ppin-prediction

Process

• Building the graph
  • Get PPIN data from somewhere
  • Get amino acid sequence + GO:funcs for each protein in the graph
    • Find the embedder that takes them both as input
    • Create embeddings
  • Combine embeddings with the graph
• Graph Analysis
  • Centrality, density...
  • Look for clusters and what they indicate
  • t-SNE plot of the embeddings
• Building the model
  • Try GAE, VGAE, or any link prediction GNN model
  • Use a GNN explainer model to optimize for large computations
  • Validate the model
• Finalizing the project
  • Interface the code out to an API
    • Publish the package
  • Look into deploying the model
  • Create a story with the data using streamlit

Viable Data Sources

• STRING
• BioGRID
• IntAct - promising
• HINT (Human Interactome)
• HIPPIE (Human Integrated Protein-Protein Interaction rEference)
• Pathway Commons