GCATSL

GCATSL predicts synthetic lethality (SL) interactions in human cancers to identify gene pairs relevant for targeted anticancer therapeutics.

Key Features:

• Graph-Based Representation Learning: Leverages multiple feature graphs constructed from diverse data sources to represent genes and capture complex biological relationships.
• Node-Level Attention Mechanism: Employs a node-level attention mechanism to capture both local and global dependencies among gene neighbors and learn nuanced node representations.
• Multi-Layer Perceptron (MLP): Uses an MLP to integrate original features with learned local and global node representations, producing feature-specific representations.
• Feature-Level Attention: Incorporates a feature-level attention mechanism to weight different feature graphs and derive final gene representations for SL interaction matrix reconstruction.

Scientific Applications:

• Benchmarking: Demonstrates superior performance compared to 14 state-of-the-art methods across three datasets under various settings.
• Novel SL Pair Identification: Identifies and prioritizes novel SL pairs to guide experimental screening of candidate interactions.
• Targeted Therapy Support: Supports selection of gene targets for targeted anticancer therapeutics by predicting SL relationships that enable selective cancer cell targeting.

Methodology:

Learns node representations from multiple feature graphs alongside a known SL interaction graph using node-level attention, integrates original and learned representations via an MLP to obtain feature-specific representations, and applies feature-level attention to reconstruct an SL interaction matrix for prediction; the architecture is presented as an alternative to matrix factorization-based methods that fail to capture neighbor dependencies or handle genes without known SL partners.