models difference between Crossdock-default2018 and default-2018

I use the built-in model --cnn crossdock_default2018 to dock my dataset firstly to get original results, and then I want to train this model using my own datasets, but I can only find default2018.model file in gnina/models/crossdocked_paper.

What should I do? If the architecture of crossdock-default2018 and default-2018 is the same, then I can compare them, right?

Sep 14 '23 12:09 Kerro-junior

default2018 is the model architecture, crossdock in the training data.

Sep 14 '23 15:09 dkoes

Sorry to interrupt, but I could not find the weght caffemodel file for default2018.model, too.

I am training default2018 for both CNNscore and CNNaffinity as in Three-Dimensional Convolutional Neural Networks and a Cross-Docked Data Set for Structure-Based Drug Design.

Sep 15 '23 09:09 JonasLi-19

For convenience, @drewnutt has uploaded the caffemodel files of all the built-in models: https://github.com/gnina/models/tree/master/crossdocked_paper

Sep 18 '23 17:09 dkoes

Hi professor, I noticed all "Gnina 1.0" related papers are published before 2018, so the caffemodel of default2018 like redock_default2018 are surely trained after that. However, I could not find the latest paper of gnina after 2018. My question is : What version of PDBbind did you use for the current caffemodel of default2018? PDBbind2020?

Dec 17 '23 14:12 RJ-Li

The models are trained on the CrossDocked2020 dataset, which is described in a 2020 paper (see the list of papers at the top of the gnina README). https://pubs.acs.org/doi/10.1021/acs.jcim.0c00411

Dec 17 '23 14:12 dkoes

I found that the default2018.model in acs2018 has no 'top: "rmsd_true"' and 'has_rmsd: false'. Then I learnt that CrossDock2020 has used docked pose so it has rmsd. So the default2018.model in acs2018 is not compatible for crossdock2018.caffemodel anymore. @Kerro-junior

Dec 19 '23 12:12 Dadiao-shuai