Accelerate Drug Design with CoCoGraph, Now Available on Vecura

What is CoCoGraph?

CoCoGraph (Collaborative Constrained Graph diffusion) is a specialized generative AI model designed for small-molecule drug discovery. Unlike traditional models that require post-hoc filtering to fix invalid chemical structures, CoCoGraph employs a discrete diffusion process using double edge-swapping (DES) operations to ensure 100% chemical validity by construction. By leveraging a paired time-prediction network, it offers precise control over the denoising trajectory to generate high-quality, drug-like molecular candidates.

What can users do with CoCoGraph on Vecura?

With CoCoGraph integrated into the Vecura platform, users can streamline their lead optimization and molecular design workflows:

• Generate Novel Molecules: Create new, chemically valid drug-like molecules using either SMILES-seeded generation to explore existing chemical space or unseeded mode for random graph generation.
• Perform Fragment-based Inpainting: Grow or modify lead compounds by attaching specific chemical fragments to a fixed scaffold, enabling precise bioisosteric replacement and structure-activity relationship analysis.
• Leverage Guided Denoising: Utilize the model's learned trajectory controller to produce stable candidates without the need for manual noise scheduling.
• Access High-Performance Variants: Choose between the efficient BASE model for rapid throughput or the fingerprint-conditioned FPS model for superior generation quality.

What the output means

The output consists of a list of generated molecular SMILES strings, along with their corresponding seed molecules and, in the case of inpainting, the specific fragment formulas utilized. Because every output is guaranteed to be chemically valid, these molecules can be seamlessly integrated into downstream pipelines, including ADMET scoring, protein-ligand docking, and chemical diversity analysis.

This output serves as a high-fidelity starting point for computational chemistry pipelines and should be used to support scientific decision-making, though it does not replace the need for subsequent experimental validation.

Why this matters

In molecular design, the efficiency of AI-driven generation is often bottlenecked by the prevalence of "chemically impossible" structures, which consume valuable computational resources during filtering and analysis. By enforcing atomic valence constraints throughout the entire diffusion process, CoCoGraph eliminates these bottlenecks, providing a cleaner, more reliable set of leads that align with real-world chemistry.

The ability to perform fragment-based inpainting directly on a target scaffold makes this an invaluable tool for researchers aiming to optimize the properties of existing drug candidates. By removing the technical overhead of complex model infrastructure, CoCoGraph on Vecura democratizes access to sophisticated generative drug design techniques.

• Developed by: Research group associated with the CoCoGraph paper (manurubo).
• Source: GitHub Repository
• Reference: arXiv:2505.16365