No abstract

Missing values are known to be problematic for the analysis of gas chromatography-mass spectrometry (GC-MS) metabolomics data. Typically these values cover about 10%–20% of all data and can originate from various backgrounds, including analytical, computational, as well as biological. Currently, the most well known substitute for missing values is a mean imputation. In fact, some researchers consider this aspect of data analysis in their metabolomics pipeline as so routine that they do not even mention using this replacement approach. However, this may have a significant influence on the data analysis output(s) and might be highly sensitive to the distribution of samples between different classes. Therefore, in this study we have analysed different substitutes of missing values namely: zero, mean, median, k-nearest neighbours (kNN) and random forest (RF) imputation, in terms of their influence on unsupervised and supervised learning and, thus, their impact on the final output(s) in terms of biological interpretation. These comparisons have been demonstrated both visually and computationally (classification rate) to support our findings. The results show that the selection of the replacement methods to impute missing values may have a considerable effect on the classification accuracy, if performed incorrectly this may negatively influence the biomarkers selected for an early disease diagnosis or identification of cancer related metabolites. In the case of GC-MS metabolomics data studied here our findings recommend that RF should be favored as an imputation of missing value over the other tested methods. This approach displayed excellent results in terms of classification rate for both supervised methods namely: principal components-linear discriminant analysis (PC-LDA) (98.02%) and partial least squares-discriminant analysis (PLS-DA) (97.96%) outperforming other imputation methods.

The origin of missing values can be caused by different reasons and depending on these origins missing values should be considered differently and dealt with in different ways. In this research, four methods of imputation have been compared with respect to revealing their effects on the normality and variance of data, on statistical significance and on the approximation of a suitable threshold to accept missing data as truly missing. Additionally, the effects of different strategies for controlling familywise error rate or false discovery and how they work with the different strategies for missing value imputation have been evaluated. Missing values were found to affect normality and variance of data and k-means nearest neighbour imputation was the best method tested for restoring this. Bonferroni correction was the best method for maximizing true positives and minimizing false positives and it was observed that as low as 40% missing data could be truly missing. The range between 40 and 70% missing values was defined as a "gray area" and therefore a strategy has been proposed that provides a balance between the optimal imputation strategy that was k-means nearest neighbor and the best approximation of positioning real zeros.

The type and use of quality control (QC) samples is a 'hot topic' in metabolomics. QCs are not novel in analytical chemistry; however since the evolution of using QCs to control the quality of data in large scale metabolomics studies (first described in 2011), the need for detailed knowledge of how to use QCs and the effects they can have on data treatment is growing. A controlled experiment has been designed to illustrate the most advantageous uses of QCs in metabolomics experiments. For this, samples were formed from a pool of plasma whereby different metabolites were spiked into two groups in order to simulate biological biomarkers. Three different QCs were compared: QCs pooled from all samples, QCs pooled from each experimental group of samples separately and QCs provided by an external source (QC surrogate). On the experimentation of different data treatment strategies, it was revealed that QCs collected separately for groups offers the closest matrix to the samples and improves the statistical outcome, especially for biomarkers unique to one group. A novel quality assurance plus procedure has also been proposed that builds on previously published methods and has the ability to improve statistical results for QC pool. For this dataset, the best option to work with QC surrogate was to filter data based only on group presence. Finally, a novel use of recursive analysis is portrayed that allows the improvement of statistical analyses with respect to the ratio between true and false positives.

No abstract