From Holo Pockets to Electron Density: GPT-style Drug Design with Density

2026-05-09

AI Summary

Traditional structure-based drug design often relies on empty binding pockets, overlooking the true physicochemical environment shaped by ligands and solvent. This work proposes EDMolGPT—a molecular generation framework based on an autoregressive decoder—that uniquely leverages low-resolution electron density point clouds, derived from either experimental or computational sources, as a unified and physically realistic conditioning signal to directly generate novel molecules with plausible 3D conformations. By incorporating such density information, the method effectively captures conformational flexibility within binding sites, mitigates structural bias, and enables joint pretraining and inference using both computational and experimental density data. Evaluations across 101 biological targets demonstrate that the generated molecules exhibit high compatibility with the actual binding environments.

Abstract

Recent advances in generative modeling have enabled significant progress in structure-based drug design (SBDD). Existing methods typically condition molecule generation on empty binding pockets from holo complexes, overlooking informative components such as the filler (ligands and solvent). Here, we leverage low-resolution electron density (ED) derived from the filler as a physically grounded condition for \textit{de novo} drug design. We consider two types of ED, calculated and cryo-EM/X-ray, obtainable from computational or experimental sources, supporting unified pre-training and experimental integration. Compared with rigid pocket representations, experimental ED naturally captures conformational flexibility and provides a more faithful description of the binding environment. Based on this, we introduce EDMolGPT, a decoder-only autoregressive framework that generates molecules from low-resolution ED point clouds. By grounding generation in physically meaningful density signals, EDMolGPT mitigates structural bias and produces molecules with 3D conformations. Evaluations on 101 biological targets verify the effectiveness. Our project page: https://jiahaochen1.github.io/EDMolGPT_Page/.

Problem

• structure-based drug design
• binding pocket
• electron density
• conformational flexibility
• de novo drug design

Innovation

• electron density
• de novo drug design
• generative modeling
• EDMolGPT
• structure-based drug design