Multivariate analysis and model building for classifying patients in the peroxisomal disorders X-linked adrenoleukodystrophy and Zellweger syndrome in Chinese pediatric patients

Abstract

Background: The peroxisome is a ubiquitous single membrane-enclosed organelle with an important metabolic role. Peroxisomal disorders represent a class of medical conditions caused by deficiencies in peroxisome function and are segmented into enzyme-and-transporter defects (defects in single peroxisomal proteins) and peroxisome biogenesis disorders (defects in the peroxin proteins, critical for normal peroxisome assembly and biogenesis). In this study, we employed multivariate supervised and non-supervised statistical methods and utilized mass spectrometry data of neurological patients, peroxisomal disorder patients (X-linked adrenoleukodystrophy and Zellweger syndrome), and healthy controls to analyze the role of common metabolites in peroxisomal disorders, to develop and refine a classification models of X-linked adrenoleukodystrophy and Zellweger syndrome, and to explore analytes with utility in rapid screening and diagnostics.

Results: T-SNE, PCA, and (sparse) PLS-DA, operated on mass spectrometry data of patients and healthy controls were utilized in this study. The performance of exploratory PLS-DA models was assessed to determine a suitable number of latent components and variables to retain for sparse PLS-DA models. Reduced-features (sparse) PLS-DA models achieved excellent classification performance of X-linked adrenoleukodystrophy and Zellweger syndrome patients.

Conclusions: Our study demonstrated metabolic differences between healthy controls, neurological patients, and peroxisomal disorder (X-linked adrenoleukodystrophy and Zellweger syndrome) patients, refined classification models and showed the potential utility of hexacosanoylcarnitine (C26:0-carnitine) as a screening analyte for Chinese patients in the context of a multivariate discriminant model predictive of peroxisomal disorders.

Keywords: C26: carnitine; Hexacosanoylcarnitine; Metabolomic signature; Newborn screening; PCA; PLS-DA; Sparse PLS-DA; Very long chain fatty acids; X-ALD; X-linked adrenoleukodystrophy; Zellweger syndrome; t-SNE.

Fig. 1

Unsupervised multivariate analysis of patient targeted metabolite panel data. (a) t-SNE visualization of 16-features in patient samples for the ZS (yellow), X-ALD (dark blue), DDE (red), and control (black) groups; (b,c) PCA analysis of targeted metabolite panel data in healthy controls, non-PD neurological patients (DDE), X-linked adrenoleukodystrophy (X-ALD), and Zellweger syndrome (ZS) patients. Score plots along the 3 principal components are shown.

Fig. 2

Assessing exploratory PLS-DA model performance and evaluating latent component and features to retain for sparse PLS-DA modeling. (a, c, e) Assessing PLS-DA model performance in the 4-class setting (Control vs. DDE vs. X-ALD vs. ZS) and the 2-class setting (X-ALD vs. ZDC; ZS vs. XDC) and selection of distance metric and number of latent components. Repeated stratified cross-validation (100  ×  5–fold CV) is used to evaluate the PLS-DA classification performance (measured by balanced error rate) for each prediction distance (max.dist, centroids.dist, and mahalanobis.dist). The balanced error rate appears to decrease negligibly after four latent components in the 4-class setting, and the balanced error rate reaches minimal value in 2-class setting with 1 latent component. (b) Cross-validation and error evaluation of the PLS-DA model in 4-class setting with 4 latent components and all 16 features. Optimal, error minimizing set of features per component are indicated with a diamond. Yellow diamond points to a 3-latent component model with 1, 15, and 1 retained feature(s) per latent components LC1, LC2, and LC3 respectively. (d) Cross-validation and error evaluation of the PLS-DA model in X-ALD vs. ZDC 2-class setting with 1 latent component and all 16 features. blue diamond points to a 1 latent component model with 8 retained features. (f) Cross-validation and error evaluation of the PLS-DA model in ZS vs. XDC 2-class setting with 1 latent component and all 16 features. Blue diamond points to a 1 latent component model with 15 retained features

Fig. 3

Sparse PLS-DA model 3LC-1-15-1 (4-class setting). (a1-2): Sample plots of the targeted metabolite panel data after a parsimonious PLS-DA model was operated on the data, depicting the patient samples with the confidence ellipses for the class labels. (b1-3) One-vs.-Others ROC curves assessing the classification performance of the PLS-DA model with 3 latent components and 1, 15, and 1 feature(s) per component (c). Feature stability per component evaluation in 5-fold – 100x cross validation

Fig. 4

Sparse PLS-DA models in the 2-class settings. (a). ROC curve assessing the classification performance of the PLS-DA model with 1 latent components and 8 features in the X-ALD vs. ZDC 2-class setting. (b). Feature stability evaluation in 5-fold – 100x cross validation in the X-ALD vs. ZDC 2-class setting. (c). ROC curve assessing the classification performance of the sparse PLS-DA model with 1 latent component and 15 features in the ZS vs. XDC 2-class setting. (d). Feature stability per in 5-fold – 100x cross validation in the ZS vs. XDC 2-class setting

Fig. 5

PLS-DA VIP model in the 4-class setting. (a1-2) Sample plots after the PLS-DA model with the VIP features was operated on the data, depicting the patient samples with the confidence ellipses for the class labels (b1-3) One-vs.-Others ROC curves assessing the classification performance of the PLS-DA model with 3 Latent Components and VIP features

Fig. 6

PLS-DA VIP model in 2-class settings. (a) ROC curve assessing the classification performance of the PLS-DA model with 1 Latent Components and VIP features in the X-ALD vs. ZDC 2-class setting (b) ROC curve assessing the classification performance of the PLS-DA model with 1 Latent Components and VIP features in the ZS vs. XDC 2-class setting