Determining spaatial patterns in gene expression using in situ hybridization and RNA sequencing data

Das, Abhinav

Permalink

https://hdl.handle.net/2142/104005 Copy

Description

Title

Determining spaatial patterns in gene expression using in situ hybridization and RNA sequencing data

Author(s)

Das, Abhinav

Contributor(s)

Varshney, Lav R.

Issue Date

2019-05

Keyword(s)

• spatial patterns in gene expressions
• in situ hybridization
• RNA sequencing data
• in situ hybridization and RNA sequencing data

Abstract

Determining spatial patterns in gene expression is crucial for understanding physiological function. Image analysis and machine learning play an important role in deriving these patterns from biological data. We first focus on the analysis of single molecule fluorescence in situ hybridization (smFISH) data, obtained from the Human Cell Atlas project. Image registration is an important step in data analysis pipelines which take in image data and output spatially resolved expression of genes. We demonstrate an efficient method to register smFISH images by using a parametric representation of images based on finite rate of innovation sampling, and by optimizing empirical multivariate information measures. We then focus on the analysis of single cell RNA-seq data. When this data is collected, precise spatial information for cells is lost. We compare different approaches to reconstruct the spatial location of cells using RNA-seq data and a reference gene expression atlas. We first compare the predictions obtained by using polynomial regression and a multilayer perceptron regressor. Using polynomial regression we obtain R2 scores of over 0.99 for the prediction of x, y, and z coordinates. Using our multilayer perceptron regressor we obtain R2 scores of 0.96-0.98. We then preselect subsets of informative genes from our original dataset and test the accuracy of our multilayer perceptron regressor using these smaller sized inputs. If we select a subset of 60 genes from our original set of 84 genes, the perceptron can predict location with only a slight loss of precision.

Type of Resource

text

Language

en

Permalink

http://hdl.handle.net/2142/104005

Owning Collections