CIPHER-SC

CIPHER-SC infers disease–gene associations by integrating single-cell transcriptome data into a context-aware graph convolutional end-to-end network to enable cell-type-specific prediction.

Key Features:

• Context-aware network architecture: Constructs a network that incorporates contextual information to integrate diverse biological data sources without introducing bias.
• Graph convolution-based end-to-end learning: Uses graph convolution in an end-to-end training framework to minimize errors associated with traditional multi-stage training.
• Single-cell transcriptome integration: Incorporates single-cell transcriptome data to capture cell-type-specific information and enable higher-resolution study of gene function.
• Unbiased data integration and improved feature selection: Reduces error accumulation and enhances feature selection accuracy through its integrated architecture.
• Performance evaluation: Demonstrates superior performance relative to four state-of-the-art approaches, achieving an AUC of 0.9501 across five-fold cross-validations and three distinct test sets.
• Ablation study results: Shows that the complete end-to-end design and unbiased integration increase AUC from 0.8727 to 0.9443 and that inclusion of single-cell data enriches predictions for cell-type-specific genes.

Scientific Applications:

• Disease–gene association prediction: Prioritizes and predicts disease-associated genes, including novel genes and genes with a genetic basis.
• Cell-type-specific gene function analysis: Enables analysis of gene functions at higher resolution by leveraging single-cell transcriptome data.
• Method benchmarking and validation: Supports rigorous benchmarking using five-fold cross-validation and evaluation on three distinct test sets.

Methodology:

Construct a context-aware network that integrates single-cell transcriptome data, apply graph convolution within an end-to-end learning framework to reduce multi-stage training errors, and evaluate performance via five-fold cross-validations, three distinct test sets, and an ablation study.