Dr. Ritambhara Singh

Spatial and temporal modeling of single-cell gene expression using deep learning

Our current understanding of the regulation of cells is akin to solving a jigsaw puzzle. Many genomic factors governing cell development have been identified, resulting in vast data collection efforts. For example, obtaining single-cell-level spatial DNA organization or gene expression measurements at a continuous time scale can reveal crucial regulatory patterns. However, technical constraints hinder our ability to collect all possible datasets, especially at a single-cell resolution. In this talk, I will present our attempt to build deep learning frameworks and leverage existing single-cell gene expression measurements to fill this data gap. First, I will discuss scGraphHiC, which performs graph deconvolution to extract genome-wide single-cell spatial interactions from bulk contact maps using single-cell gene expression as a guiding signal. Next, I will cover scNODE, which predicts single-cell gene expression at missing time points to enable cell developmental analysis. Our models allow researchers to predict spatial and temporal information of single-cell datasets for improved biological insights from existing experiments.