search

HannesStark / EquiBind

EquiBind: geometric deep learning for fast predictions of the 3D structure in which a small molecule binds to a protein
MIT License
477 stars 109 forks source link

Baseline details, global docking configuration, GLIDE. #32

Closed luwei0917 closed 2 years ago

luwei0917 commented 2 years ago

Hello, In the paper, you mentioned the GLIDE runtime, but we couldn't find the setting for GLIDE, and other baselines. Since those methods typically perform a local docking, requiring the specification of pocket center, and box sizes, it will be great if you can provide more details for performing global docking using GLIDE and others. Thanks!

HannesStark commented 2 years ago

Hi! The runtime of GLIDE in the paper is much higher than that of other methods because we could not parallelize it and it only ran on a single CPU instead of 16 which the others used. From here (https://www.schrodinger.com/kb/1165), I gather that with multiple licenses, GLIDE can be run in parallel. (described in the asterisk next to the runtime) I think this would reduce the runtime by 16 times.

The pocket center and box size were chosen such that the whole receptor is in the box with an additional 4 Angström in each direction. The reason for the 4 A is that this is the default setting of GNINA's autobox_ligand parameter.

For the other parameters, we use the default settings. Except for QVina-W where we increase the exhaustiveness to 64.