[논문 리뷰] Generating 3D Molecules for Target Protein Binding

Meng Liu, Youzhi Luo외 3명|arXiv (Cornell University)|2022. 04. 19.

Computational Drug Discovery Methods인용 수 43

한 줄 요약

그래프BP는 등가 변환에 대한 로컬 좌표 시스템과 흐름 기반 생성을 이용하여 연속적으로 3D 공간에 원자를 배치하고 주어진 단백질 결합 부위에 결합하는 3D 분자를 생성한다.

ABSTRACT

A fundamental problem in drug discovery is to design molecules that bind to specific proteins. To tackle this problem using machine learning methods, here we propose a novel and effective framework, known as GraphBP, to generate 3D molecules that bind to given proteins by placing atoms of specific types and locations to the given binding site one by one. In particular, at each step, we first employ a 3D graph neural network to obtain geometry-aware and chemically informative representations from the intermediate contextual information. Such context includes the given binding site and atoms placed in the previous steps. Second, to preserve the desirable equivariance property, we select a local reference atom according to the designed auxiliary classifiers and then construct a local spherical coordinate system. Finally, to place a new atom, we generate its atom type and relative location w.r.t. the constructed local coordinate system via a flow model. We also consider generating the variables of interest sequentially to capture the underlying dependencies among them. Experiments demonstrate that our GraphBP is effective to generate 3D molecules with binding ability to target protein binding sites. Our implementation is available at https://github.com/divelab/GraphBP.

연구 동기 및 목표

• Address structure-based drug design by generating 3D molecules conditioned on a target protein binding site.
• Capture both 3D geometric structure and chemical interactions in the generation process.
• Allow continuous atom placement without discretizing the 3D space while maintaining transformation equivariance.
• Model dependencies among atom type and spatial coordinates through sequential generation.

제안 방법

• Represent the protein-ligand complex as a growing 3D graph and encode context with a 3D GNN.
• Select a local reference atom via auxiliary classifiers to build a local spherical coordinate system for equivariant placement.
• Place a new atom by generating its type and relative position (d, theta, phi) using an autoregressive flow model.
• Generate variables sequentially (a_t, d_t, theta_t, phi_t) to capture dependencies among atom type and geometry.
• Train with a joint objective including atom placement likelihood and auxiliary classifier losses (L_ap, L_cc, L_fc).

실험 결과

연구 질문

• RQ1 Can GraphBP generate 3D molecules that exhibit binding affinity to specified protein binding sites?
• RQ2 Does the sequential generation with a local coordinate system preserve equivariance under rigid transformations?
• RQ3 How does GraphBP's performance compare to LiGAN variants on structure-based drug design tasks?
• RQ4 What is the impact of the sequential generation order on capturing dependencies among atom type and geometry?

주요 결과

• GraphBP outperforms baselines in generating 3D molecules with binding affinity to target binding sites.
• The method preserves equivariance by constructing a local spherical coordinate system for atom placement.
• Sequential generation captures dependencies among atom type and geometry, improving placement quality.
• Experiments use CrossDocked2020 and show significant improvements over LiGAN-prior and LiGAN-posterior baselines.
• Atoms are generated in any continuous position rather than discretized grids, enabling flexible placement.