Contrastive Geometric Learning Unlocks Unified Structure- and Ligand-Based Drug Design

TL;DR

ConGLUDe tackles the fragmentation between structure-based and ligand-based drug design by introducing a unified contrastive geometric learning framework that jointly trains a geometric protein encoder and a lightweight ligand encoder. It extends VN-EGNN with a global protein node and integrates a 3-axis InfoNCE objective to align protein, pocket, and ligand representations, enabling pocket prediction without predefined pockets and efficient pocket-conditioned screening. Across virtual screening, zero-shot target fishing, and pocket-related tasks, ConGLUDe achieves state-of-the-art or competitive performance while maintaining high computational efficiency, demonstrating strong cross-domain generalization. This work advances toward general-purpose foundation models for drug discovery by leveraging both structural complexes and large-scale bioactivity data.

Abstract

Structure-based and ligand-based computational drug design have traditionally relied on disjoint data sources and modeling assumptions, limiting their joint use at scale. In this work, we introduce Contrastive Geometric Learning for Unified Computational Drug Design (ConGLUDe), a single contrastive geometric model that unifies structure- and ligand-based training. ConGLUDe couples a geometric protein encoder that produces whole-protein representations and implicit embeddings of predicted binding sites with a fast ligand encoder, removing the need for pre-defined pockets. By aligning ligands with both global protein representations and multiple candidate binding sites through contrastive learning, ConGLUDe supports ligand-conditioned pocket prediction in addition to virtual screening and target fishing, while being trained jointly on protein-ligand complexes and large-scale bioactivity data. Across diverse benchmarks, ConGLUDe achieves state-of-the-art zero-shot virtual screening performance in settings where no binding pocket information is provided as input, substantially outperforms existing methods on a challenging target fishing task, and demonstrates competitive ligand-conditioned pocket selection. These results highlight the advantages of unified structure-ligand training and position ConGLUDe as a step toward general-purpose foundation models for drug discovery.

Figures (4)

• Figure 1: Joint training of ConGLUDe on structure- and ligand-based data enables (ligand-conditioned) pocket prediction, virtual screening and target fishing.
• Figure 2: ConGLUDe architecture and training proceduce. A: Message-passing scheme of ConGLUDe's protein encoder based on VN-EGNN: 1. message exchange between residue nodes, 2. residue nodes to virtual pocket nodes, 3. pocket nodes to residue nodes, 4. residue nodes to virtual protein node, 5. virtual protein node to residue nodes. B: The protein encoder supplies a representation of the whole protein $\bm{p}$, and of each detected pocket $\bm{b}_k$. The ligand encoder encodes each small molecule into a protein matching representation $\bm{m}_{\mathrm{p}}$ and a pocket-matching representation $\bm{m}_{\mathrm{b}}$. C: Contrastive loss functions used in our approach. Structure-based losses include $\mathcal{L}_{\mathrm{p2m}}$: InfoNCE between a concatenated protein-pocket representation and all ligand representations from the batch, $\mathcal{L}_{\mathrm{m2p}}$: InfoNCE between a ligand and all protein representations in the batch, and $\mathcal{L}_{\mathrm{m2b}}$: InfoNCE between a ligand and all pocket representations from the corresponding protein. The NCE loss between a protein and annotated ligand representations ($\mathcal{L}_{\mathrm{LB}}$) is used on ligand-based data.
• Figure 31: Distributions of maximum Tanimoto similarities between ECFP count fingerprints (radius 2, length 2048) of test-set molecules and those in the structure- and ligand-based training sets.
• Figure 61: t-SNE projection of protein and ligand embeddings for the DUD-E target with PDB ID 2FSZ.