MOKAN: A Multi-Omics Data Analysis Framework Using Kolmogorov-Arnold Networks.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-throughput sequencing generates vast multi-omics data crucial for understanding cancer.
• Existing multi-omics integration methods struggle with feature engineering and data heterogeneity.
• A robust framework is needed to comprehensively analyze cancer through integrated multi-omics data.

Purpose of the Study:

• To propose MOKAN, a novel multi-omics integration framework based on Kolmogorov-Arnold Networks (KAN).
• To address limitations in existing methods by capturing data heterogeneity and enabling complementary data analysis.
• To enhance cancer classification and biomarker identification through effective multi-omics data integration.

Main Methods:

• Developed MOKAN, a framework utilizing Kolmogorov-Arnold Networks (KAN) for multi-omics data integration.
• Employed a sample-weighted random sampler to balance class distribution during training.
• Leveraged KAN's learnable activation functions and flexible structure to capture diverse feature spaces.
• Utilized KAN's decomposition property to create and integrate low-dimensional subspace representations.

Main Results:

• MOKAN effectively integrates heterogeneous multi-omics data, preserving individual omics contributions.
• The framework successfully breaks down high-dimensional data into manageable subspaces for analysis.
• Experimental results show MOKAN outperforms existing methods in cancer classification tasks.

Conclusions:

• MOKAN provides a powerful and flexible approach for multi-omics data integration.
• The framework enhances the understanding of cancer by leveraging the complementary nature of different molecular layers.
• MOKAN represents a significant advancement in computational cancer research and analysis.