Interpretable neural network for risk stratification and drug target discovery based on PBMC transcriptomes.

Area of Science:

• Genomics
• Immunology
• Computational Biology
• Artificial Intelligence

Background:

• Peripheral blood mononuclear cells (PBMCs) are crucial for immune surveillance and disease development.
• PBMCs offer a minimally invasive method for molecular screening.
• Transcriptomic profiling of PBMCs provides insights into immune responses and disease states.

Purpose of the Study:

• To develop an intrinsically interpretable deep learning framework (ME-NET) for multi-disease screening and risk stratification using PBMC transcriptomic data.
• To enable biologically grounded interpretation of disease-associated immune programs by connecting predictions to pathway-level signals and contributing genes.
• To identify potential therapeutic repurposing candidates by integrating ME-NET-prioritized targets with drug-gene interaction resources.

Main Methods:

• Development of ME-NET, a deep learning framework incorporating pathway priors in a pathway-informed layer.
• Training and validation using harmonized PBMC transcriptomic datasets from heterogeneous clinical backgrounds.
• Comparison of ME-NET performance against standard machine-learning baselines.
• Integration of ME-NET-prioritized targets with drug-gene interaction databases.

Main Results:

• ME-NET demonstrated consistent performance across diverse datasets and outperformed standard machine-learning baselines.
• The pathway-informed layer enabled biologically grounded interpretation, linking predictions to specific pathways and genes.
• Identification of candidate compounds for therapeutic repurposing through integration with drug-gene interaction resources.

Conclusions:

• Interpretable deep learning applied to blood-derived transcriptomes is effective for scalable multi-disease screening.
• ME-NET facilitates the prioritization of biomarkers and drug targets for various diseases.
• This approach holds promise for advancing precision medicine and drug discovery.