Optimization of Smoking Classification by Applying Neural Network with Variable Importance Using Cytokine Biomarkers

Area of Science:

• Biomarker Research
• Machine Learning in Medicine
• Computational Biology

Background:

• Cigarette smoking is a major cause of preventable death, increasing risks for heart disease, stroke, and cancer.
• Smoking-induced endothelial dysfunction is linked to inflammatory cytokines, which can serve as predictive biomarkers.
• Advances in biomarker research and machine learning are crucial for precision diagnosis and therapeutics.

Purpose of the Study:

• To classify individuals as smokers or non-smokers using machine learning algorithms based on cytokine profiles.
• To identify the most impactful cytokine biomarkers for distinguishing between smokers and non-smokers.
• To evaluate the efficacy of a Neural Network model in smoking status classification.

Main Methods:

• Utilized a Neural Network (NN) algorithm to classify smokers versus non-smokers based on 63 distinct cytokines.
• Employed cross-validation and hyperparameter tuning to optimize NN performance.
• Identified the top 10 most influential cytokines for classification and compared model performance using all 63 versus the top 10 cytokines.

Main Results:

• The NN model using all 63 cytokines achieved an Area Under the Receiver Operating Characteristic curve (AUROC) of 0.949.
• A refined model using the top 10 cytokines demonstrated superior performance with an AUROC of 0.995.
• The 10 most impactful cytokines identified were I-TAC, IL-22, IL-2R, IL-3, HGF, IL-18, G-CSF-CSF-3, MIF, SDF-1alpha, and MMP-1.

Conclusions:

• Machine learning, particularly Neural Networks, effectively classifies smokers using cytokine profiles.
• Specific cytokines like I-TAC and IL-22 are highly predictive of smoking status.
• Cytokine biomarkers combined with machine learning hold significant potential for early disease prediction and novel treatment strategies.