Application of wavelet transform and cascade correlation neural networks to mass and ion mobility spectra

Prediction of toxicity and substructures is useful for screening environmentally hazardous compounds from gas chromatography-mass spectrometric (GC-MS) data. Classification of mass spectra by toxicity and substructure allows the detection of chromatographic peaks from potentially hazardous compounds that are missing from a reference database. Artificial neural networks may predict substructures and toxicity from mass spectra without first determining the exact configurational structure of the pesticides. Temperature constrained cascade neural networks (TCCCN) are self-configuring networks that train rapidly and are applied to classifying mass spectra. TCCCN models are used to mathematically resolve peaks of different classes in the chromatograms by substructure and toxicity.; Fourier transform (FT) and wavelet transform (WT) are powerful noise reduction and data compression methods. Ion mobility spectrometry (IMS) is a sensing technique that can generate a large amount of data in a short-time monitoring event. A two-dimensional Fourier compression method has been developed for IMS data. The comparative study for different wavelet denoising has been accomplished. The results indicated WT denoising methods are generally better than Fourier denoising method.; WT preprocessing offers two advantages over non-preprocessing: data compression and noise reduction. The compression makes the training faster and makes the training possible for large data sets. The wavelet compression in which data are compressed by frequency usually provides better compression efficiency for most data. The reduction of noise is important in the multivariate analysis because many methods tend to overfit the model. WT preprocessed data can be directly input to other data processing methods, such as singular value decomposition (SVD), linear discriminant analysis (LDA), and TCCCN, eliminating the need to restore data back to the original domain. Therefore, in the case that a large number of data need to be compressed for storage and for future processing, wavelet processing is a powerful method. The ion mobility spectra of three aldehyde and three alcohol compounds are used to evaluate the WT-SVD, WT-LDA, and WT-TCCCN methods and the results indicate that the WT preprocessing is advantageous.