Abstract

Large, high capacity, deep learning algorithms and models, trained on large amounts of data have shown to achieve impressive performance and generalize well. As newer, cheaper types of data querying biological systems are currently more available than ever, the popularity of deep learning in computational biology has increased. However, with deeper models comes greater responsibility. Is the complexity of such models warranted and can they bring new insight into scientific decision-making? In this talk, I will survey a number of challenges in computational biology, with a focus on the problem of interpretable feature selection in spatial single cell RNA sequencing data. We will discuss linear approaches to select relevant features using lasso, and extensions to the context of deep learning using variational autoencoders and the famous gumbel softmax trick. This is joint work with Nabeel Sarwar and Soledad ViIllar.