Neurobiology and Neurooncology are challenging fields of study due to the complexity of the brain as an organ as well as the difficulty in attaining human samples of brain tissue. These challenges have increasingly been addressed by the use of mouse models and molecular profiling that have uncovered complex sets of cell types. Fundamentally all cell types contain the same DNA but can be stratified by their RNA profile. This approach has taken advantage of single cell RNA sequencing (scRNA-Seq) which allows for single cells to be measured individually and for many cell types to be uncovered. In the past 5 years these experiments have become ever more prolific with large numbers of labeled datasets of single cells. These datasets can be used to explore a variety of questions involving the stratification of cell types to the distribution of these cell types in tissue. One of the most promising areas is the use of these cell types to better understand tumor heterogeneity and response to treatment. A promising avenue to leverage these diverse datasets for the study of heterogeneity is Transfer Learning. Transfer Learning is the subfield of Machine Learning that applies information learned from source data to target data to generate more generalizable or accurate models. We will develop methods to combine datasets into more robust models of tissue heterogeneity using transfer learning in both normal and Glioblastoma brain. These models can be used to predict clinical outcomes and study unique cellular relationships between datasets.
The brain is a very complex organ with a wide variety of cell types -- the more granular of which have just recently been uncovered. We develop transfer-learning methods to integrate multiple datasets across species and disease states to glean information that will improve our understanding of the brain at the cellular level. We reframe Glioblastoma's genetic associations to survival as mixtures of quantifiable cell types effectively eliminating the ?curse of dimensionality? via semi-supervised transfer-learning.