I want to reproduce the default2018 model, but ...

[Dadiao-shuai]

Issue summary

I am new to gnina, and I have just learnt that there are two papers about default2018.model.
Given that someone asked for the weightfile of default2018 but did not get answer, I guess you group didn't provide the result caffemodel of redock/general/crossdock_default2018 to the public. So now I want to train it myself to get their weight file.

Questions

However, I am confused about the huge amount of datasets in gnina/model/data, there many datasets but I am not sure which one should I use if I wanna reproduce *_default2018.model.

After I read the article, I guess the following is for general_def2018 :

As for crossdock_def2018，I need the dir wierbowski_cd?

As for redock_def2018.model, I just need PDBbind_refined_2019, right?

Your system configuration

Operating system: a docker that installed with gnina
Python3 version:3.9

[dkoes]

Trained model weights in pytorch format can be found here: https://github.com/RMeli/gnina-torch
You can get the caffemodel weights for built-in models with the dumpmodel binary.
For training a crossdocked model you need the crossdock2020 dataset:
https://bits.csb.pitt.edu/files/crossdock2020/

[drewnutt]

The weights files for all default2018 models have been uploaded with (gnina/models#31).

Training of crossdock_default2018 requires the CrossDock2020 dataset.
redock_default2018 is trained on the redocked subset of the same dataset.