Pathway reconstruction is a crucial group of algorithms that helps us make sense of complex cellular networks. Cells are essentially intricate networks where genes and proteins interact in highly coordinated ways. Pathway reconstruction describes the process of connecting genes or proteins of interes t— represented as nodes—by selecting the most relevant interactions between them, which we represent as edges in a protein-protein interaction network.
This capstone project will explore a Graph Neural Networks (GNNs) formulation for data-driven pathway reconstruction on Human Protein-Protein Interaction (PPI) Networks.
- Approach: Leverage GNNs for data-driven pathway reconstruction on Human Protein-Protein Interaction networks, focusing on edge selection to define pathways.
- Technical Framework: Formulate pathway reconstruction as an edge classification problem (Selected ∈ {0,1}) where node inclusion is determined automatically from selected edges.
Directory Structure:
- overall dataprep --> making union ppi, proc3essing pc bulk and individual pathways,
- gnn --> makes gnn node samples, create gnn data object, train and eval gnn
- oi2 --> make oi2 data, oi2 config file, pruning hyperparamters
- evaluation -->