3D STRUCTURAL INFORMATION - FROM PROPERTY PREDICTION TO SUBSTRUCTURE RECOGNITION WITH NEURAL NETWORKS

Two applications of neural networks in molecular recognition, incorpor ating at different levels the structural information, are presented. I nteratomic distances are the basis of the search for a given 3D substr ucture, with a Hopfield network using either a Boltzmann machine or a ''mean held annealing'' algorithm (according to a model we previously proposed by analogy with the ''travelling salesman problem''). Besides atom spatial locations, the model can incorporate characteristic poin ts featuring selected electronic or steric features, and add supplemen tary constraints on the nature of these points or some property value on them. For model compounds, this approach retrieves the correct (fli pped) orientations in binding the adenosine Al receptor. In QSAR field , we use a three layers feed forward neural network to predict the act ivity of polychlorinated dibenzofurans toward the AcH receptor. Due to the high homogeneity of the studied population input data only consis t here of a topological descriptor, refined by a cross validation proc ess. Results compete favorably with the previous approaches, with no n eed for complex held calculations.