Two Applications Of Artificial Neural Networks In The Identification Of Bacterial Genus-species Relationship And The Calibration Analysis Of Chromatography

Artificial neural networks(ANNs),with their excellent abilities of information storage and data processing in parallel,have been widely applied in biology,chemistry, mathematics,computer,automation and other disciplines.They were related to the fields of pattern identification,optimization,signal processing,and so on.Two specific applications of ANNs in bacterial genus-species relationship and the calibration analysis of chromatography were explored in this paper.In the part one,main work was focused on exploring the relationship of the volatile organic compounds(VOCs) from bacterial pure cultures and the bacterial genus and species.Bacterial VOCs are a variety of compounds produced in the metabolism of bacteria. They are formed during both the primary metabolism and the secondary metabolism as side-products,mainly in the metabolic oxidation of glucose from various intermediates. The production of VOCs is greatly affected by microbial species,growth phase and cultural conditions.They were usually employed as biomarkers in the monitoring analyses related to microbiology,environment and microbial physiology.In this part,the relationships of VOCs from bacterial pure cultures and the bacterial genus and species distributed in six genuses were investigated.The VOCs were detected by means of gas chromatography-flame ionization detector(GC-FID) with a previous concentration by head space solid phase microextraction.Factors,including the concentration of Luria-Bertani(LB) medium and cultural time,which affected the production of VOCs and the experimental repeatability,were also studied,using Escherichia coli MG1655 as the reference strain.The results indicated that a better chromatogram repetitiveness of VOCs was obtained at the conditions:bacterium in exponential phase of growth was inoculated into 50-time diluted LB medium with the inoculum concentration of 1%(v/v),then cultured at 35℃for 24 h before detected by GC-FID.The chromatograms of VOCs obtained from bacterial pure cultures were then transformed to data matrices consisted by normalized responses,and cluster analyses were conducted to the matrices.Two cluster analysis methods were employed in this paper,namely hierarchical cluster analysis and Kohonen self organizing map(KSOM). The results shown that both the cluster analysis methods could only differentiate the bacterial genuses partially and the rates of accuracy in their identifying the bacterial species through the chromatograms of VOCs were both less than 60%.Those may indicate that a further study should be conducted,if the VOCs of bacterial pure culture, influenced by a number of factors,were utilized to identify the bacterial genus and species.In the part two,main work was focused on modeling the nonlinear calibration curves of gas chromatography-electron capture detector(GC-ECD) by means of ANNs,aiming to extend their working calibration ranges.The applications of GC-ECD always suffered from its narrow linear range.A radial basis function neural network(RBFNN) method was developed for the first time to model the nonlinear calibration curves of four hexachlorocyclohexane(HCH) isomers,aiming to extend their working calibration ranges in GC-ECD.Other 14 methods,including seven parametric curve fitting methods,two nonparametric curve fitting methods,and five other ANNs methods,were also developed and compared.Their performance of fitting the nonlinear calibration curves of HCH isomers was assessed and compared.Only the RBFNN method,with logarithm-transform and normalization operation on the calibration data,was able to model the nonlinear calibration curves of the four HCH isomers adequately.The RBFNN method accurately predicted the concentrations of HCH isomers within and out of the linear ranges in certified test samples.Furthermore,no significant difference(p＞0.05) was found between the results of HCH isomers concentrations in water samples calculated with RBFNN method and ordinary least squares(OLS) method(R~2＞0.9990).The working calibration ranges of the four HCH isomers were extended from 0.2～20 ng/ml,0.6～60 ng/ml,0.4～20 ng/ml and 0.08～20 ng/ml to 0.2～1000 ng/ml,0.6～1000 ng/ml,0.4～1000 ng/ml and 0.08～1000 ng/ml,respectively.The outstanding nonlinear modeling capability of RBFNN,along with its universal applicability to various problems as a "soft" modeling method,should make the method an appealing alternative to traditional modeling methods in the calibration analyses of various systems besides the ECD.