A correction method for systematic error in 1H-NMR time-course data validated through stochastic cell culture simulation

Abstract

BackgroundThe growing ubiquity of metabolomic techniques has facilitated high frequency time-course data collection for an increasing number of applications. While the concentration trends of individual metabolites can be modeled with common curve fitting techniques, a more accurate representation of the data needs to consider effects that act on more than one metabolite in a given sample. To this end, we present a simple algorithm that uses nonparametric smoothing carried out on all observed metabolites at once to identify and correct systematic error from dilution effects. In addition, we develop a simulation of metabolite concentration time-course trends to supplement available data and explore algorithm performance. Although we focus on nuclear magnetic resonance (NMR) analysis in the context of cell culture, a number of possible extensions are discussed.ResultsRealistic metabolic data was successfully simulated using a 4-step process. Starting with a set of metabolite concentration time-courses from a metabolomic experiment, each time-course was classified as either increasing, decreasing, concave, or approximately constant. Trend shapes were simulated from generic functions corresponding to each classification. The resulting shapes were then scaled to simulated compound concentrations. Finally, the scaled trends were perturbed using a combination of random and systematic errors. To detect systematic errors, a nonparametric fit was applied to each trend and percent deviations calculated at every timepoint. Systematic errors could be identified at time-points where the median percent deviation exceeded a threshold value, determined by the choice of smoothing model and the number of observed trends. Regardless of model, increasing the number of observations over a time-course resulted in more accurate error estimates, although the improvement was not particularly large between 10 and 20 samples per trend. The presented algorithm was able to identify systematic errors as small as 2.5 % under a wide range of conditions.ConclusionBoth the simulation framework and error correction method represent examples of time-course analysis that can be applied to further developments in 1H-NMR methodology and the more general application of quantitative metabolomics.Electronic supplementary materialThe online version of this article (doi:10.1186/s12918-015-0197-4) contains supplementary material, which is available to authorized users.

Effect of an amyloidogenic SARS-COV-2 protein fragment on α-synuclein monomers and fibrils

“…While these figures show a systematic deviation, the percentual change is small and the regression plots in the Supplemental Figure S1 lead to large Pearson correlation coefficients. We remark that the deviations between numerical and NMR measurements are consistent with previously reported coarse-grained molecular dynamic studies 32,33 , and may arise from either systematic errors in the NMR measurements due to reference quantification or a dilution effect 34 , or from use of non-polarizable force fields, leading to systematic errors in the calculations due to the absence of polarization in the H α atoms 35 . Note that, the chemical shift values do not differ between the various trajectories of the isolated α-synuclein monomer suggesting sufficient sampling to achieve equilibrium.…”

Role of microRNA-4458 in patients with non-small-cell lung cancer

“…The green fluorescence intensity was analyzed using Gene Pix Pro version 6.0 (Axon Instruments; Molecular Devices, LLC). The median value correction method was used to obtain the corrected value (26). The expression was regarded as upregulated if the ratio of the miR fluorescence corrected values of the NSCLC and paracarcinoma specimens was ≥1.5.…”

Identifying model error in metabolic flux analysis – a generalized least squares approach

“…NMR spectra acquisition, metabolite quantification, and internal standard correction are described in [ 33 ]. In brief, samples were scanned on a Bruker Avance 600 MHz spectrometer using the first increment of a 1D-NOESY pulse sequence with metabolite quantification carried out using Chenomx NMR Suite 8.1 (Chenomx Inc., Edmonton, Canada).…”