PIKE-R2P

PIKE-R2P predicts protein abundance from single-cell RNA sequencing (scRNA-seq) data by embedding protein-protein interaction (PPI) networks into a graph neural network to model interdependent proteins in a multi-label framework.

Key Features:

• Multi-Label Prediction Framework: Formulates protein abundance prediction as a multi-label task that models interdependencies among multiple proteins at the single-cell level.
• Integration of Protein-Protein Interactions (PPI): Embeds prior knowledge from protein-protein interaction networks into the predictive model to capture relationships among proteins.
• Graph Neural Network Architecture: Uses a graph neural network to integrate PPI data with scRNA-seq expression information and learn complex biological network patterns.

Scientific Applications:

• Cross-Modality Prediction: Predicts protein abundances from RNA expression levels at single-cell resolution to bridge transcriptomic and proteomic modalities.
• Enhanced Predictive Performance: Leverages PPI integration to achieve smaller errors and higher correlations with gold-standard protein abundance measurements compared to methods that omit network priors.

Methodology:

Embeds protein-protein interaction networks into a graph neural network and formulates prediction as a multi-label problem that integrates scRNA-seq expression with PPI-derived context.