The project develops a novel model selection method to automate choice of deep neural network architectures for effective integration of diverse biological data to predict their functional interactions.
Specifically, we design principled model selection methods using Bayesian nonparametrics to infer the most plausible architectures of deep neural networks given the data, and develop efficient techniques to make the use of the inference methods computationally tractable. The project develops deployable computational tools based on the proposed techniques to simulate functional aspects of genes and proteins (e.g., physical or genetic interactions).