SLMGAE

SLMGAE predicts synthetic lethal interactions in human cancers using a multi-view graph auto-encoder framework to integrate SL graphs, protein-protein interaction (PPI) data, and Gene Ontology (GO) information for improved SL prediction.

Key Features:

• Synthetic lethal interaction prediction: Predicts synthetic lethal (SL) gene pairs in human cancers.
• Graph-based representation: Represents genes as nodes and SL interactions as edges to model genetic relationships.
• Multi-view Graph Auto-Encoders (GAEs): Uses multiple GAEs to process distinct network views and reconstruct their graphs.
• Support views: Incorporates a primary SL graph view and additional views derived from protein-protein interactions (PPI) and Gene Ontology (GO).
• Attention mechanism: Applies an attention mechanism to dynamically weight and integrate information from different support views.
• Training objectives: Trains the model by minimizing reconstruction error of each view and prediction error for SL interactions.
• Evaluation dataset: Empirically evaluated using the SynLethDB dataset and reported to outperform existing methods.
• Case studies: Includes case studies of novel predicted SLs to demonstrate potential biological relevance.

Scientific Applications:

• Targeted anticancer therapy discovery: Supports identification of gene pairs exploitable as targets for targeted cancer therapies.
• Prioritization for experimental validation: Ranks novel SL candidates to guide experimental follow-up and validation.
• Reducing experimental burden: Provides computational predictions intended to reduce cost and variability of experimental SL screens across cell lines.

Methodology:

Constructs graphs with genes as nodes and SL edges, derives additional views from PPI and GO, processes each view with a separate Graph Auto-Encoder to reconstruct graphs, uses an attention mechanism to weight views, and trains by minimizing reconstruction and prediction errors with empirical evaluation on SynLethDB.