Neural networks molecular representations

J.R.M. Smits, W.J. Meissen, G.J. Daalmans and G. Kateman, Using molecular representations in combination with neural networks. A case study prediction of the HPLC retention index. Computers Chem., 18 (1994) 157-172. 

Although neural networks have been extensively investigated, the main efforts in chemistry research focus on the appropriate representation of data for neural networks. For instance, finding the adequate descriptor for the representation of chemical structures is one of the basic problems in chemical data analysis. The solution to these problems is a mathematical transformation of the molecular data into a vector of fixed length. On the basis of these vectors, several methods of data analysis can be performed statistical evaluation, evaluation of complex relationships, and fast and effective simulation and prediction of molecular features. 

Detail-Filtered One-Level Decomposition The detail-filtered (high-pass) coefficients of the first resolution level represent a new type of descriptor that reveals special aspects of data, like trends, breakdown points, and discontinuities in higher derivatives. It is useful as alternative molecular representation for neural networks in classification and prediction tasks (Figure 5.20). 

We have seen that RDF descriptors are one-dimensional representations of the 3D structure of a molecule. A classification of molecular structures containing characteristic structural features shows how the descriptor preserves effectively the 3D structure information. For this experiment, Cartesian RDF descriptors were calculated for a mixed data set of 100 benzene derivatives and 100 cyclohexane derivatives. Each compound was assigned to one of these classes, and a Kohonen neural network was trained with these data. The task for the Kohonen network was to classify the compounds according to their Cartesian RDF descriptors. 

E. Tafeit, W. Estlelberger, R. Horejsi, R. Moeller, K. Oettl, K. Vrecko, and G. Reibnegger, ]. Mol. Graphics M.odel., 14,12 (1996). Neural Networks as a Tool for Compact Representation of Ab Initio Modell, Molecular Potential Energy Surfaces. 

ANN = artificial neural network CPG = counterpropagation 3D-MoRSE = 3D molecular representation of structures based on electron diffraction FREL = fragment reduced to an environment that is limited. 

---