Use FlowDock

The amino acid sequence(s) corresponding to the chains of the receptor protein.

An optional already docked protein-ligand(s) complex.

Input small molecules for docking. All inputs are converted to SMILES format, with SMILES strings being the preferred input for optimal results.

The number of samples to generate.

Number of sampling steps to perform for each sample.

FlowDock employs both Ordinary Differential Equation (ODE) and Variance Diminishing ODE (VD-ODE) solvers to integrate its learned vector fields during the docking process, with VD-ODE being the default. The ODE solver ensures smooth and accurate trajectories between the unbound (apo) and bound (holo) states, maintaining consistency with the model's learned flow field. In contrast, the VD-ODE solver introduces variance damping during early timesteps, sharply interpolating toward the final prediction in later stages for greater stability and efficiency.

The Variance Diminishing ODE (VD-ODE) sampler includes a variance diminishing factor that controls the balance between exploration and exploitation during the sampling process. This factor offers a trade-off, with a value of 1.0 encouraging broader sampling to explore diverse structural conformations, while values greater than 1.0 prioritize exploitation, focusing on refining specific predictions and converging on high-confidence solutions. The default value is 1.0, ensuring a balanced approach suitable for most docking scenarios.

FlowDock uses two main priors to initialize unbound (apo) structures: the ESMFold prior for proteins and the harmonic prior for ligands. The ESMFold prior, which is the default for proteins, predicts apo structures directly from their sequences using the ESMFold model and adds a small amount of Gaussian noise to prevent overfitting and capture natural variations. The harmonic prior generates initial ligand conformations based on the ligand's bond graph, ensuring physically plausible starting structures. These priors provide reliable and biologically meaningful foundations for mapping apo structures to their bound (holo) states during docking.

Whether to use the ground-truth binding site for sampling, if it is available.

Whether to detect covalent bonds between the input receptor and ligand, if present.