NSF/BIO-UKRI/BBSRC: Integrative Deep Learning and Statistical Models for 3D Multimodal Analysis of Brain Structure

About This Grant

The brain is the most complex organ in the body, composed of different types of cells connected by intricate networks essential for sensory perception, cognition, and behavior. Understanding how these cells are organized, connected, and functionally coordinated within the brain remains a significant scientific challenge. Recent technological advances have produced large amounts of structural and functional data of the brain, yet extracting meaningful biological insights from these complex datasets remains difficult. This research addresses this critical gap by creating computational tools integrating molecular data with structural imaging, offering a detailed view of brain organization. Specifically, the project leverages datasets from advanced imaging methods revealing fine anatomical details with spatial transcriptomics, a technique profiling gene expression within tissues. By integrating these complementary data types using novel Bayesian and deep learning modeling frameworks, the project aims to build the first comprehensive 3-dimensional molecular and structural map of brain areas for olfaction in mice, an important model system for understanding neural circuits. This map will provide fundamental insights into how molecular characteristics of neurons relate to their structural connections, enhancing our understanding of how brain circuits’ function. Broader impacts include training a new generation of interdisciplinary scientists skilled in the intersection of artificial intelligence (AI), data science, and brain science. The project further introduces students to exciting careers in science, technology, engineering, and mathematics (STEM), and fosters greater public awareness of brain science research and its potential benefits for health and society. Technically, this project develops a novel computational framework combining Bayesian statistical modeling and Graph Neural Networks (GNNs) to integrate spatial transcriptomics with high-resolution structural data from X-ray nano-holotomography. The Bayesian component quantitatively models the spatial distribution and molecular identities of neuronal structures (e.g., glomeruli) in the olfactory bulb, accounting for biological variability and measurement uncertainty. The GNN-based approach dynamically integrates multimodal spatial data—molecular, morphological, and connectivity—to capture complex neuronal relationships, by explicitly incorporating uncertainty-aware learning aligned with the Bayesian framework. The methods employ efficient graph-sampling algorithms and multi-view contrastive learning to achieve scalability for analyzing large-scale, high-resolution brain datasets. Experimental validation involves direct integration with detailed molecular and anatomical datasets from the mouse olfactory bulb, ensuring biological accuracy and interpretability. Expected outcomes include open-source computational tools that significantly advance our ability to quantify and interpret complex biological data. The developed methods promise broad applicability in multi-modal spatial omics datasets beyond brain science, potentially transforming data analysis across biological and biomedical research contexts. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.