Electronic Thesis and Dissertation Repository

Thesis Format

Monograph

Degree

Master of Science

Program

Biochemistry

Supervisor

Andrews, Tallulah

Abstract

Spatial transcriptomics is a new technology that enables measurement of the whole transcriptome of a tissue slide at high resolution. One challenge with this technology is that tissue slices from different biological conditions rarely align perfectly, thus it is difficult to identify when a gene or cell-type changes its spatial distribution between different samples. To address this challenge, we have adapted methods designed to remove batch effects from single-cell RNA-seq data to identify conserved spatial regions which can be used to measure relative locations and positional changes between different samples of spatial transcriptomics. This was accomplished by developing our Graph-based Integration and Analysis of Spatial Transcriptomics (GIAST) tool. We used synthetic data with a known ground truth and published datasets with sequential slices of spatial transcriptomics and manually annotated conserved structures to compare our tools performance to existing single-cell RNA-seq integration tools. We demonstrated our tools ability to consistently perform well compared to other tools on these datasets while at the fraction of the run time as well as determining change in location.

Summary for Lay Audience

New technologies allow researchers figure out which genes are active in different areas of tissue slide samples, while also keeping track of where those areas are. This way, researchers can get a better understanding of how cells work together and can even compare healthy samples with diseased samples to get a better understanding of how they develop. Specifically, using this technology we can even potentially determine any differences in the location of these cells relative to the overall sample. However, this can be very challenging as a result of samples being processed at different laboratories since this can lead to small differences that are not related to what is actually being studied. We created a tool to address this problem with this new technology by altering approaches that exist to address this issue in similar technologies and improve it to work efficiently and effectively. To improve our tool we created synthetic samples to determine the best approach and applied it on real samples. Comparing our tool to existing tools we demonstrated its ability to consistently perform well and perform the fastest on both synthetic and real data. We also demonstrated our tools ability to determine changes in location between groups of similar cells in different samples.

Download

Share

COinS