Integration of metabolomics, lipidomics and clinical data using a machine learning method

Abstract

Background: The recent pandemic of obesity and the metabolic syndrome (MetS) has led to the realisation that new drug targets are needed to either reduce obesity or the subsequent pathophysiological consequences associated with excess weight gain. Certain nuclear hormone receptors (NRs) play a pivotal role in lipid and carbohydrate metabolism and have been highlighted as potential treatments for obesity. This realisation started a search for NR agonists in order to understand and successfully treat MetS and associated conditions such as insulin resistance, dyslipidaemia, hypertension, hypertriglyceridemia, obesity and cardiovascular disease. The most studied NRs for treating metabolic diseases are the peroxisome proliferator-activated receptors (PPARs), PPAR-α, PPAR-γ, and PPAR-δ. However, prolonged PPAR treatment in animal models has led to adverse side effects including increased risk of a number of cancers, but how these receptors change metabolism long term in terms of pathology, despite many beneficial effects shorter term, is not fully understood. In the current study, changes in male Sprague Dawley rat liver caused by dietary treatment with a PPAR-pan (PPAR-α, -γ, and -δ) agonist were profiled by classical toxicology (clinical chemistry) and high throughput metabolomics and lipidomics approaches using mass spectrometry.

Results: In order to integrate an extensive set of nine different multivariate metabolic and lipidomics datasets with classical toxicological parameters we developed a hypotheses free, data driven machine learning approach. From the data analysis, we examined how the nine datasets were able to model dose and clinical chemistry results, with the different datasets having very different information content.

Conclusions: We found lipidomics (Direct Infusion-Mass Spectrometry) data the most predictive for different dose responses. In addition, associations with the metabolic and lipidomic data with aspartate amino transaminase (AST), a hepatic leakage enzyme to assess organ damage, and albumin, indicative of altered liver synthetic function, were established. Furthermore, by establishing correlations and network connections between eicosanoids, phospholipids and triacylglycerols, we provide evidence that these lipids function as a key link between inflammatory processes and intermediary metabolism.

Fig. 1

The design of the PPAR-pan agonist treatment study. a Four week old Sprague Dawley rats were acclimatised for approximately 3 weeks and were killed after a 13 week dose period. b Recovery animals were acclimatised for 3 weeks, dosed with the test compound for 13 weeks and kept for an additional 4 week dose free period. Before terminal kills urine and plasma samples were collected for urinalysis and clinical chemistry. After the terminal kills liver weight were recorded and liver samples were collected for metabolomic analysis

Fig. 2

Methodological work flow for data integration. Random forest (RF) classification was used to select subsets of metabolites from the combination of all metabolite data sets. Data sets which are the best in predicting the dose of PPAR-pan administered were assessed by calculating classification error values. The variables from the individual datasets were selected by a backward elimination approach, and the final set of metabolites were used for network analysis. As a separate strategic workflow, an RF regression approach was used to link liver metabolites with classical clinical chemistry parameters. Datasets which explain the variation of the classical clinical chemistry parameters were calculated, and individual variables were selected using permutation tests. Again, the final set of metabolites and the explained clinical chemistry parameters were selected for network analysis

Fig. 3

Random Forest (RF) classification approach for the determination of class error (how well the PPAR-pan dose level is predicted) and the selection of variables (which variables contribute to PPAR-pan dose level prediction) in each different dataset. a Class error of metabolomic and lipidomic dataset comparing values using the full set of variables and selected variables for calculations. b The number of variables contained within each dataset (in blue) and the number of metabolites after variable selection (in orange). For example, the number of total acyl-carnitines is 40 (in blue) and only four were selected (in orange)

Fig. 4

Network of selected metabolites. a A partial correlation network of the most discriminatory metabolites (12) differentiating between different doses of the PPAR-pan treatment concentrations. The solid lines denote positive and dotted lines denote negative correlations, and the thickness of the lines indicate the strength of the associations. b Biological pathways and their potential connections associated with the selected 12 metabolites. Note that metabolites of interest that were detected by our RF approach are color-coded on both pathway maps

Fig. 5

Variations explained by metabolomic and lipidomic datasets in relative liver weight and clinical chemistry parameters using the random forest (RF) regression approach. a Variation explained (Q²) in relative liver weight with and without variable selection. b The number of selected variables compared to the original (full dataset) number of variables. c Variation explained (Q²) with all the metabolomic and lipidomic data linking with phenotypes associated with plasma clinical chemistry from liver analysis. In total four parameters showed the largest variation explained across the different data sets

Fig. 6

A partial correlation network of the nine selected variables linking with the relative liver weight (ratio between the measured body weight and the measured liver weight of each animal). Different types of data are shown in different colours. The thickness of the lines relate to the extent of the correlation, where straight lines indicate positive and dotted lines indicate negative correlations

Fig. 7

a A correlation network of the 14 selected variables based on a Pearson correlation coefficient of more than 0.8 (r > 0.8) linking with the relative liver weights. Lipid mediators (lipids both shorter and longer than eicosanoids) are shown in green whereas phospholipids are shown in blue. b Partial correlation with 14 variables are shown