Advancing computational modeling of disease metabolism by integrating AI and systems biology

Project summary Metabolic pathways are crucial for cellular energy and function. Dysregulated metabolism is a distinctive feature of various diseases, including cancer, diabetes, cardiovascular disease, inflammatory, and neurodegenerative diseases.

Given its integral role in disease pathology, accurately and comprehensively characterizing metabolic alterations offers immense potential for enhancing our understanding of disease biology, improving clinical diagnostics, prevention, and management. Specifically, such characterizations can lead to: (1) a deeper understanding of metabolic variation and

heterogeneity in disease tissue microenvironments; (2) the identification of novel drug targets or metabolic biomarkers for early diagnosis and treatment optimization; and (3) the development of tailored nutritional and dietary recommendations to enhance patient health quality. While significant progress has been made in studying metabolic

activities across species, a pivotal gap exists. On one hand, current analyses often fail to delineate the variations and heterogeneities of metabolic activities in highly altered disease microenvironment. Existing methodologies tend to present an averaged view of heterogeneous cell populations within tissues, overlooking the intricate metabolic

heterogeneity and exchanges that occur in complex tissues. This is especially concerning given that cells are known to adapt their metabolism in response to various biochemical conditions. On the other hand, compared to other areas like transcriptional regulation or immune response, there exists a marked gap in leveraging omics data to characterize

metabolic variations, necessitating tailored systems biology models and tools. In this MIRA project, we aim to address these essential gaps by overcoming four challenges. First, we seek a systematic and data-driven approach to characterize metabolic landscapes in disease systems, recognizing that metabolic pathways are multifaceted and

variations could span from flux to network topology. Second, we intend to harness the power of multi-omics data and extant knowledge to bridge gaps in metabolic modeling, aiming for a thorough elucidation of the metabolic fluxome and thereby achieving a holistic characterization of metabolic activities. Third, we aim to mechanistically decode the

functional roles of metabolic variations in diseases, emphasizing the heterogeneity and adaptability of metabolic phenotypes. Finally, by applying our systematic research framework into our -omics testbeds covering several disease types, we aim to pinpoint metabolic abnormalities, that could aid in precision medical interventions, and diet and

nutrient recommendations. Overall, our project promises to develop a suite of unparalleled computational tools: a data-driven research framework built upon a novel biophysics-informed neural network, dynamic models for in-depth metabolic analysis, cutting-edge statistical metrics, and a natural language processing tool to mine disease-specific

metabolic insights from literature. This is complemented by tools for specific analytical tasks, a curated omics testbed, and a publicly accessible knowledge base. Together, these innovations promise to revolutionize our understanding of disease metabolism, with validations planned on selected systems.