Predictive flip regression: a technique for QSAR of derivatives of symmetric molecules

We have for the first time used the flip regression technique to predict the activity of unknown drugs given a substantial series of derivatives of symmetrical molecules such as benzene, with known activity. As descriptors we have used quantum theoretic parameters such as orbital energies and the orientation of pi-orbital nodes and Mulliken charges of atoms. Flip regression is a technique for dealing with multiply substituted derivatives of symmetric molecules, such as phenethylamine or benzenesulfonamide. The method has been tried on two large data sets: thrombin inhibitors and carbonic anhydrase (CA) inhibitors. Multiple validation tests were run, randomly splitting the data into training and test sets and successfully predicting the activity of the test sets, with predictive R2 of around 0.9 for the thrombin inhibitors and 0.7 for the CA inhibitors.