Abstract:
|
Single cell sequencing has enabled high throughput quantification of gene expression and other features at the level of individual cells, enabling the study of cell type diversity in tissue during homeostasis and disease. I will describe deep learning methods for analysis of single cell data, in particular for data denoising and multimodal imputation. Data denoising addresses the sparsity of single cell RNA sequencing data, where only a small proportion of the transcripts in each cell are sequenced. We propose a transfer learning framework to learn from related single cell data sets to improve expression estimates, allowing us to leverage the expanding resources of publicly available data such as the Human Cell Atlas. Building on this denoising framework, I will describe the Single Cell Transcriptome to Proteome Prediction Deep Neural Network (cTP-Net) and its extensions for multi-omic imputation. I discuss the limits of data pooling in single cell genomics: What can be learned across cell types, tissues, and species? How useful are data from other domains in improving the estimates from your own study?
|