Comparative Analysis of POLB Mutations Across Cancer Types Using a Machine Learning-Driven Web Application

In this paper, we introduce a web application that is powered by machine learning and is specifically designed for the analysis of POLB gene mutations linked with different cancer types. The POLB gene encodes a vital enzyme in the base excision repair (BER) pathway, and any alterations in this gene’s sequence can cause genomic instability that might lead to tumor development. The application is fed with more than 1000 carefully selected POLB mutation samples from TCGA and COSMIC datasets, where every sample comes with various functional mutation scores like PolyPhen-2, SIFT, PROVEAN, and CADD, etc. To improve the predictive power of the system, it employs sophisticated preprocessing methods such as missing-value handling, IQR-based outlier removal, standardization of data using StandardScaler, class balancing through SMOTE, and creation of new mutation descriptors as the input data. The classifiers are evaluated in the context of a machine learning pipeline, where XGBoost—fine-tuned through GridSearchCV—reaches the maximum accuracy of 98%, which is the highest among all classifiers. The web application based on Flask provides a user-friendly interface for single and batch mutation risk prediction, interactive graph visualization, CSV file analysis, user authentication, and efficient error handling. The results obtained corroborate the capability of the system to operate on dynamic genomic data and to classify POLB mutations with great accuracy.