In pediatric acute leukemias, relapse is the major cause of mortality. Unfortunately, this occurs in 20-40% of patients, but methods to identify those at higher risk of relapse are lacking, and treatment options for such patients remain limited. To address this need, Kara Davis is generating large-scale datasets from leukemia samples and employing deep neural networks to improve relapse prediction. By using clinically annotated primary patient samples, Davis and her team will build a single-cell atlas of leukemia cells by applying a combination of mass cytometry and single-cell sequencing. They will then use deep learning models and neural networks to detect single-cell phenotypes predictive of relapse in the context of hematopoietic development.
Neural network deep learning models are widely used in image analysis, marketing and social networking. Their application to clinical medicine is in its infancy. However, with the growing application of single-cell, high-parameter technologies in patients and electronic medical records, the possibility exists to leverage neural networks to approach these large datasets for diagnostic, prognostic and therapeutic decision making. The use of neural network models in classifying healthy and leukemic cell types and further determining relapse risk based on their classification has never been done. Establishing these tools will be essential for improving patient survival and guiding standards of care as treatment evolves to incorporate more personalized medicine approaches in oncology.