Graph Neural Networks in Precision Medicine: prediction of clinical outcomes in infectious diseases

• Grant number: 24/21755-4
• Support Opportunities: Scholarships in Brazil - Master
• Start date: May 01, 2025
• End date: December 31, 2026
• Field of knowledge: Biological Sciences - Immunology - Immunogenetics
• Principal Investigator: Helder Takashi Imoto Nakaya
• Grantee: Natalia Von Staa Mansur
• Host Institution: Faculdade de Ciências Farmacêuticas (FCF). Universidade de São Paulo (USP). São Paulo , SP, Brazil

Abstract

Understanding how gene expression influences the response to pathogens, such as Mycobacterium tuberculosis, can reveal new insights into pathogenesis, therapeutic targets, and diagnostic methods. Current diagnostic methods for tuberculosis have low sensitivity and fail to characterize different clinical presentations. Therefore, transcriptomic signatures are being sought to improve disease diagnosis. Machine learning (ML) techniques have been employed; however, data sparsity can hinder these methods, leading to overfitting and instability. Due to the heterogeneity of biological data, it is suggested that models considering biological relationships between genes are more effective for accurate predictions. This project aims to integrate transcriptomic data into networks and apply graph neural networks (GNN) to train a predictive model. It is expected that this approach, which considers gene relationships, will produce more robust and precise models. In addition to evaluating the model as a diagnostic tool, genes with the highest predictive relevance will be investigated. Systems biology methods will be used to generate insights into the pathophysiology of tuberculosis and the contribution of host response heterogeneity. Methodologies that predict prognosis and identify transcriptomic signatures associated with clinical conditions can support personalized treatment and precision medicine.