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

Seema Singh Saharan ^1,2,3, Pankaj Nagar ¹, Kate Townsend Creasy ⁴

⁰Department of Statistics, University of Rajasthan, Jaipur, India.

Proceedings. International Conference on Computational Science and Computational Intelligence +

|

November 4, 2024

View abstract on PubMed

Summary

Machine learning identified key plasma cytokines that distinguish smokers from nonsmokers, improving early disease detection and enabling precision medicine interventions.

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.

Keywords:

Classification Cytokines Prediction Accuracy Principal Component Analysis Support Vector Machine