Meeting Banner
Abstract #1513

Non-Negative Principal Component Analysis Based Scaling: Application on NMR Spectroscopic Metabolomics

Lingli Deng1, 2, Jiyang Dong1, Zhong Chen1, 2

1Department of Electronic Science, Xiamen University, Xiamen, Fujian, China; 2Department of Communication Engineering, Xiamen University, Xiamen, Fujian, China

Scaling is an important data preprocessing procedure prior to multivariate statistical analysis for NMR spectroscopic metabolomics. The commonly used methods scale each variable of the data independently, which ignores the chemical meaning of the spectra and may make the subsequent analysis be hard to interpret. A new scaling method based on non-negative principal component analysis (NPCA) is proposed in this paper. It aims to perform scaling on the concentration of the metabolites rather than on the variables. Analysis results of simulated and real 1H NMR spectra presents itself as an interpretable pretreatment method for multivariable data analysis.

Abstract PDF

Keywords

aims aligned anal application approximation audience available basic better bins biochemical bucked chem chemical china common commonly communication component concentration concentrations consisted contains contribute correct corrected correlates dataset dong electronic employed engineering ensure equal equally example factorize greatly hard hence identification ignores important improved independently individual influence inter interpret interpretable interval kept last literature loading lost make manually matrix medical metabolic metabolite metabolites model multivariate natural negative noise nuclear obviously offers original package paper pattern peak physical plots potential preprocessing principal prior procedure processing profiles proposed rather real realistic recognition regarded relevant residual respectively scale scaled scaling science scientists scores simulated simulation spectra spectral spectroscopic spectroscopy stability stacked statistical subsequent synthetic target unit urine variable variables variance vast view width yang