How to Use FlowDock

FlowDock is a generative protein-ligand docking model that predicts both bound complex structures and binding-affinity-related outputs. The paper introduces it as a conditional flow-matching approach that can dock multiple ligands, map apo structures toward holo states, and return pose confidence together with affinity estimates.

On Neurosnap, researchers provide receptor Input Receptor sequences, optionally add an Input Template structure, and supply one or more Input Molecules for docking. That makes the workflow useful for flexible docking, target-focused pose generation, and small virtual-screening batches where both structural plausibility and affinity ranking matter.

The output is organized around sampled poses and comparative metrics, so the tool is best used to rank and inspect candidate complexes rather than to treat a single generated pose as definitive.

The key methodological change is conditional geometric flow matching: the model learns a trajectory from unbound to bound protein-ligand states rather than directly classifying or scoring static poses. In the paper, that enabled FlowDock to handle flexible docking and multi-ligand settings while also producing confidence and affinity estimates for each sampled complex.

On Neurosnap, Number of Samples and Number of Steps control search breadth and refinement depth. Sampler, Sampler ETA, and Prior Type let researchers decide how aggressively the model should explore structural space and whether initialization should lean on sequence-based protein priors or harmonic ligand priors. Optional Detect Covalent and Exact Prior settings are useful when the chemistry or binding-site information is unusually constrained.

Researchers typically use the results page to compare sampled poses, inspect whether receptor context looks reasonable, and combine pose-level metrics with predicted affinity before selecting ligands for rescoring or experiment.

• Flexible docking setup: FlowDock accepts receptor sequences, optional structural templates, and one or more ligands in a single workflow.
• Apo-to-holo modeling: The flow-matching formulation is designed to move unbound inputs toward bound complexes rather than score only fixed poses.
• Sampling controls: Sample count, step count, sampler choice, and prior selection give researchers meaningful control over exploration.
• Decision-ready results: Ranked poses plus affinity-related metrics make it easier to decide which complexes deserve rescoring or experimental follow-up.