Predicting binding affinity remains the holy grail in computational drug discovery. It may works on some target but not on others. Even extremely compute-intense approaches seem not to work, that work consistently well. One reason is that we are over-estimating the quality of our data (the PDB structures) we are working with. Some flaws in this data are well-known low resolution or high temperature factors could be taken into account but checking the electorn density we find many more issues. After all the crystal structure is also just a model that more or less fits to the primary experimental data. Next, the path from a raw PDB file to the input necessary for detailled molecular modelling is not always obvious. We've analyzed this carefully and provide a novel solution. Finally water poses a particular problem in the modeling process. Some water molecules are crucial, others should be replaced to gain affinity. However, careful analysis again shows that also the water molecules in the crystal structure model are not all "well defined". Therefore we will have a look at water molecules and how we can measure their experimental support. In summary, in many cases a quantitative assessment is next to impossible but it is good to "know your enemies" and a qualitative assessment may still be very helpful. E.g. for the prioritization of compounds no absolute predictions are needed, but structure-activity relations and trends in the affinity data. We will show how the Hyde scoring pinpoints issues in the data and provides at least a qualitative assessment that is extremely helpful for the every-day tasks of SAR-analysis and compound prioritization.

Predicting binding affinity doesn't work, or does it?

251st National Meeting (2016)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.