Towards Efficient Molecular Property Optimization with Graph Energy Based Models

2025-02-17

AI Summary

Chemical space is vast and complex; attribute-driven molecular generation typically relies on explicit conditional modeling, limiting generalizability and design flexibility. To address this, we propose an implicit molecular optimization framework that operates without attribute labels: a Graph Energy-Based Model (G-EBM) built upon graph neural networks, trained via contrastive divergence, and optimized via implicit gradients for efficient sampling on the learned energy landscape. This approach abandons conventional conditional generation paradigms, enabling end-to-end, label-free, property-directed molecular generation. Evaluated on standard chemical benchmarks—including QM9 and ZINC250k—our method significantly outperforms state-of-the-art conditional generative and reinforcement learning approaches in terms of target property optimization, structural diversity, and synthetic validity. Results demonstrate its efficacy, robustness, and practical utility for de novo drug discovery.

Technology Category

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

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• 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

Abstract

Optimizing chemical properties is a challenging task due to the vastness and complexity of chemical space. Here, we present a generative energy-based architecture for implicit chemical property optimization, designed to efficiently generate molecules that satisfy target properties without explicit conditional generation. We use Graph Energy Based Models and a training approach that does not require property labels. We validated our approach on well-established chemical benchmarks, showing superior results to state-of-the-art methods and demonstrating robustness and efficiency towards de novo drug design.

Problem

• Optimizing molecular properties efficiently
• Generating molecules without property labels
• Improving de novo drug design robustness

Innovation

• Graph Energy Based Models
• implicit property optimization
• label-free training approach