A kernel-based integration of genome-wide data for clinical decision support

Area of Science:

• Bioinformatics
• Computational Biology
• Translational Oncology

Background:

• Transcriptomic data alone is insufficient for fully understanding tumor biology due to factors like alternative splicing and post-translational modifications.
• Integrating multiple genome-wide data sources (genome, transcriptome, proteome, epigenome) is crucial for a comprehensive biological understanding.
• The growing volume of omics data necessitates a robust methodological integration framework.

Purpose of the Study:

• To develop and validate a kernel-based framework for integrating multiple genome-wide data sources for clinical decision support.
• To enhance the predictive performance of cancer outcome prediction models by fusing diverse omics data.
• To apply the framework to rectal and prostate cancer datasets for predicting multiple clinical outcomes.

Main Methods:

• A kernel-based approach was employed, integrating data at the kernel matrix level within the patient domain.
• A weighted least squares support vector machine was utilized as the supervised classification algorithm.
• The framework was tested on rectal cancer (microarray, proteomics) and prostate cancer (microarray, genomics) datasets.

Main Results:

• For rectal cancer, combining microarray and proteomics data yielded the highest leave-one-out AUC values (0.927–0.987).
• For prostate cancer, integrating microarray and genomics data improved prediction for all four outcomes, with AUCs ranging from 0.786 to 0.987.
• In both cancer types, multi-modal data integration consistently outperformed single-data source predictions.

Conclusions:

• Integrating multiple genome-wide data sets significantly enhances the predictive accuracy of clinical decision support models for cancer.
• The findings underscore the importance of comprehensive multi-modal data for advancing personalized cancer therapy.
• While initial investment in multi-modal data is substantial, it is essential for developing cost-efficient, tailored treatment strategies.