Over the past few weeks I have been toiling on quite a few virtual docking studies for my paper on structure-based drug design for beta-2 adrenergic receptors. I observed this common peculiarity in all the studies using different docking software (GLIDE, DOCK, AUTODOCK, etc.) that the rankings do not always paint the whole picture.

The highest ranking molecules don’t necessarily show the best binding during receptor binding assays. This inaccuracy can obviously be attributed to our incomplete knowledge over chemical conformations thereby resulting in such unsatisfactory software programs. But what could be done to improve these programs? The obvious step would be to better understand ligand-protein interactions and other chemical interactions. The next option would be to use some data mining techniques to analyze the data obtained from actual assays and compare them with the rankings. We could use this data to derive commonalities and improve the present programs. I wonder if we have enough data for this to be possible?