MagicDock: Toward Docking-oriented De Novo Ligand Design via Gradient Inversion

2025-10-10

AI Summary

Existing ligand de novo design methods suffer from three key limitations: (i) pseudo–de novo generation (not truly zero-shot), (ii) non-differentiable and oversimplified docking modeling, and (iii) inflexible support for diverse ligand types—collectively resulting in suboptimal binding affinity. Method: We propose a docking-oriented, end-to-end differentiable gradient inversion framework that (i) employs learnable 3D point-cloud representations and differentiable surface modeling to accurately capture protein–ligand interactions, (ii) directly optimizes ligand geometry via backward gradient flow to maximize docking scores, and (iii) enables flexible generation of heterogeneous ligands (e.g., small molecules, peptides). Contribution/Results: The method is theoretically sound and computationally efficient. On nine benchmark tasks, it achieves average improvements of 27.1% and 11.7% over SOTA in protein–ligand and molecule–ligand design, respectively, significantly enhancing binding efficacy and practical applicability.

Technology Category

• Search and Optimization: Learning to Search
• Machine Learning: Deep Neural Architectures and Foundation Models
• Game Theory and Economic Paradigms: Mechanism Design

Application Category

• Economics, Online Markets and Human Computation: Uses of LLMs and GenAI for marketplace design, bidding, and strategic interactions
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

De novo ligand design is a fundamental task that seeks to generate protein or molecule candidates that can effectively dock with protein receptors and achieve strong binding affinity entirely from scratch. It holds paramount significance for a wide spectrum of biomedical applications. However, most existing studies are constrained by the extbf{Pseudo De Novo}, extbf{Limited Docking Modeling}, and extbf{Inflexible Ligand Type}. To address these issues, we propose MagicDock, a forward-looking framework grounded in the progressive pipeline and differentiable surface modeling. (1) We adopt a well-designed gradient inversion framework. To begin with, general docking knowledge of receptors and ligands is incorporated into the backbone model. Subsequently, the docking knowledge is instantiated as reverse gradient flows by binding prediction, which iteratively guide the de novo generation of ligands. (2) We emphasize differentiable surface modeling in the docking process, leveraging learnable 3D point-cloud representations to precisely capture binding details, thereby ensuring that the generated ligands preserve docking validity through direct and interpretable spatial fingerprints. (3) We introduce customized designs for different ligand types and integrate them into a unified gradient inversion framework with flexible triggers, thereby ensuring broad applicability. Moreover, we provide rigorous theoretical guarantees for each component of MagicDock. Extensive experiments across 9 scenarios demonstrate that MagicDock achieves average improvements of 27.1% and 11.7% over SOTA baselines specialized for protein or molecule ligand design, respectively.

Problem

• Generating ligands that effectively dock with protein receptors
• Addressing limitations in pseudo de novo and docking modeling
• Creating flexible ligand types through differentiable surface modeling

Innovation

• Uses gradient inversion for ligand generation
• Employs differentiable surface modeling with point-clouds
• Integrates flexible ligand types via unified framework