MAgentOmics

MAgentOmics employs a multi-agent architecture and an extended ant colony optimization algorithm to perform unsupervised feature selection and integrate multi-omics data for cancer prediction and molecular characterization.

Key Features:

• Multi-Agent Architecture: Uses a multi-agent system to consider and integrate diverse omics data types simultaneously for holistic feature selection.
• Extended Ant Colony Optimization Algorithm: Adapts ant colony optimization to iteratively construct and evaluate candidate feature subsets in high-dimensional multi-omics spaces.
• Innovative Fitness Function: Implements a novel fitness function that evaluates feature subsets without relying on external prediction targets such as patient survival time.
• Unsupervised Methodology: Operates without labeled outcomes to assess features based on intrinsic properties and inter-omics relationships.

Scientific Applications:

• Cancer Prediction: Integrates multi-omics data to support identification of molecular features relevant to cancer prediction and outcome analyses.
• Clinical Understanding and Decision-Making: Identifies cross-omics features to inform clinical interpretation of cancer molecular profiles and support oncology decision-making.

Methodology:

Evaluated on TCGA ovarian cancer multi-omics data from 176 patients using 5-fold cross-validation and compared against state-of-the-art supervised multi-view methods.