Scoring using merged feature maps

Here is an approach to identifying overlap of poses with multiple fragment hits.
The procedure is described in this notebook: https://github.com/tdudgeon/jupyter_mpro/blob/master/SuCOSStein.ipynb

In brief, RDKit feature maps for existing fragment hits are merged to generate a ‘Frankenstein’ feature map that can be used to score candidate poses. That merged feature map should contain all the important features for interaction that the fragment hits is trying to tell us about.
The distribution of scores looks reasonable and a quick visual inspection of the poses of the best and worst scoring poses confirms that it seems to work.

3 Likes