search

drorlab / combind

Integrated physics-based and ligand-based modeling.
Other
60 stars 13 forks source link

Question regarding poses.csv output #6

Open hnisonoff opened 2 years ago

hnisonoff commented 2 years ago

I ran combind and received the following output for poses.csv

ID,POSE,COMBIND_RMSD,GLIDE_RMSD,BEST_RMSD
ligand_6,0,0.883745269850613,0.883745269850613,0.22876830274335147
ligand_7,0,-1.0,-1.0,-1.0
ligand_8,1,-1.0,-1.0,-1.0
ligand_9,0,-1.0,-1.0,-1.0
ligand_a,0,-1.0,-1.0,-1.0
ligand_b,0,-1.0,-1.0,-1.0

Should I be concerned that only one ligand has RMSD's that are not -1? Can you help me understand what this output means. Thank you!

jpaggi commented 2 years ago

Hi Hunter, the POSE column gives the index of the pose that was selected by combind and the remaining RMSD columns provide the RMSD to the native pose of the combind pose, glide pose, and the best (lowest-rmsd) pose in the pose list.

An RMSD of -1 reflects that no native structure was provided for that ligand in the "combind featurize" step. So having many -1's is common. If you thought that you had provided native poses for the other ligands, there is probably a mismatch between the file names of the native poses and the entry titles in the docking results.

Hope that helps!

hnisonoff commented 2 years ago

Thank you! And should I interpret pose 0 as meaning that Glide would have selected these as the top pose without the combind correction factor?

jpaggi commented 2 years ago

Yes exactly. In this example, the poses for all but ligand_8 are unchanged from the Glide prediction.

hnisonoff commented 2 years ago

Great, thank you for all the quick replies! I really enjoyed the paper!

hnisonoff commented 2 years ago

@jpaggi one other question. In the SI of the paper it sounded like you only use the shape similarity features for virtual screening, but I noticed in the code and README that shape similarity is used in addition to MCSS similarity for normal docking pose prediction. Do you have any guidance for whether I should be using both or just one? Can I ignore shape similarity if I am using MCSS?