This thesis will address biological problems that can be resolved through sequential model-based optimization. First, we propose protein fold family-regularization as a form of ML-assisted directed evolution. Next, we use sequential Bayesian optimization to automatically configure laboratory-based experimental designs. This includes developing synchronous and asynchronous parallel algorithms, as well using a cloud-based platform for experimental automation. Finally, we develop algorithms for automating vaccine design that can be used in-line with laboratory-based experimentation, and show how our optimization framework can augment a deep-learning approach in order to automatically generate candidate epitopes for vaccine inclusion.
Chris Langmead (CMU, Chair)
David Koes (PITT)
Russell Schwartz (CMU)
Jennifer Listgarten (UC Berkeley)
Zoom Participation. See announcement.