Deep learning-driven proteomics analysis for gene annotation in the renin-angiotensin system

Area of Science:

• Cardiovascular Biology
• Genomics
• Artificial Intelligence in Medicine

Background:

• The renin-angiotensin system (RAS) plays a crucial role in cardiovascular diseases, but many of its gene functions remain poorly understood.
• Understanding these functions is vital for developing effective treatments for conditions like hypertension and cardiomyopathy.

Purpose of the Study:

• To develop and validate a multi-label deep learning model for systematic annotation of RAS gene functions and their biological pathway involvement.
• To identify novel RAS components and interaction pathways relevant to cardiovascular disease pathogenesis.

Main Methods:

• Processed 39,463 RAS-related publications using TF-IDF and PCA for feature generation.
• Applied a Multi-Layer Perceptron (MLP) for multi-label classification, evaluating performance with Precision, F1-Score, Ranking Loss, and ROC-AUC.
• Validated in silico predictions using extracellular vesicle (EV) proteomics and capillary Western assays in a hypertensive mouse model.

Main Results:

• The deep learning model achieved high performance (Precision: 0.7474, ROC-AUC: 0.8697), outperforming traditional methods.
• Grouping genes into biological branches further improved model interpretability and performance (Precision: 0.8312, ROC-AUC: 0.9182).
• Identified AGTR2, IRAP, Ywhas, EDNRA, and ESR2 as critical RAS genes, with IRAP significantly upregulated in hypertension and interacting with key signaling molecules.

Conclusions:

• This study presents the first integration of multi-label AI with EV proteomics for RAS pathway annotation, offering a powerful framework for understanding complex gene-pathway relationships.
• Revealed a novel IRAP/Ywha(s)/Nedd4-2-ACE2 interaction axis, highlighting a potential therapeutic target for hypertension and related cardiovascular diseases.