Fergus Imrie

A public crystallographic structure and affinity dataset for enteroviral 2A protease (925 binding events, 699 compounds), with docking, cofolding and screening baselines. Fragment fine-tuning of OpenFold3-p2 improves follow-on predictions.

This paper improves diffusion-based molecular docking by using reinforcement learning to enforce physical validity, enhancing pose accuracy without increasing computational cost.

MolSnapper improves structure-based drug design by conditioning diffusion models with 3D pharmacophores, generating more valid molecules with higher structural and chemical similarity.

This paper evaluates RNN-based de novo molecular generation under limited training data and proposes a curriculum learning approach to improve diversity and control, while highlighting limitations of SMILES representations.

STRIFE improves fragment elaboration by incorporating pharmacophoric information from protein targets, outperforming existing methods in generating ligand-efficient structures.

This paper introduces DeepCoy, a deep learning method to generate decoy molecules that match active molecules' properties, improving virtual screening performance by reducing bias and enhancing benchmarking.

This paper introduces DEVELOP, a generative model that incorporates 3D pharmacophoric constraints to improve molecule design quality and control.

This paper introduces DeLinker, a 3D graph-based generative model for compound design that outperforms baselines by incorporating structural information.