Implementation of PCA enabled Support Vector Machine using cytokines to differentiate smokers versus nonsmokers

Area of Science:

• Biomarker discovery
• Translational medicine
• Computational biology

Background:

• Smoking is linked to severe diseases like COPD, cancer, and cardiac conditions.
• Cytokines play a role in inflammatory responses associated with smoking-related illnesses.
• Early diagnosis and intervention are crucial for managing smoking-related diseases.

Purpose of the Study:

• To investigate the association between elevated plasma cytokine levels and smoking status.
• To develop a machine learning model for differentiating smokers from nonsmokers using cytokine profiles.
• To identify key cytokine biomarkers for disease prognosis and diagnosis.

Main Methods:

• Applied Support Vector Machine (SVM) algorithm to analyze 65 plasma cytokines and traditional biomarkers.
• Utilized Principal Component Analysis (PCA), 10-fold cross-validation, and variable importance for optimization.
• Evaluated classification performance using Area Under the Receiver Operating Curve (AUROC).

Main Results:

• SVM achieved an AUROC of 89.2% (95% CI: 85.4%, 93.1%) in differentiating smokers and nonsmokers.
• Key cytokines identified include I-TAC, G-CSF-CSF-3, and MDC-CCL22.
• Optimizing with the top five cytokines improved AUROC to 93% (95% CI: 90.1%, 99.5%).

Conclusions:

• Machine learning, specifically SVM, effectively identifies smoking status based on plasma cytokine profiles.
• Selected cytokines serve as potent biomarkers for distinguishing smokers, aiding in early disease detection.
• These findings support the application of machine learning in translational and precision medicine for smoking-related diseases.