GPN WHPC Presentation: Joint Identification of Neuron Types and Type-Specific Activity-Regulated Genes with Coupled Autoencoders
Special day this month: Thursday at Noon CDT
Joint Identification of Neuron Types and Type-Specific Activity-Regulated Genes with Coupled Autoencoders by Dr. Yeganeh Marghi
Abstract: Recent advances in single-cell transcriptomics revealed an enormous diversity among neuronal cells. While the previous studies confirmed the existence of broad neuron classes, they also pointed towards a complicated landscape, where neuron types often appear to overlap or form gradients in gene expression. Therefore, a crucial step toward elucidating the neuronal identity is to jointly identify the discrete and the continuous factors of variability. Taking advantage of deep learning approaches, we study this problem in a variational framework by utilizing multiple interacting autoencoders, designed to disentangle the discrete and continuous aspects of neuronal diversity. We demonstrate the application of our method to a stand-alone single-cell RNA sequencing dataset, which defined over 100 transcriptomic neuron types in the mouse cortex. Our results suggest that the proposed method can refine the existing classifications of neurons by joint identification of discrete types and type-specific, activity-regulated genes.
Speaker Bio: Yeganeh Marghi received her Ph.D. in Electrical Engineering from Northeastern University where she specialized in machine learning and signal processing. Currently, Dr. Marghi is a Scientist at the Allen Institute for Brain Science, where her research focuses on cellular analysis and cell taxonomies using multimodal integration of electrophysiology and transcriptomic data. Her keynote talk during or WHPC June Monthly Meeting will address a joint identification of neuron types and type-specific activity-regulated genes with coupled autoencoders.