MolBB: Molecular Graph Generation via Latent Brownian Bridge Diffusion Model

Due to the vastness of chemical space and the structural complexity of molecules, designing novel molecular graphs with desired properties remains a central challenge in computational chemistry. Although deep generative models have achieved remarkable progress in recent years, existing approaches still suffer from high modeling complexity and a lack of chemical plausibility in generated structures. To address these issues, we propose MolBB, a novel molecular graph generation framework that integrates the Brownian Bridge Diffusion Model (BBDM) with a chemically-regularized variational graph autoencoder. MolBB introduces a Chemically Regularized Latent Space (CRLS) to replace traditional Gaussian priors, lever-aging scaffold-informed Brownian bridges to enhance the chemical validity of generated molecules while preserving structural diversity. As far as we know, MolBB is the first model to incorporate Brownian bridge diffusion for molecular graph generation. Extensive experiments on multiple benchmark datasets demonstrate that MolBB achieves state-of-the-art performance in both unconditional molecule generation and molecular property optimization tasks. © 2025 IEEE.