Improved classification accuracy in 1- and 2-dimensional NMR metabolomics data using the variance stabilising generalised logarithm transformation

摘要: Classifying nuclear magnetic resonance (NMR) spectra is a crucial step in many metabolomics experiments. Since several multivariate classification techniques depend upon the variance of data, it important to first minimise any contribution from unwanted technical arising sample preparation and analytical measurements, thereby maximise wanted biological between different classes. The generalised logarithm (glog) transform was developed stabilise DNA microarray datasets, but has rarely been applied data. In particular, not rigorously evaluated against other scaling used metabolomics, nor tested on all forms NMR including 1-dimensional (1D) 1H, projections 2D 1H J-resolved (pJRES), intact (JRES). Here, effects glog are compared two commonly stabilising techniques, autoscaling Pareto scaling, as well unscaled four methods terms data accuracy following analysis, latter achieved using principal component analysis followed by linear discriminant analysis. For three datasets analysed, accuracies were highest transformation: 100% for discriminating 1D hypoxic normoxic invertebrate muscle, JRES fish livers sampled rivers. third dataset, pJRES urine breeds dog, equal accuracies. Additionally we extended algorithm effectively suppress noise, which proved critical spectra. We have demonstrated that transforms datasets. This significantly improves discrimination classes resulted higher unscaled, autoscaled or scaled confirmed broad applicability approach disparate samples spectra,