Bio-JARVIS: Development of a microservice based on generative AI for generating clinical interpretation texts of pathogens in metagenomics tests

• Grant number: 24/17790-9
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: January 01, 2025
• End date: December 31, 2025
• Field of knowledge: Engineering - Biomedical Engineering
• Principal Investigator: Deyvid Emanuel Amgarten
• Grantee: Gustavo Bezerra de Andrade
• Host Institution: Instituto Israelita de Ensino e Pesquisa Albert Einstein (IIEPAE). Sociedade Beneficente Israelita Brasileira Albert Einstein (SBIBAE). São Paulo , SP, Brazil

Abstract

Clinical metagenomics is an advanced approach that combines genetic sequencing and bioinformatics to identify pathogens directly from raw patient samples. Compared to traditional diagnostic methods, its key advantages include the elimination of the need for microorganism cultures (bacteria and fungi) or targeted molecular assays, such as PCR, for viruses. As metagenomic testing becomes increasingly prevalent in Brazil and globally, this technology is becoming ever more critical for decision-making in precision medicine. The present project aims to develop a generative artificial intelligence-based microservice, called Bio-JARVIS (Just an Artificial Reasoning and Very Interpretative System), with the objective of automating the generation of interpretative reports for metagenomic tests. This tool will support physicians and patients in understanding the identified pathogens by providing accessible and reliable clinical interpretations. To achieve this, we will evaluate leading generative AI providers (OpenAI, Google, and Amazon) to identify the most suitable solutions for implementing Bio-JARVIS. The microservice development will be carried out using the Django (Python) and Angular (TypeScript) frameworks, with rigorous validation through A/B testing with physicians and analysts. The primary success metrics will include accuracy, cost-efficiency, and acceptance by healthcare professionals. At the end of the project, the Bio-JARVIS source code will be made publicly available, with the expectation that this tool will become a practical and scalable solution, significantly improving the clinical interpretation of metagenomic tests.