UBC Math Bio Seminar: Ian Wong

Animal tissues exhibit complex spatial architectures that emerge from cell-cell interactions. For instance, transitions between individual cells and multicellular groups occur widely in embryonic development, wound healing, and tumor progression. Although these collective behaviors are readily apparent to human observers, automated and unsupervised classification remains challenging. In this seminar, I will present recent results using topological data analysis for human interpretable machine learning of cellular architectures. First, we show that Wasserstein distances between persistence diagrams reveal distinct individual, clustered, and branching organization in epithelial cells. Second, we demonstrate the use of persistence images to classify spot and stripe patterning that occur due to self-sorting of two distinct cell types (“differential adhesion”). Finally, we consider topological cell shape generators trained using human neutrophils from healthy donors and septic patients. These results highlight the promise of topological approaches to elucidate the underlying structure of noisy and sparsely sampled biological datasets.