Heterogeneous multiple kernel learning for breast cancer outcome evaluation.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning in Oncology

Background:

• Breast cancer is a leading cancer affecting women, necessitating advanced detection and outcome prediction methods.
• Gene expression data analysis is crucial for predicting cancer prognosis.
• Kernel methods show promise for cancer outcome evaluation, but optimal kernel and parameter selection remain challenges.

Purpose of the Study:

• To develop and evaluate a novel heterogeneous multiple kernel learning (HMKL) algorithm for breast cancer outcome prediction.
• To address the challenge of selecting appropriate kernels and their parameters in cancer outcome evaluation.
• To demonstrate the efficacy of HMKL using real-world microarray datasets.

Main Methods:

• Utilized a heterogeneous kernel set including Hadamard, RBF, and linear kernels.
• Computed mixed kernel coefficients via quadratic programming.
• Integrated particle swarm optimization (PSO) within HMKL for kernel parameter selection.
• Applied the developed HMKL algorithm for breast cancer outcome evaluation.

Main Results:

• The HMKL method demonstrated superior performance compared to Random Forest, Decision Tree, GA with Rotation Forest, BFA+RF, SVM, and MKL.
• HMKL successfully selected optimal kernel functions and parameters.
• The Hadamard kernel was identified as effective within the HMKL framework.

Conclusions:

• HMKL provides an effective tool for breast cancer evaluation, aiding physicians in understanding patient conditions.
• HMKL offers a robust mechanism for automatic kernel function and parameter selection.
• The study validates the effectiveness of the Hadamard kernel in HMKL and suggests broader applicability of HMKL to real-world problems.