Autocorrelation of molecular electrostatic potential surface properties combined with partial least squares analysis as alternative attractive tool to generate ligand-based 3D-QSARs.
A database of 106 human A3 adenosine receptor antagonists was used to derive two alternative PLS models: one starting from CoMFA descriptors and the other starting from the autocorrelation descriptors. The peculiarity of this work is the introduction of autocorrelation vectors as molecular descriptors for the PLS analysis. The autocorrelation allows comparing molecules (and their properties) with different structures and with different spatial orientation without any previous alignment. In particular, Molecular Electrostatic Potential (MEP) was the property computed and its information encoded in autocorrelation vectors. The 3D spatial distribution and the values of the electrostatic potential is in fact largely responsible for the binding of a substrate to its receptor binding site. Validation was done with an external test set and the results of the two models were compared. Interestingly, our preliminary results seem to indicate that this new alternative approach could robustly compete with the already well consolidated CoMFA approach. In particular, we have suggested that it could be a very interesting tool to filter large structural database in several virtual screening applications.