IBEX: Information-Bottleneck-EXplored Coarse-to-Fine Molecular Generation under Limited Data

2025-08-14

AI Summary

In structure-based drug design, data scarcity of protein–ligand complexes severely limits the generalizability of generative models and exacerbates overfitting. Method: We propose IBEX, a “coarse-to-fine” generative framework grounded in PAC-Bayesian information bottleneck theory. It is the first to employ the information bottleneck to quantify the information density of masking strategies and derive generalization bounds—revealing that scaffold hopping effectively increases model capacity. Building upon TargetDiff, IBEX preserves binding-pocket constraints during generation and integrates L-BFGS optimization of physical energy terms and rigid-body degrees of freedom for conformational refinement. Results: IBEX achieves a 64% zero-shot docking success rate (up from 53%), an average AutoDock Vina score of −8.07 kcal/mol, and attains the best median binding energy on 42 of 57 targets. It improves QED by 25%, achieves state-of-the-art validity and diversity, and significantly reduces extrapolation error.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Information Extraction
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Three-dimensional generative models increasingly drive structure-based drug discovery, yet it remains constrained by the scarce publicly available protein-ligand complexes. Under such data scarcity, almost all existing pipelines struggle to learn transferable geometric priors and consequently overfit to training-set biases. As such, we present IBEX, an Information-Bottleneck-EXplored coarse-to-fine pipeline to tackle the chronic shortage of protein-ligand complex data in structure-based drug design. Specifically, we use PAC-Bayesian information-bottleneck theory to quantify the information density of each sample. This analysis reveals how different masking strategies affect generalization and indicates that, compared with conventional de novo generation, the constrained Scaffold Hopping task endows the model with greater effective capacity and improved transfer performance. IBEX retains the original TargetDiff architecture and hyperparameters for training to generate molecules compatible with the binding pocket; it then applies an L-BFGS optimization step to finely refine each conformation by optimizing five physics-based terms and adjusting six translational and rotational degrees of freedom in under one second. With only these modifications, IBEX raises the zero-shot docking success rate on CBGBench CrossDocked2020-based from 53% to 64%, improves the mean Vina score from $-7.41 kcal mol^{-1}$ to $-8.07 kcal mol^{-1}$, and achieves the best median Vina energy in 57 of 100 pockets versus 3 for the original TargetDiff. IBEX also increases the QED by 25%, achieves state-of-the-art validity and diversity, and markedly reduces extrapolation error.

Problem

• Addresses scarcity of protein-ligand complex data
• Improves molecular generation transfer performance
• Enhances docking success rate and Vina score

Innovation

• PAC-Bayesian information-bottleneck theory for sample analysis
• Coarse-to-fine pipeline with L-BFGS optimization
• Physics-based refinement of molecular conformations