Prediction of Smoking Behavior From Single Nucleotide Polymorphisms With Machine Learning Approaches

Area of Science:

• Genetics
• Computational Biology
• Behavioral Science

Background:

• Smoking dependence has a high heritability, up to 50%, suggesting a strong genetic component.
• Identifying genetic susceptibility variants for smoking is crucial for developing targeted prevention strategies.
• Previous genetic studies have shown limited predictive power for smoking behavior.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting smoking behavior based on genomic profiles.
• To assess the efficacy of Support Vector Machine (SVM) and Random Forest (RF) algorithms in predicting smoking dependence.
• To identify the most effective feature selection methods for building accurate predictive models.

Main Methods:

• Applied machine learning algorithms, including SVM and RF, for predictive modeling of smoking behavior.
• Utilized a 10-fold cross-validation approach on a dataset of African ancestry individuals (1,431 smokers, 1,503 non-smokers).
• Employed logistic regression and LASSO regression for feature selection, identifying top single nucleotide polymorphisms (SNPs).
• Validated models on an independent dataset (213 smokers, 224 non-smokers).

Main Results:

• The SVM model, using 500 top SNPs selected by logistic regression, achieved an Area Under the Curve (AUC) of 0.720 on the independent test set.
• The RF model with 500 top SNPs selected by logistic regression achieved an AUC of 0.667 on the independent test set.
• A combined feature selection approach (logistic and LASSO regression) with SVM yielded the highest AUC of 0.897 on the independent test set.

Conclusions:

• Machine learning methods, particularly SVM, demonstrate significant potential for building accurate predictive models of smoking behavior.
• Genomic profiling combined with advanced computational approaches offers a promising avenue for personalized smoking dependence prevention.
• Further research is warranted to refine these models and translate findings into clinical applications for smoking cessation and prevention.