Optimized Gaussian naïve Bayes for prostate cancer detection in White patients using gene expression data

Background Prostate cancer remains a major health concern worldwide. Racial disparities in prostate cancer diagnosis underscore the need for diagnostic approaches to manage the disease across different racial groups. Machine learning techniques, particularly Gaussian naïve Bayes, offers a promising method to improve diagnostic accuracy by identifying complex patterns in gene expression data. Gaussian naïve Bayes performs well with small datasets, minimizing the risk of overfitting. Method Gene expression and phenotype datasets were retrieved from the University of California, Santa Cruz (UCSC) Xenabrowser. Differentially expressed gene (DEG) analysis and receiver operating characteristic (ROC) evaluation were performed for race-specific populations. The resulting features were incorporated into multiple scenarios to develop machine learning models, combining feature selection, balancing techniques, and hyperparameter tuning within the Gaussian naïve Bayes algorithm to optimize prostate cancer diagnosis. Furthermore, Local Interpretable Model-Agnostic Explanations (LIME) was employed to interpret the impact of the selected features. Results The study evaluated 96 scenarios involving combinations of 4, 7, 13, and 139 gene features with balancing techniques and hyperparameter tuning to build machine learning models using the Gaussian naïve Bayes algorithm for prostate cancer detection. The highest accuracy for White race samples was achieved with 13 genes, resulting in 95% accuracy. LIME analysis identified the gene features with the greatest influence on model predictions. The model was further validated on Black race samples, achieving an accuracy of 93%. Conclusion The study demonstrates that Gaussian naïve Bayes, when combined with race-specific feature selection and hyperparameter tuning, can detect prostate cancer with high accuracy. These findings highlight the potential of Gaussian naïve Bayes for addressing prostate cancer detection in race-specific populations.