Beginner's question, how to judge the quality of results based on several scores?

[stan1233]

Issue summary

Steps to reproduce

Dear developer,

I am a beginner in this area and have tested using examples in example colab notebook.

I only know that the lower the affinity, the stronger the binding. Does the other indicator intramol represent higher instability? Are CNN pose score and CNN affinity better when they are higher or lower? Are there any benchmarks for these indicators, below or above which can represent very stable or strong binding?

In the re-scoring, what does the high or low value of CNN variance represent?

Best,
Stan

              _             
             (_)            
   __ _ _ __  _ _ __   __ _ 
  / _` | '_ \| | '_ \ / _` |
 | (_| | | | | | | | | (_| |
  \__, |_| |_|_|_| |_|\__,_|
   __/ |                    
  |___/                     

gnina v1.1 master:e4cb380+   Built Dec 18 2023.
gnina is based on smina and AutoDock Vina.
Please cite appropriately.

Commandline: ./gnina -r rec.pdb -l lig.pdb --autobox_ligand lig.pdb -o docked.sdf --seed 0
==============================
*** Open Babel Warning  in PerceiveBondOrders
  Failed to kekulize aromatic bonds in OBMol::PerceiveBondOrders

Using random seed: 0

0%   10   20   30   40   50   60   70   80   90   100%
|----|----|----|----|----|----|----|----|----|----|
***************************************************

mode |  affinity  |  intramol  |    CNN     |   CNN
     | (kcal/mol) | (kcal/mol) | pose score | affinity
-----+------------+------------+------------+----------
    1       -8.50       -0.79       0.9491      6.845
    2       -8.51       -0.84       0.8983      6.785
    3       -8.30       -0.72       0.4567      6.462
    4       -6.12       -0.53       0.3316      6.002
    5       -6.27       -0.62       0.2152      5.766
    6       -6.13        0.76       0.2049      5.790
    7       -6.84       -0.98       0.1468      5.442
    8       -6.90       -0.77       0.1340      5.333
    9       -6.92       -0.47       0.1271      5.910

!./gnina -r rec.pdb -l lig.pdb --autobox_ligand lig.pdb --score_only

              _             
             (_)            
   __ _ _ __  _ _ __   __ _ 
  / _` | '_ \| | '_ \ / _` |
 | (_| | | | | | | | | (_| |
  \__, |_| |_|_|_| |_|\__,_|
   __/ |                    
  |___/                     

gnina v1.1 master:e4cb380+   Built Dec 18 2023.
gnina is based on smina and AutoDock Vina.
Please cite appropriately.

Commandline: ./gnina -r rec.pdb -l lig.pdb --autobox_ligand lig.pdb --score_only
==============================
*** Open Babel Warning  in PerceiveBondOrders
  Failed to kekulize aromatic bonds in OBMol::PerceiveBondOrders

## Name gauss(o=0,_w=0.5,_c=8) gauss(o=3,_w=2,_c=8) repulsion(o=0,_c=8) hydrophobic(g=0.5,_b=1.5,_c=8) non_dir_h_bond(g=-0.7,_b=0,_c=8) num_tors_div
Affinity: -8.23943 (kcal/mol)
CNNscore: 0.97413 
CNNaffinity: 6.98467
CNNvariance: 0.07986
Intramolecular energy: -0.51286
Term values, before weighting:
##  84.31120 1224.00134 2.82601 33.60834 2.67216 0.00000

[dkoes]

You want a larger CNNaffinity (this is a pK value) and higher CNNscore (this is a pose quality score). Affinity you want lower (it is an energy). The CNNvariance is the variance between the models in the ensemble and gives some idea how uncertain the neural networks are, but we haven't done any calibration with it so I can't tell you what a good/bad value is.

CNN_VS, which is in the output sdf, is CNNscore*CNNaffinity.

[stan1233]

Thank you for your explanation! I have made more attempts and encountered some issues that I would like to consult with you about.

I performed blind docking on some proteins that were from AF2, and used GNINA for rescoring. If the Affinity becomes a positive value, does it mean that this docking result is unreliable?

Can the value of intramol be used to determine whether this output complex is reliable?

I analyzed the interaction and indeed felt that the ligand and receptor are relatively far apart from each other.

If I use --minimize for energy minimization, and the final score is still a positive value, does this mean that the docking was unsuccessful?

[dkoes]

If Affinity is positive, there are steric clashes.