Multi-agent Feature Selection for Integrative Multi-omics Analysis.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Multi-omics data integration is crucial for understanding complex diseases like cancer.
• High-dimensionality of multi-omics data poses challenges for accurate cancer prediction.
• Existing feature selection methods often analyze omics data types independently.

Purpose of the Study:

• To develop a novel multi-agent architecture, MAgentOmics, for integrated multi-omics feature selection.
• To address the curse of dimensionality in cancer prediction using an unsupervised approach.
• To improve the discovery of molecular mechanisms underlying cancer through enhanced feature selection.

Main Methods:

• MAgentOmics extends ant colony optimization for multi-omics data integration.
• An unsupervised fitness function evaluates feature subsets without prediction targets.
• The method was evaluated on TCGA ovarian cancer multi-omics data using 5-fold cross-validation.

Main Results:

• MAgentOmics demonstrated superior integration power compared to state-of-the-art supervised multi-view methods.
• The unsupervised approach effectively identifies relevant features from high-dimensional multi-omics datasets.
• The proposed method shows promise for advancing cancer subtyping and biomarker discovery.

Conclusions:

• MAgentOmics offers an effective unsupervised strategy for multi-omics feature selection in cancer research.
• Integrated feature selection enhances clinical understanding and accelerates decision-making.
• The publicly available code facilitates further research in multi-omics data analysis.