How can I use model trained by gnina-torch to perform dock in gnina?

[RJ-Li]

Hi, I want to know if there is an easy way to use model trained by gnina-torch to transfer into Caffe-model-weight file,
or directly use your gnina-torch model to dock (I didn't find it's possible according to your README)?

[RMeli]

Hi @RJ-Li, unfortunately gnina-torch does not provide docking capabilities. In principle it should be possible to transfer PyTorch weights to Caffe weights (you can see #43 (comment) as an example of how to do the opposite), but I expect this to be rather tedious and I haven't triesd it myself. Maybe @dkoes or @drewnutt have tried and can provide some guidance?

In GNINA 1.0: molecular docking with deep learning we suggest to use the deep learning scoring function for re-scoring the poses generated with a classical scoring function, which provides a good tradeoff between accuracy and computational cost. Therefore, one possibility, would be to perform docking with GNINA with --num_modes set to the same value of --num_mc_saved, so that you will output all the generated poses that are usually re-scored internally (you might also want to turn off RMSD filtering). See Figure 1 of the paper for a visual explanation of the two parameters. You can then re-score all the pose outputted by GNINA with gnina-torch and retain only the top $N$ ones (where $N$ is what you would have used as --num_modes, 9 by default).