A significant challenge in treating some aggressive cancers and neurological diseases is the lack of understanding regarding the spatial-temporal molecular heterogeneity of the diseased tissue/organ. Molecular characteristics and interaction vary significantly across different spatial sub-units of the tissue/organ. The spatial pattern also changes over time as the disease progresses. The capability of mapping out the spatial-temporal patterns of molecular biomarkers is instrumental for elucidating the biological underpinning of disease formation, progression, and treatment resistance. This mapping, however, is extremely challenging, because it would require dense sampling of the diseased tissue/organ of a living person by highly invasive biopsy, which is infeasible. In reality, only a few samples can be taken, leaving a vast amount of unknown blank regions. On the other hand, recent years have witnessed the rapid advances of biomedical imaging technologies, which create structural and functional images of various modalities. Multimodality images can be taken non-invasively and for the entire diseased tissue/organ; importantly, they provide a complementary phenotypic presentation of underlying molecular characteristics. This creates an unprecedented opportunity to generate inverse-estimates of the underlying spatial-temporal molecular characteristics from the images to fill in the "blank regions." Once achieved, such inverse-estimates would help decipher the complex biological system of the diseased tissue/organ and inform new effective treatments with unparalleled precision adapted to spatial-temporal molecular heterogeneity.
The objective of this project is to develop a suite of new statistical models for inverse mapping/estimation of the spatial-temporal heterogeneity of molecular biomarkers from multimodality image phenotype. The investigators propose a novel modeling framework that integrates data-driven and biological-principle-driven mechanistic models, and meanwhile fuses global-scale image data and sparsely-sampled local biopsy measurements. This framework embraces modeling approaches to characterize both spatial heterogeneity and temporal dynamics of the disease. Furthermore, to account for patient similarity and specificity, the investigators propose a robust transfer learning model for integrating each patient?s data with information selectively transferred from other patients to avoid the negative transfer. Also, the project tackles joint modeling of a biomarker panel for characterizing spatial-temporal biomarker interaction. The proposed models will be validated in two applications: glioblastoma and Alzheimer's Disease. This project is expected to generate significant insight for unraveling the complex biological systems underlying these diseases and provide the groundwork for new treatment intervention. Additionally, the proposed modeling framework integrates statistical and bio-mechanistic models, which bridges two traditionally separate research fields together. The research team is committed to educating the next generation statisticians and biomedical researchers with cross-disciplinary skills, recruiting minority and women students, and disseminating research results in both statistical and bio/biomedical communities.
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.