Radial Basis Function Neural Network For Classification And Discrimination Of Bacterial Maldi Time-of-flight Mass Spectrometry

The traditional Radial basis function (RBF) neural networks and improved RBF networks were applied to differentiate bacteria cultured at different times (24,48 and 72 hours, respectively) based on matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF-MS). To speed up the network training and reduce the number of variables of the original MS data, wavelet transformation was used to compress the original data from 13826 to 328. The feature peaks of the MALDI mass spectra were maintained in the compressed data. The classification of different species of bacteria cultured at different times was studied and the effect of network parameters was investigated, which may provide more information for biological studies.With the traditional RBF networks, the classification correctness rates of the five bacteria are 71.4%, 61.5%, 80%, 84.6% and 92.8%, respectively. And with the improved RBF networks, better result can be achived. The correctness rates are 85.7%, 69.2%, 80%, 92.3% and 92.8%.The MALDI-TOF-MS can be applied to obtain characteristic spectra from bacteria by using whole cells. This makes the analyses simpler and much faster. But the MALDI-TOF-MS has a high sensitivity to very small changes in chemical composition and the changes in protein composition can be affected by minor variations in environmental conditions during the ongoing processes in the cells, slightly differences in the culture process of bacteria may lead to prominent differences in the spectra. So even using seemingly controlled experimental conditions, the spectra produced from identical strains are rarely identical. This enhances the difficulties in bacterial classification based on comparison of spectra from reference and unknowns.Artificial Neural Network (ANN) is a non-linear, self-adaptive, dynamic system modeled on the human brain. It has superiority in processing incomplete, inaccurate and even fuzzy information. Using ANN, we can find a smooth representation of the underlying trends in the data.Owing to the complexity of biological effects in bacterial growth, more rigid control of bacterial culture conditions seems a critical factor for improving the rate of correctness for bacterial classifications.