COSNet

COSNet performs graph-based classification on biological networks using a Hopfield Network–based learning framework optimized for highly imbalanced class distributions.

Key Features:

• Graph-Based Classification: Analyzes biological networks where nodes represent biological entities and edges encode functional or physical relationships.
• Imbalanced Learning Optimization: Addresses severe class imbalance common in biological network classification tasks.
• Parametric Hopfield Network Model: Implements a modified Hopfield Network in which neuron activation thresholds and output levels are treated as learnable parameters.
• Label Propagation Dynamics: Propagates label information across network structures through iterative neural network dynamics.
• Scalable Computation: Operates with quasi-linear time complexity for efficient analysis of large-scale biological graphs.

Scientific Applications:

• Disease Gene Prioritization: Identifies candidate genes associated with diseases using biological network data.
• Functional Gene Annotation: Predicts gene functions by propagating functional labels across biological networks.
• Protein–Protein Interaction Network Analysis: Supports classification tasks within protein interaction networks.

Methodology:

COSNet applies a parametric Hopfield Network model to biological graphs, learning neuron activation thresholds and output levels to propagate label information across nodes while accounting for highly imbalanced class distributions.