Deciphering systemic biological networks through AI-driven multi-omic integration
PI: Gustavo Nader (Kinesiology)
Tuition will be covered from other grants the PIs have.
Junior PI: Yogasudha Veturi, Assistant Professor of Biobehavioral Health and Statistics
Project description: We will integrate publicly available multi-organ, multi-omic dataset generated by the Molecular Transducers of Physical Activity Consortium (MoTrPAC). This project encompasses 9,466 assays across 19 organs (including muscle, liver, heart, lung, adipose tissue, blood, brain), 25 molecular platforms from rats exposed to an 8-week endurance training program. The datasets were generated at various time points to define temporal trajectories at various levels of organization including genomics, transcriptomics, proteomics, metabolomics, lipidomics, phosphoproteomics, acetylproteomics, ubiquitylproteomics, and immunomics. These unprecedented datasets will provide a unique opportunity to decipher key biological mechanisms and networks involved in physiological function at the whole organism level. We will integrate these data to investigate the molecular hierarchies that establish organ cross-talk and optimal organismal physiological adaptations and function. The emergent properties decoded by our analytical pipelines will inform novel system-level strategies for combating common conditions (e.g. diabetes, heart disease, aging) in humans. Notably, preliminary analyses indicate organ-specific sexually dimorphic responses, highlighting the importance of understanding these networks in a sex-specific manner.
Our computational approaches will include matrix factorization methods (e.g. MoFA, MEFISTO), neural network-based methods (e.g. DCCA, DeepMaps) and graph-based clustering methods (e.g., Seurat) to integrate the longitudinal multi-organ multi-omic data. AI-driven methods will combine vertical integration (e.g., integrating distinct sets of features like transcriptomics and epigenomics that aren’t directly comparable) and horizontal integration (focusing on a single modality like transcriptomics and combining data temporally) across each of the 19 organs. Thus, for any organ, we will develop data-driven models that can predict an omic dataset (layer) from the rest as well as identify a set of features (genes/proteins/metabolites) whose levels at later time points can be predicted from key features/networks enriched at earlier time points. This will reveal whether tissue-specific networks can predict organ crosstalk at a whole-organism level. Finally, we will also identify feature sets (i.e., biomarkers/networks) that yields best possible prediction for a trait of interest (e.g. blood glucose levels pre- and post-endurance training) and can illuminate biological and physiological functions characteristic of optimal health. Once our baseline networks or features are established, we can adapt this knowledge to humans to address which networks or features can restore a healthy state from a diseased state. Our approach can help identify novel biomarkers and, importantly, can be cross-referenced with drug-induced genomic responses to predict potential therapeutic interventions. Specifically, we can ask: Which drug or compound can restore or activate the appropriate networks or features in a disease-specific context?
Students: We are looking for students with expertise in statistical modeling, bioinformatics, as well as machine learning/AI-based algorithms for high-dimensional data analysis. The student should preferably be post-comps from Bioinformatics and Computational Biology, Molecular Cell and Integrative Biosciences, Statistics or Computer Science programs/departments. We will collect preliminary results demonstrating a snapshot of identified features/networks temporally for a small subset of omics/organs that we will be published in a scientific journal as well as be used towards extramural funding applications. We will meet as a group at Westgate W375 on Wednesday mornings at 9 AM.