Neural Network Classification of Mutagens Using Structural Fragment Data

Abstract

A neural network was applied to a large, structurally heterogeneous data set of mutagens and non-mutagens to investigate structure-property relationships. Substructural data comprising a total of 1280 fragments were used as inputs. The training of the back-propagation networks was directed by an algorithm which selected an optimal subset of fragments in order to maximize their discriminating power, and a good predictive network. The system comprised three programs: the first used a keyfile of 100 fragments to generate training and test files, the second was the network itself and a procedure for ranking the effectiveness of these fragments and the third randomly replaced the lowest fragments. This cycle was then repeated. After running on a 386/33 PC several networks produced approximately 11% failures in the test set and 6% in the training set. By simplifying the output of the hidden layer it was possible to describe the hidden layer states in terms of clusters of mutagens and non-mutagens. Some of these clusters were structurally homogeneous and contained known mutagenic and non-mutagenic structural classes. This analysis provided a useful means of demonstrating how the network was classifying the data.

A comparative study of molecular similarity, statistical, and neural methods for predicting toxic modes of action

“…These models can be used for prediction, classification, or optimization, and are relatively independent of data types or distributions [39]. Neural networks have recently received attention in the fields of applied chemistry and QSARs [16–18]. In this study, the Learning Vector Quantization (LVQ) classification network was used [40].…”

“…Quantitative molecular similarity analysis methods have been used for estimation of physicochemical [13,14] and toxicologic [13,15] properties as well. Recent studies have also utilized neural network methods for estimating properties from structural descriptors [16–18]. Discriminant analysis is a statistical technique commonly used for classification situations and has been used to model diverse physicochemical and toxicologic properties [19,20].…”

QSAR in Toxicology. 3. Prediction of Chronic Toxicities

“…Another related approach has been the use of a neural network to Note that there is no longer a bilinear relationship with hydrophoidentify fragments that discriminate between mutagens and non-bicity and the difference between HOMO and LUMO energies (remutagens taken from the RTECS data base [19]. Ultimately this lating to molecular stability) is now important to predict provides a model with good predictive capabilities.…”

The Challenges Involved in Modeling Toxicity Data In Silico: A Review

“…A recent comprehensive review [232] of different in silico models and approaches for predictions of genotoxic outcome, shows that most of the earlier approaches described for the prediction of Ames mutagenicity produced good specificity and sensitivity values (prediction accuracy of up to 85%). Depending on the descriptors and the statistical methods used, some of the models offer simple SAR information [82,244], whilst others are harder to interpret due to the choice of chemical descriptors derived from structural information [245,246].…”

“…Different QSAR and machine learning methods have been used to derive in silico predictions about the Ames outcome of the chemicals. These include Ames test QSAR models using PLS, NN, RF, and SVM [46,[244][245][246][247][248][249][250].…”