Radial basis function (RBF) neural network models for the simultaneous estimation of flash point (T f) and boiling point (T b) based on 25 molecular functional groups and their first-order molecular connectivity index (1χ) have been developed. The success of the whole modeling process depended on a network optimization strategy based on biharmonic spline interpolation for the selection of an optimum number of RBF neurons (n) in the hidden layer and their associated spread parameter (σ). The RBF networks were trained by the Orthogonal Least Squares (OLS) learning algorithm. After dividing the total database of 400 compounds into training (134), validation (133), and testing (133), the average absolute errors obtained for the validation and testing sets ranges from 10 °C to 12 °C and 11 °C to14 °C for T f and T b, respectively, and are in agreement with the experimental value of about 10 °C. Results of a standard Partial Least Square (PLS) regression model for single output predictions range from 23 °C to 24 °C and 18 °C to 20 °C for T f and T b, respectively, indicating the superior predictive ability of the neural model and strongly suggests that a nonlinear relationship exists between the input and target parameters of the data. The robustness of the neural model was successfully examined by a random split cross validation based on pooling together of the validation and test data sets. The study shows that the simple numerical coding of a molecule based on its formula together with its 1χ is an attractive way of estimating the flammability properties of organic compounds via an RBF neural network.

No abstract

Pattern recognition techniques (factor analysis and neural networks) were used to investigate and classify human brain tumors based on the 1H NMR spectra of chemically extracted biopsies (n = 118). After removing information from lactate (because of variable ischemia times), unsupervised learning suggested that the spectra separated naturally into two groups: meningiomas and other tumors. Principal component analysis reduced the dimensionality of the data. A back-propagation neural network using the first 30 principal components gave 85% correct classification of meningiomas and nonmeningiomas. Simplification by vector rotation gave vectors that could be assigned to various metabolites, making it possible to use or to reject their information for neural network classification. Using scores calculated from the four rotated vectors due to creatine and glutamine gave the best classification into meningiomas and nonmeningiomas (89% correct). Classification of gliomas (n = 47) gave 62% correct within one grade. Only inositol showed a significant correlation with glioma grade.

1H nuclear magnetic resonance (NMR) spectra of tumors and normal tissue include signals from all hydrogen-containing metabolites and can therefore be considered multicomponent multivariate mixtures. We have obtained 1H spectra from perchloric acid extracts of three normal tissues (liver, kidney, and spleen) and five rat tumors (GH3 prolactinoma, Morris hepatomas 7777 and 9618a, LBDS1 fibrosarcoma, and Walker 256 carcinosarcoma). We have applied several different chemometric methods to analyze the data. First, we used principal component analysis, cluster analysis, and an optimized artificial neural network to develop a classification rule from a training set of samples of known origin or class. The classification rule was then assessed using a set of unknown samples. We were able to successfully determine the class of each unknown sample. Second, we used the chemometric techniques of factor analysis followed by target testing to investigate the underlying biochemical differences that are detected between the classes of samples.