Single Cell Data Analysis Algorithms

National Institute Of Diabetes And Digestive And Kidney Diseases

Investigators

Abstract

With advances in single-cell techniques, collecting a large quantity of data has become more accessible and efficient. In contrast, the increased complexity of data has made it more challenging to unravel underlying biological mechanisms. Thus, it is critical to develop novel computational methods capable of dealing with such complexity and of providing some predictive deductions from the data. Following up on our developing Drosophila network inference, we are tackling larger developmental datasets. There are two steps to this. First, we are developing methods for inferring a pseudotime hand-in-hand with a dynamical system for measured gene expression values in single-cells. Second, we are developing deep learning methods to make robust low-dimensional approximations for high-dimensional sparse single-cell data that are invertible, so that dynamics inferred in low dimensionos can be projected back to the original gene expression space. The deep learning networks are trained to solve a data-driven Schwinger-Dyson equation to learn the probability distribution of the original data. By factoring the deep learning network output of the log of the probability density through a low-dimensional layer, we are attempting to obtain a geometrically accurate stochastic representation of the original high-dimensional data without making unfounded assumptions about the statistical characteristics of the measured single-cell RNA-seq data. With the help of new algorithms such as Dynamo that predict velocity in the space of RNA expression measurement, we are combining our low-dimensional deep learning representation to push this velocity forward to the low-dimensional representation. The importance of this push-forward map is that the dynamical equations of the gene expression dynamics can be determined in the low-dimensional space, and then the invertibility of the low-dimensional representation can be used to project this dynamics into the original RNA measurement space in a more accurate determination of dynamics.