An Accurate Tool for Uncovering Cancer Subtypes by Fast Kernel Learning Method to Integrate Multiple Profile Data.

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Cancer remains a significant global health threat.
• Accurate subtyping of cancers is vital for personalized medicine and drug development.
• The Cancer Genome Atlas (TCGA) provides valuable multi-omics data for cancer research.

Purpose of the Study:

• To develop an accurate method for classifying lung and renal cancer subtypes.
• To leverage multi-omics data (gene expression, isoform expression, DNA methylation) for improved cancer classification.
• To investigate the effectiveness of Multiple Kernel Learning (MKL) for integrating diverse genomic datasets.

Main Methods:

• Data acquisition from The Cancer Genome Atlas (TCGA) for lung and renal cancers.
• Feature dimension reduction using Principal Component Analysis (PCA).
• Application of Multiple Kernel Learning (MKL) with various weighting methods (KTA, FKL, HSIC, Mean) for kernel fusion.
• Classification using Support Vector Machines (SVM) with combined kernels.

Main Results:

• Principal Component Analysis effectively reduced feature dimensionality and improved computational speed.
• Multiple Kernel Learning integrated multiple data types, enhancing classification accuracy.
• High classification accuracy achieved: 0.978 for renal cell carcinoma subtypes (MKL with HSIC) and 0.990 for lung cancer subtypes (MKL with FKL).

Conclusions:

• The proposed MKL-based approach integrating PCA and SVM is highly effective for cancer subtype classification.
• This method demonstrates significant potential for advancing personalized cancer treatment strategies.
• The findings underscore the utility of multi-omics data integration in oncology research.