Weighted K-means support vector machine for cancer prediction

Area of Science:

• Computational biology
• Machine learning in bioinformatics
• Statistical learning for medical applications

Background:

• Support Vector Machines (SVM) are established tools in biomedical research for disease subtype identification and genetic variant pathogenicity.
• Existing SVM methods may benefit from enhanced weighting strategies to improve predictive performance.

Purpose of the Study:

• To introduce novel weighted Support Vector Machine (SVM) algorithms: weighted K-means SVM (wKM-SVM) and weighted SVM (wSVM).
• To investigate the numerical relationship between SVM objective functions and imposed weights.
• To enhance predictive accuracy by integrating boosting algorithms with the proposed weighted SVM models.

Main Methods:

• Development of weighted K-means SVM (wKM-SVM) and weighted SVM (wSVM) by incorporating weights into the SVM loss term.
• Demonstration of numerical relationships between SVM objective functions and weighting parameters.
• Application of the boosting algorithm to the wKM-SVM and wSVM for ensemble learning.
• Validation through simulation studies and analysis of TCGA pan-cancer methylation data for breast and kidney cancer.

Main Results:

• The proposed weighted KM-SVM (wKM-SVM) and weighted SVM (wSVM) demonstrate superior predictive performance compared to standard KM-SVM and SVM.
• Boosting integration further enhances the accuracy of the weighted SVM models.
• The methods show effectiveness on both simulated datasets and large-scale real-world cancer methylation data.

Conclusions:

• Weighted KM-SVM (wKM-SVM) and wSVM significantly increase the accuracy of classification models in biomedical applications.
• These advanced SVM techniques can aid in disease diagnosis and inform clinical treatment decisions.
• A publicly available software package (wSVM) facilitates the adoption of these methods in research and practice.