure-free CPIs

Prediction Accuracy above 90%

Applicable in Virtual High-throughput Screening for FIC and difficult-to-drug targets, 3-6 hours for 1M molecule screening for drug discovery

Wet-lab validation can be reduced to the scale of 100 and fewer, with capabilities to identify allosteric hits/leads

Interaction based

Model description: a multi-objective neural network to predict both non-covalent interactions and binding affinity for a given compound-protein pair, with interpretability of the identified local features for compound-protein pairs, and provide useful insights about the molecular mechanisms of compound-protein interactions, help advance the drug discovery process

Applicable in first-in-class and fast follow drug discovery projects

Outperform other state-of-the-art methods in predicting compound-protein binding affinities

Next-gen Model

Model description: a next-generation multi-objective neural network to predict both non-covalent interactions and binding affinity for a given compound-protein pair, leveraging SBDD, FBDD, LBDD and interpretability of the identified local features for compound-protein pairs, and provide useful insights about the molecular mechanisms of compound-protein interactions, help advance the drug discovery process

Applicable in first-in-class and fast follow drug discovery projects