Binding pocket specification

[askusay]

I apologize if I have missed something obvious in the documentation! I remember in the AF3 paper, a feature for specifying binding pockets is described, i.e. to predict a ligand binding poses in a specific region of the protein. Could you please let me know if this is possible and how to do it?

[joshabramson]

Hi, yes, the weights we used for the Posebusters benchmark (which were trained to an earlier cutoff date, to ensure no crossover with the Posebusters set), were finetuned for one ablation to allow conditioning on the pocket position. As this is not the AlphaFold3 model, we are not releasing these weights or conditioning code.