Automated Machine Learning to Predict Interaction Between Human-Pathogen Proteins: Exploring Topological Information

• Grant number: 25/01309-2
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate (Direct)
• Start date: September 05, 2025
• End date: December 03, 2025
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: André Carlos Ponce de Leon Ferreira de Carvalho
• Grantee: Bruno Rafael Florentino
• Supervisor: Alexander Schonhuth
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Institution abroad: Bielefeld University, Germany
• Associated to the scholarship: 24/00830-8 - BioPrediction: democratizing machine learning in the study of molecular interactions, BP.DD

Abstract

With the advancement of modern techniques such as next-generation sequencing, the increasing amount of biological data available has posed challenges in extracting relevant molecular-level knowledge. One of the most significant challenges is predicting interactions between biological sequences such as DNA, RNA, and proteins that play crucial roles in complex processes like gene regulation and immune response, accelerating the study of diseases and therapies. Using Machine Learning (ML) algorithms in these problems has shown promising prospects. Considering this, we propose BioPrediction, an end-to-end framework built on automated ML, designed to uncover novel interactions between sequences without requiring specialized ML expertise. The goal is to develop an accessible classification model for researchers in the field of biological sciences, facilitating the application of ML in molecular biology problems. In this proposal, we are developing an enhanced version of BioPrediction to predict human-pathogen interactions. This version integrates both the structural information of the proteins involved and validated interaction networks, providing a richer foundation for decision-making. We are also using Graph Attention Networks (GAT) for robust topological features, linking the primary structure information of each protein with interaction network data to build new and resilient features. This novel approach aims to enhance prediction performance for human-pathogen protein interactions. Additionally, BioPrediction emphasizes interpretability, enabling researchers to comprehend the algorithm's decision-making process in contexts like metabolic network mapping and protein-protein interaction analysis. In summary, BioPrediction aims to boost research in molecular biology, contributing to the understanding of viral and bacterial infection processes and also to the development of therapies, such as the discovery of molecular targets for drugs.