How To Obtain True Positive And True Negative Datasets To Be Used In Training A Docking Predictor

Question

[Kevin]

Hi, I am trying to train a predictor from some docking results. To do this I will need both true negative and true positive datasets. These datasets must have a PDB, small molecule and an indication of whether they interact.

I can probably make a true positive dataset by finding all PDBs which are bound to a small molecules but I'm not sure about the true negatives. I'm hoping that these datasets exist already from other publications.

Does anyone know where I might find these datasets? Thanks, Kevin

Answer 1

[Vladimir Chupakhin]

Most common method is to use DUD or newer version - DUDe. The generated compounds - called decoys, active compounds can be found in ChEMBL.

[Istvan Albert]

Dude! This tool is worthy of inclusion in CRAC: funny and/or weird names for bioinformatics tools