Accelerating Lead Discovery: DrugFlow is Now Integrated into Vecura

What is DrugFlow?

DrugFlow is a sophisticated, structure-based de novo drug design model that generates 3D candidate ligand molecules within specified protein binding pockets. By integrating continuous flow matching for 3D coordinate generation with discrete Markov bridges for atom and bond type determination, it produces chemically valid, drug-like small molecules that are physically plausible. It is especially useful for early-stage drug discovery, enabling researchers to explore chemical space without requiring prior knowledge of specific chemical series.

What can users do with DrugFlow on Vecura?

With DrugFlow on Vecura, users can:

• Generate Novel Ligands: Create batches of chemically valid, 3D drug-like candidate molecules tailored to specific protein pockets.
• Utilize Advanced Variants: Choose between base models, uncertainty-aware versions (OOD), preference-aligned models for improved drug-likeness, or the FlexFlow extension for joint sampling of protein side-chain angles.
• Customize Generation: Control generation parameters such as molecule size, sampling depth, and pocket definition to focus the search on specific binding environments.
• Implement Post-hoc Quality Control: Apply integrated filters (including PoseBusters validity and ChEMBL ring system checks) to ensure high-quality, relevant results.

What the output means

The output provides candidate molecule SMILES strings and a multi-record SDF file containing 3D atomic coordinates of the generated poses within the binding pocket.

This output should be used to support scientific decision-making in the hit-generation phase. It does not replace the necessity for experimental validation or more rigorous binding affinity assessments.

Why this matters

De novo drug design remains a significant challenge in computational chemistry, particularly when bridging the gap between valid 2D structures and physically constrained 3D pocket interactions. DrugFlow addresses this by providing a robust framework that sidesteps the common artifacts found in discrete-only generation methods, offering a practical tool for rapid hit discovery.

By democratizing access to these generative techniques, Vecura enables medicinal chemists and structural biologists to move beyond traditional screening libraries and generate customized candidates that are optimized for specific protein targets.

• Developed by: LPDI-EPFL (Arne Schneuing, Ilia Igashov, Adrian Dobbelstein)
• Source: ICLR 2025 Paper, GitHub
• Reference: Schneuing et al. (2025). DrugFlow: Structure-based de novo drug design with flow matching and Markov bridges.