Differential Analysis

Volcano Plots

This combines fold change and statistical significance (e.g. p-values from ANOVA) to visualize and identify metabolites that exhibit both significant changes and substantial fold differences between species.

Examples:

  • “Bioinformatic Insight into Portulaca Oleracea L. (Purslane) of Bulgarian and Greek Origin,” Acta Biologica Cracoviensia s. Botanica, 2020, https://doi.org/10.24425/abcsb.2020.131662.
  • “Integrated metabolomics and transcriptome analysis on flavonoid biosynthesis in safflower (Carthamus tinctorius L.) under MeJA treatment” 2020

Peak Areas

This involves comparing individual variables (peaks) across species using techniques like t-tests, ANOVA, or non-parametric equivalents. It helps identify specific metabolites that significantly differ among species.

Examples:

PCA

Methods like Principal Component Analysis (PCA), Partial Least Squares Discriminant Analysis (PLS-DA), can be used for (1) visualising clusters of the data and (2) analyzing if there is a statistical difference between species based on the components using multivariate tests e.g. MANOVA.

For visualisations, https://plotly.com/python/pca-visualization/ is nice.

Examples:

  • Florence Souard et al., “Metabolomics Fingerprint of Coffee Species Determined by Untargeted-Profiling Study Using LC-HRMS,” Food Chemistry 245 ( April 2018): 603–12, https://doi.org/10.1016/j.foodchem.2017.10.022.
  • Aaron M. Goodpaster and Michael A. Kennedy, “Quantification and Statistical Significance Analysis of Group Separation in NMR-Based Metabonomics Studies,” Chemometrics and Intelligent Laboratory Systems 109, no. 2 (December 2011): 162–70, https://doi.org/10.1016/j.chemolab.2011.08.009.

(Hierarchical) Clustering

Clustering algorithms (e.g., Hierarchical Clustering, k-means) can reveal patterns or clusters in the data, highlighting differences or similarities among species based on multiple variables simultaneously.

Examples:

  • Florence Souard et al., “Metabolomics Fingerprint of Coffee Species Determined by Untargeted-Profiling Study Using LC-HRMS,” Food Chemistry 245 ( April 2018): 603–12, https://doi.org/10.1016/j.foodchem.2017.10.022.
  • Shaurya Chanana et al., “Hcapca: Automated Hierarchical Clustering and Principal Component Analysis of Large Metabolomic Datasets in R,” Metabolites 10, no. 7 (July 21, 2020): 297, https://doi.org/10.3390/metabo10070297.

Machine Learning Classification

We can also set up a model to predict species/regions based on the LCMS data, doing this can support that metabolite data can be used to distinguish the species.

Examples:

General Notes

Normalization

Careful preprocessing, including data normalization and transformation, is crucial to ensure accurate comparisons. Normalization methods like median scaling, log transformation, or autoscaling are commonly used in LC-MS data analysis.

Statistical Significance Correction

To control the error associated with multiple tests we need to a correction procedure. To control for the family-wise error rate, there are a few common methods. Bonferroni correction is often used (e.g. here and here), though this is a weak method. Holm-Bonferroni Correction is uniformly more powerful than the Bonferroni method, and Hochberg Correction is a still more powerful option if the p values are independent. Tukey’s HSD can be used when comparing means, and is used here, as well as in Compound Discoverer. These can be carried out in a python script e.g. see statsmodels multiple tests and statsmodels multiple comparisons.

Some analyses use False Discovery Rate correction, which are more powerful but less stringent than correcting for family-wise error.

Compound Discoverer (v 3.1), does some adjustments by default. In the differential analysis, the p-value for the sample group is calculated by running the Tukey HSD test (posthoc) after an analysis of variance (ANOVA) test. An ‘adjusted’ p value is also provided using the Benjamini-Hochberg algorithm to account for the false discovery rate. Note that in the volcano plots the former is used.