ARAX-MGKG2: Advanced Reasoning Agent with Multiomics, Molecular, Genetics, and RTX-KG2 Knowledge Provider Integration for Translational Research

Abstract The Biomedical Data Translator has already established itself as a powerful platform for integrating diverse biomedical data and enabling complex reasoning across a federated architecture. In the Performance Phase, we aim to elevate Translator to a new level of utility and impact by addressing key limitations and expanding its capabilities. Our overarching goal is to transform Translator into a highly versatile and user-centric tool, capable of delivering precise, actionable insights to support both basic and translational research. To achieve this, we will expand Translator's ability to handle a broader range of queries by incorporating new data types such as clinical trials, pharmacogenomics, multiomic, and metagenomic datasets. These enhancements will enable researchers to pose more sophisticated questions utilizing their own data in the context of Translator, uncovering connections and pathways that were previously inaccessible. In parallel, we will improve the transparency and provenance of Translator's outputs by developing new methodologies for evidence tracking and confidence assessment, ensuring that users can fully trust and understand the information they receive. We aim to closely collaborate with the Translator User Interface team to expose these new query and data types, as well as enhance the clarity and interpretability of evidence, provenance, and confidence associated with query results. High performance and scalability are also central to our vision. We will optimize our tools’ architecture to handle larger, more complex datasets, with improved speed and efficiency. This will involve refining knowledge graph integration, decreasing API response time and call frequency, enhancing data caching, and streamlining update processes, all of which will contribute to a more responsive and reliable system. Crucially, we will also expand user engagement, making Translator more accessible and valuable to a broader audience of researchers and clinicians. By implementing advanced user feedback mechanisms and developing new tools for user-supplied data integration, we aim to foster a dynamic, collaborative ecosystem where users can actively shape the future development of Translator. This project brings together an interdisciplinary team from Penn State University, the Institute for Systems Biology, Oregon State University, The Broad Institute, and Grenoble University. Each institution’s team contributes unique expertise, ranging from reasoning agent development and knowledge graph construction to user engagement and molecular data integration. This multi-institutional collaboration is crucial for addressing the complex challenges of expanding and optimizing the Translator system, ensuring that it continues to provide cutting-edge tools and insights for the biomedical research community.