Predicting binding affinity is still regarded as the holy grail. With HYDE (embedded in SeeSAR), however, we are one step closer to accurately and reliably predict a protein-ligand's binding affinity. What one often neglects is the accuracy of the data, and that, especially in brute force correlation analysis, we might compare apples with oranges.
HYDE will tell you where the problems are. So you can understand, without looking at any numbers, where you might have to optimize or where a crystal structure might not be credible. Because they are not, as Derek Lowe once put it, a "message from God", but full of assumptions themselves. If we start to understand incorporate this in our work, we will see calculated binding affinity in a whole new (green) light.