Metabonomics Studies Of Lung Cancer Serum By Optimized Partial Least-squares Discriminant Analysis Coupled With~1H Nuclear Magnetic Resonance Spectroscopy

Metabonomics is an important part of system biology including genomics, transcriptomics and proteomics. One of the crucial issues is to use chemometric algorithms to mining the maximal information inherent in the obtained data sets in metabonomics. Partial least-squares discriminant analysis (PLS-DA) has been widely used in data analysis in metabonomic community, due to its simple parameter structure and good model interpretability. The issues of suboptimum and overfitting, however, often occur in PLS-DA modeling. Based on the characteristics of PLS-DA and excellent optimization capability of the particle swarm optimization (PSO), the following two works have been carried out in the current paper:(1) Particle swarm optimization (PSO) has been invoked to improve the performance of PLS-DA via simultaneously selecting the optimal variable subset as well as the associated weights and the best number of latent variables in PLS-DA, forming an optimized version of PLS-DA method, namely, PSO-PLSDA. In this chapter, combined with’H NMR-based metabonomics, PSO-PLSDA is applied to recognize the lung cancer patients from the healthy controls. Compared with the recognition rates of86%and65%for the training and test sets yielded by PLS-DA, PSO-PLSDA those of99%and85%were offered by PSO-PLSDA. Furthermore, PSO-PLSDA also identified several potential discriminative metabolites to aid the diagnosis of lung cancer, including lactate, proline, glycoprotein, glutamate, alanine, threonine, taurine, glucose (α-and β-), trimethylamine, glutamine, glycine, and myo-inositol.(2) In this chapter, another new optimized PLS-DA has been designed by using discrete PSO to simultaneously select the optimal sample as well as variable subsets and the best number of latent variables in PLS-DA, called PSO-SV-PLSDA. A new objective function was formulated to decide the appropriate items involved in PLS-DA, based on the model complexity and accuracy. Combined with ’H NMR-based metabonomics, PSO-SV-PLSDA compared with PLS-DA was also employed to recognize the lung cancer patients from the healthy controls. The results revealed that PSO-SV-PLSDA hold superior recognition capabilities to PLS-DA. Moreover, several most discriminative metabolites were also identified to aid the diagnosis of lung cancer, including lactate, valine, proline, glycoprotein, glutamine, threonine, taurine, glucose (α-and β-), trimethylamine, and lipid.