Quantum topological QSAR models based on the MOLMAP approach.

Quantum topological molecular similarity produces a two-dimensional array of descriptors for each molecule while a three-dimensional array is obtained by placing the descriptor data matrices of a set of molecules beside each other. Here, we use the multiway data analysis method called molecular maps (MOLMAP) of atom-level properties in a new way. We transferred the three-dimensional array of quantum topological molecular similarity descriptors into new two-dimensional parameters using Kohonen networks, followed by partial least squares. Six different data sets were analyzed by the proposed procedure, which were previously analyzed (Eur. J. Med. Chem. 2006 41 862) by partial least squares applied to unfolded data. They include: (i) the pK(a) of imidazoles, (ii) the ability of a set of indole derivatives to displace [(3)H] flunitrazepam from binding to bovine cortical membranes, (iii) the inhibitory effect of a set of benzimidazoles on the influenza virus, (iv) the interaction of amides with liver alcohol dehydrogenase, (v) inhibition of carbonic anhydrase by sulfonamides and (vi) the toxicity of a set of chlorophenols. Overall, the results showed better statistical results compared with simple unfolding. Furthermore, variable important in projection plots confirmed previous findings about active centers and even in some cases showed more accurate results.