| Abstract
| - We describe a method for the automatic generation of weakly correlated descriptors for molecular data sets.The method can be regarded as a statistical learning procedure that turns the molecular graph, representingthe 2D formula of the compound, into an adaptive whole molecule composite descriptor. By translating themolecular graph structure into a dynamical system, the algorithm can compute an output value that is highlysensitive to the molecular topology. This system can be trained by gradient descent techniques, which relyon the efficient calculation of the gradient by back-propagation. We present computational experimentsconcerning the classification of the Developmental Therapeutics Program AIDS antiviral screen data set onwhich the performance of the method compares with that of approaches based on substructure comparison.
|
