Why are the results from bash and python different?

[xywawawa]

here is the python code:

from DockQ.DockQ import load_PDB, run_on_all_native_interfaces

model = load_PDB("4g3y/complex.1.pdb")
native = load_PDB("4g3y/native.pdb")

chain_map = {"L": "L", "H": "H", "C": "C"}

results, total_score = run_on_all_native_interfaces(model, native, chain_map=chain_map)

print(f"Total DockQ score: {total_score}")

And i get a result:

Total DockQ score: 1.0878364906515292

Interface between chains LH:
DockQ: 1.000
iRMSD: 0.000
LRMSD: 0.000
fnat: 1.000
fnonnat: 0.000
F1: 1.000
clashes: 0

Interface between chains LC:
DockQ: 0.044
iRMSD: 11.872
LRMSD: 23.521
fnat: 0.000
fnonnat: 1.000
F1: 0.000
clashes: 2

Interface between chains HC:
DockQ: 0.044
iRMSD: 11.486
LRMSD: 23.521
fnat: 0.000
fnonnat: 1.000
F1: 0.000
clashes: 2

I also run a command:

DockQ 4g3y/complex.1.pdb 4g3y/native.pdb --mapping LHC:LHC

get a result:

Total DockQ over 3 native interfaces: 0.363 with LHC:LHC model:native mapping
Native chains: L, H
        Model chains: L, H
        DockQ: 1.000
        iRMSD: 0.000
        LRMSD: 0.000
        fnat: 1.000
        fnonnat: 0.000
        F1: 1.000
        clashes: 0
Native chains: L, C
        Model chains: L, C
        DockQ: 0.044
        iRMSD: 11.872
        LRMSD: 23.521
        fnat: 0.000
        fnonnat: 1.000
        F1: 0.000
        clashes: 2
Native chains: H, C
        Model chains: H, C
        DockQ: 0.044
        iRMSD: 11.486
        LRMSD: 23.521
        fnat: 0.000
        fnonnat: 1.000
        F1: 0.000
        clashes: 2

Why are the total dockq scores different? The result from bash seems to be a avarage result from (1.0000 + 0.044 + 0.044) / 3. But since the receptor is fixed(chain L,H) when generating decoys(model), it's unfair to calculate the score using chain L,H.

[clami66]

Hi,

The "Total DockQ" reported in the bash call is the average, as you said.

The total reported by python are not averaged, just summed together, as they are what DockQ tries to maximize when optimizing the mapping native and model interfaces. In your case, I would suggest you extract from the results dictionary the DockQ scores for the relevant interfaces (LC and HC) and combine those as desired.