SIMULATING LIPOPHILICITY OF ORGANIC-MOLECULES WITH A BACKPROPAGATION NEURAL-NETWORK

From a training set of 7200 chemicals, a back-propagation neural netwo rk (BNN) model was developed for calculating the 1-octanol/water parti tion coefficient (log P) of molecules containing nitrogen, oxygen, hal ogen, phosphorus, and/or sulfur atoms. Chemicals were described by mea ns of autocorrelation vectors encoding hydrophobicity, molar refractiv ity, H-bonding acceptor ability, and H-bonding donor ability. A 35/32/ 1 composite network composed of four configurations was selected as th e final model (root-mean-square error (RMS) = 0.37, r = 0.97) because it provided the best simulation results (RMS = 0.39, r = 0.98) on an e xternal testing set of 519 molecules. This final model compared favora bly with a recently published BNN model using variables (atoms and bon ds) derived from connection matrices.