Recursive ensemble feature selection provides a robust mRNA expression signature for myalgic encephalomyelitis/chronic fatigue syndrome

Abstract

Myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) is a chronic disorder characterized by disabling fatigue. Several studies have sought to identify diagnostic biomarkers, with varying results. Here, we innovate this process by combining both mRNA expression and DNA methylation data. We performed recursive ensemble feature selection (REFS) on publicly available mRNA expression data in peripheral blood mononuclear cells (PBMCs) of 93 ME/CFS patients and 25 healthy controls, and found a signature of 23 genes capable of distinguishing cases and controls. REFS highly outperformed other methods, with an AUC of 0.92. We validated the results on a different platform (AUC of 0.95) and in DNA methylation data obtained from four public studies on ME/CFS (99 patients and 50 controls), identifying 48 gene-associated CpGs that predicted disease status as well (AUC of 0.97). Finally, ten of the 23 genes could be interpreted in the context of the derailed immune system of ME/CFS.

Figure 1

Results of the REFS algorithm run ten times on the mRNA expression data of 118 samples from the CAMDA dataset. (a) The optimal number of predictor genes to distinguish 93 cases from 25 controls was 23 (red vertical line). (b) mRNA expression levels of the 23 predictor genes for 93 cases (light blue) and 25 controls (dark blue) in box-and-whisker plots. Outliers were omitted for visualization purposes.

Figure 2

ROC curves for (a) REFS, (b) IWGCNA, and (c) univariate analysis applied to the same mRNA expression data of 118 samples from the CAMDA dataset. (d) ROC curve for the validation of the obtained 23-gene signature on a separate dataset (GSE14577). The 5-fold cross validation was performed with the eighteen genes available in GSE14577.

Figure 3

Results of the REFS algorithm run ten times on the merged DNA methylation datasets restricted to 278 probes associated with the 23 candidate genes. (a) The optimal number of predictor CpGs to distinguish 99 cases from 50 controls was 48 (red vertical line). (b) ROC curve of the 48 predictor CpGs. c) DNA methylation levels (normalized using Standard scaler) of the 48 predictor CpG sites for 99 cases (light blue) and 50 controls (dark blue).

Figure 4

Visualization of the functions and locations of ten proteins in a hypothetical immune cell setting. All ten mRNA transcripts were downregulated in PBMCs of ME/CFS patients compared to healthy controls. Created with BioRender.com.

Figure 5

Overview of the REFS algorithm using the 10-fold global accuracy > 70% as a stop parameter.

Figure 6

Overview of the pipeline to reduce the number of methylation data probes for REFS by selecting CpGs associated with the candidate genes.