Endocrine-mediated pubertal brain network development: Bridging datasets with machine learning - PROJECT SUMMARY This proposal aims to address the BRAIN Initiative goals of developing and applying technologies for innovative study of the relationships between biological processes, neural structure, and brain function, to further our understanding of how sex hormones (e.g., estradiol, testosterone, progesterone) exert organizing effects on brain development during puberty. Animal and postmortem human research has revealed that these effects include cytoarchitectonic changes, including apoptosis and changes to dendrite structure, but technology has limited in vivo study in humans. While neuroimaging has provided insight into associations between sex hormone levels during puberty and the brain’s macro-scale and functional architecture, this literature overlooks potential neurobiological mechanisms (e.g., cytoarchitecture changes) and the impact of phasic versus tonic (i.e., annual versus weekly) hormone changes on brain development. The proposed human neuroimaging project addresses these gaps by integrating estimates of cytoarchitecture from advanced biophysical modeling of diffusion-weighted imaging, to complement commonly used macro-scale and functional architecture measures, and by assessing both sparsely and densely sampled hormone levels using three independent studies of youth at various stages of pubertal maturation. Furthermore, the proposed research applies machine learning to facilitate multi-dataset integration and investigation of how hormone levels are related to correspondence between the brain’s structural and functional architectures. The candidate has a background in magnetic resonance imaging, including with developmental populations, is experienced in developing and applying data-driven tools for studying large-scale brain networks, and seeks further training in neuroendocrinology, machine learning, and diffusion-weighted imaging to become an independent researcher in this field. During the mentored phase, she will (1) develop and train a model that predicts sex hormone levels from functional connectomics in a large adolescent dataset, test its accuracy predicting the same and novel hormones in an independent adolescent dataset, and share the pre-trained model for use by the broader research community and (2) assess roles of sex hormone levels during puberty in developing structure- function associations using multivariate, cross-decomposition approaches. Training from leaders in adolescent neuroendocrinology, machine learning, multi-dataset analysis, and diffusion-weighted imaging will aid in this work and prepare her for the independent phase, when she will (3) apply the pretrained model to predict (a) average hormone levels, to further assess the generalizability of identified sex hormone-related brain networks, and (b) weekly sex hormone levels, to assess the role of phasic hormone changes on the identification of tonic neurobiological associations. Then, she will (4) apply further cross-decomposition approaches to model the role of hormones in associations between the brain’s functional architecture and each macro- and cytoarchitecture across pubertal development in three independent datasets.
