Improving the Development of QSAR Prediction Models with the use of Approximate Similarity Approach

The improvement on the QSAR prediction of the trans-stilbenes affinity for the beta-amyloid peptide (employed for detecting the Alzheimer disease) achieved by means of using approximate similarity measurement is presented in this work. A wide spectrum of similarity methods is described, and results obtained by approximate similarity are compared with those obtained by constitutional, fingerprint and descriptor-based similarity. The fact of using similarity corrections by considering distances between the non-isomorphic fragments (the approximate similarity concept) led to accurate QSAR models (Q(2) > 0.80). The high predictive ability achieved by simple methods is remarked.