Metabolomics and partial least square discriminant analysis to predict history of myocardial infarction of self-claimed healthy subjects: validity and feasibility for clinical practice.

Nornazliya Mohamad,Rose Iszati Ismet,Manikandan Selvaraj,Thomas Hennessy,Kamarudzaman Md.Isa,Zakaria Bannur,Adzroolidzwan Ismail,Mohdsalleh Rofiee,Mohd Zaki Salleh,Thuhairahhasrah Abdul Rahman,Fadzilahmohd Nor,Aminuddin Ahmad,Lay Kek Teh

doi:10.1186/s13336-015-0018-4

Abstract

BackgroundThe dynamics of metabolomics in establishing a prediction model using partial least square discriminant analysis have enabled better disease diagnosis; with emphasis on early detection of diseases. We attempted to translate the metabolomics model to predict the health status of the Orang Asli community whom we have little information. The metabolite expressions of the healthy vs. diseased patients (cardiovascular) were compared. A metabotype model was developed and validated using partial least square discriminant analysis (PLSDA). Cardiovascular risks of the Orang Asli were predicted and confirmed by biochemistry profiles conducted concurrently.ResultsFourteen (14) metabolites were determined as potential biomarkers for cardiovascular risks with receiver operating characteristic of more than 0.7. They include 15S-HETE (AUC = 0.997) and phosphorylcholine (AUC = 0.995). Seven Orang Asli were clustered with the patients’ group and may have ongoing cardiovascular risks and problems. This is supported by biochemistry tests results that showed abnormalities in cholesterol, triglyceride, HDL and LDL levels.ConclusionsThe disease prediction model based on metabolites is a useful diagnostic alternative as compared to the current single biomarker assays. The former is believed to be more cost effective since a single sample run is able to provide a more comprehensive disease profile, whilst the latter require different types of sampling tubes and blood volumes.Electronic supplementary materialThe online version of this article (doi:10.1186/s13336-015-0018-4) contains supplementary material, which is available to authorized users.

Highlights

Cardiovascular disease (CVD) is known as the major cause of death worldwide [1]
Lipid measurements including the concentrations of high-density lipoprotein cholesterol (HDL) and low-density lipoprotein cholesterol (LDL) were routinely used as indicators to assess cardiovascular risks [3]
The subjects were classified into three groups: urban population of cardiovascular patients with history of myocardial infarction (MI; n = 31) and healthy volunteers (HT; n = 23), and self-claimed healthy Orang Asli (OA; n = 34)

Summary

Results

Metabolite profiles and biomarkers discovery A total of eighty-six (86) metabolites were annotated and identified from an average of 1658 spectra detected from LC/MS-QTOF analysis. Two (2) metabolites; 15S-HETE (AUC = 0.997) and phosphorylcholine (AUC = 0.995) with the highest potential of being the serum biomarkers for patients were identified Both metabolites were significantly up-regulated in the patients as compared to the healthy volunteers. The biomarkers were capable of discriminating patients from the healthy volunteers with the AUC of 0.998, 0.998 and 0.991, and average accuracy of 0.947, 0.961 and 0.963, respectively These results denoted that the fourteen (14) serum metabolites were potential classifiers between the patients and healthy volunteers. Metabotypes prediction of the Orang Asli A cross-validated PLSDA model with satisfactory discriminating ability was constructed using the fourteen (14) metabolites to assess the metabolic differences between the patients and healthy volunteers (Figure 4). Class of metabolite Metabolite p-value (p= 0.005) Log fold change AUC Sensitivity Specificity

Conclusions

Introduction

Materials and methods

C22 Sulfatide

Conclusion

14. Bellwood P