Duanhua Cao

OmniSyn generates and optimizes target-specific molecules with synthesis traces using a sequence-conditioned LLM, achieving state-of-the-art performance on drug discovery benchmarks.

FlexiTAC generates controllable PROTAC linkers using a Bayesian flow network, improving validity and pose quality over existing methods.

This paper introduces MolGenBench, a benchmark for evaluating molecular generative models in drug discovery, highlighting gaps between current models and real-world pharmaceutical workflows.

PBCNet2.0 predicts protein-ligand binding affinity using a cartesian tensor-based Siamese Neural Network, achieving high accuracy and efficiency compared to physics-based simulations.

The paper introduces ConfSeq, a conformation description language for 3D molecular structures, enabling sequence-based modeling and achieving state-of-the-art results in 3D molecular tasks.

SurfDock improves protein-ligand complex prediction by integrating multiple protein representations into a diffusion model, achieving higher accuracy and physical plausibility than existing methods.

PackDock predicts accurate protein-ligand complex conformations using a diffusion model for side-chain packing, validated through multiple docking experiments.

EquiScore is a protein-ligand interaction scoring method that outperforms existing methods on external test sets and shows potential for lead compound optimization.