Fig. 2

Partial least squares-discriminant analysis for differentially expressed lipids between patients with and without severe OSA. A Individual loading for the two first components. B The rank of variable importance of differentially expressed lipids on the two first components. Lipid features included in the final model. Feature selection was based on a combination of random forest and automated model selection using the Akaike Information Criterion. B Receiver operating characteristic curves for predicting severe OSA using the lipid signature. The AUC (95% CI) for the model is shown. OxCer, oxidized ceramide; OxTG, oxidized triglyceride; mass at RT, mass at retention time; OR, odds ratio