MUMA: a multi-omics meta-learning algorithm for data interpretation and classification.

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• Multi-omics data integration offers a comprehensive view of biological mechanisms.
• Challenges include data noise, heterogeneity, and high dimensionality, hindering accurate analysis.
• Existing methods struggle to extract meaningful insights without overfitting.

Purpose of the Study:

• To introduce a novel algorithm, the Multi-Omics Meta-learning Algorithm (MUMA), for robust multi-omics data integration.
• To enhance diagnostic performance and interpretability in analyzing complex biological datasets.
• To improve the extraction of biological information from noisy and high-dimensional omics data.

Main Methods:

• Developed MUMA, featuring self-adaptive sample weighting to handle noise.
• Incorporated interaction-based regularization to leverage relationships between omics modalities.
• Validated MUMA using simulations and eighteen real-world multi-omics datasets.

Main Results:

• MUMA demonstrated superior performance in classifying biological samples, including cancer subtypes.
• The algorithm effectively selected relevant biomarkers from noisy multi-omics data.
• Outperformed state-of-the-art methods in various multi-omics data analysis tasks.

Conclusions:

• MUMA provides a powerful and interpretable tool for multi-omics data integration.
• The algorithm facilitates a deeper understanding of biological systems and disease mechanisms.
• MUMA aids researchers in extracting reliable biological insights from complex omics data.