Data Processing and Analysis

Question 1

Peak picking (convolution)

Answer 1

Used to align and integrate data across multiple samples. Software such as XCMS and mzMine are freely available and commonly applied. The resulting data matrix is typically constructed with thousands of ‘metabolite features’ (m/z-retention time pairs) and tens-to-thousands of samples.

Question 2

Probabilistic quotient normalisation (PQN)

Answer 2

Normalisation method used to reduce the effect of artificial correlations in the data in the case of large differences between the groups

Question 3

Missing values in metabolomics datasets can be observed for three reasons

Answer 3

(1) metabolite is detected in one sample but is not present at any concentration in another sample
(2) metabolite is present in a sample but at a concentration less than the analytical method’s limit of detection
(3) metabolite is present in a sample at a concentration greater than the analytical method’s limit of detection but the data processing software has not detected and reported the metabolite.

Question 4

Missing value imputation (MVI)

Answer 4

Applied to logically replace missing values with a non-zero value while maintaining the data structure. This approach is primarily applied in multivariate analysis which typically operates most robustly with a dataset not containing missing values

Question 5

Missing value imputation methods

Answer 5

• small value replacement (SV)
• mean replacement (MN)
• median replacement (MD)
• k-nearest neighbour (KNN)
• Bayesian PCA (BPCA)
• multivariate imputation by chained equation (MICE)
• Sangster’s method

Question 6

Scaling

Answer 6

Scaling is performed to adjust for differences in fold change between metabolites which may be caused by large differences in the variation of the measured responses; however, the use of a scaling factor reduces such large differences to a relative value which is not dependant on the absolute abundance.