A Machine Learning Algorithm for Quantitatively Diagnosing Oxidative Stress Risks in Healthy Adult Individuals Based on Health Space Methodology: A Proof-of-Concept Study Using Korean Cross-Sectional Cohort Data

Figures

• 

  Figure 1. Schematic of the analysis pipeline. The eligible datasets were split into training and hold-out validation datasets. The least absolute shrinkage and selection operator (LASSO)/elastic net regression algorithm was trained using the training dataset and further evaluated in the hold-out validation dataset.

• 

  Figure 2. Regression coefficients for the features selected in 200 different 10-CV least absolute shrinkage and selection operator (LASSO)/elastic net regularized generalized linear models. The color scale beneath the heatmap represents a range of coefficient values from negative (blue) to red (positive). CV, cross-validation; RDW, red blood cell distribution width; LDL-C, low-density lipoprotein cholesterol; RFS, recommended food score; MDA, malondialdehyde; GPT, glutamate pyruvate transaminase; GGT, γ-glutamyl transferase; Hct, hematocrit; MCHC, mean corpuscular hemoglobin concentration; UA, uric acid; RBC, red blood cell count; WBC, white blood cell count; Hb, hemoglobin; BMI, body mass index; HbA1c, glycosylated hemoglobin.

• 

  Figure 3. Selection of the best performing model. (A) Cross-validation estimate of the mean squared prediction error for elastic net (left) and least absolute shrinkage and selection operator (LASSO, right), as a function of log λ. Grey bars are one-standard errors, and the vertical dashed lines are the location of the minimum misclassification error. Numbers on the top of the panel indicate the number of features selected. (B) Receiver operating characteristic curve for the elastic net and LASSO model on hold-out samples. The diagonal line represents the reference line of 0.5. AUC, area under the curve; CI, confidence interval.

• 

  Figure 4. Performance of the best performing machine learning model on two datasets. (A) Training dataset and (B) hold-out dataset. The diagonal line represents the reference line of 0.5. Light blue indicates 95% CIs of the ROC curve estimate. CI, confidence interval; ROC, receiver operating characteristic.

• 

  Figure 5. Application of the final model for qualitative diagnosis of oxidative stress risks. (A) Calibration of oxidative stress risks of subjects having zero (n = 379), one (n = 282), and two metabolic risk factors (n = 162) and oxidative disease cases (n = 505). (B) Application of oxidative stress risks of subjects with other diseases not related to oxidative stress (n = 386) in comparison to the healthy controls (n = 379) and oxidative disease cases (n = 505). The violin plot represents the distribution of data by using the kernel density function, and the width of the violin plot represents the sample size at this level. The red dot on each box plot represents the mean value. The box in the violin plot represents the median and quartile; the extension from the thin black line represents the 95% confidence interval. The different lowercase letters indicate a significant difference between the health status by Duncan’s test (p < 0.05).