Conditional Diffusion Based on Discrete Graph Structures for Molecular Graph Generation

Learning the underlying distribution of molecular graphs and generating high-fidelity samples is a fundamental research problem in drug discovery material science. However, accurately modeling rapidly novel remain crucial challenging goals. To accomplish these goals, we propose Conditional Diffusion model based on discrete Graph Structures (CDGS) for graph generation. Specifically, construct forward diffusion process both structures inherent features through stochastic differential equations (SDE) derive as condition reverse generative processes. We present specialized hybrid noise prediction that extracts global context local node-edge dependency from intermediate states. further utilize ordinary equation (ODE) solvers efficient sampling, semi-linear structure probability flow ODE. also combine with gradient guidance molecule property predictor similarity-constrained optimization. Experiments diverse datasets validate effectiveness our framework. Particularly, proposed method still generates high-quality limited number steps.