I think this has been reported before, but I can't find a previous post about it.
For ligands, the score reported by current.GetSegmentScore may be much lower than the total of the subscores reported by current.GetSegmentSubscore.
The recipe getligscores demonstrates the problem.
As the image below shows, on puzzle 2634, the segment information window reported a total score of 6,605 for ligand segment 369.
But current.GetSegmentScore says the score for segment 369 is only 330.
The total of the current.GetSegmentSubscore values for the subscores listed in the segment information window matches the 6,605 value shown.
It's not immediately clear to me where the discrepancy of 6,275 comes from.
Hi Loci, Thanks for the report. – Is this happening with the starting structure, or are there particular solutions which show this behavior? (In the latter case, please Share with Scientists one of the examples where it happens.)
For puzzle 2637, the starting pose shows the problem. Segment 369 reports score -159, but the total of segment 369 subscores is -3,185, for a difference of -3,206. The segment information window shows score -3,185 for 369. The image shows the segment information window and the output from getligscores for the starting pose.
For 2637, other poses show the problem as well. One solution shared with scientists as "ligand score mismatch".
I think I've got it – The key insight was looking at the fold difference, rather than the delta difference. The GetSegmentScore values are consistently 20 fold less than the GetSegmentSubscore values. This is because in ligand design puzzles we consistently upweight the score contribution of the ligand. (The factor happens to be 20 for this particular puzzle.) Checking the code, the GetSegmentSubscore value (and the tab window) gets the correct weighting factor applied, but the GetSegmentScore is reporting the (incorrect) pre-upweighting score.
I've put together a patch for the next devprev.
