Introduction
Spatially resolved transcriptomics (SRT) is a novel, disruptive technology set to push the boundaries of exploring gene regulatory networks while providing both spatial and temporal resolution. It is expected to be exponentially adopted by the transcriptomics community after being named Nature’s Method of the Year in 2020. Despite this, analysis packages for SRT datasets are still in their infancy, with a clear forerunner yet to emerge. A comprehensive review of the performance of commonly used packages on the same datasets is lacking and there is a need to determine their performance in correctly labelling spatially variable genes (SVGs) in a tissue section.
Methods
To establish which of the current packages is most effective in identifying SVGs within data generated using the same technology, I am creating a benchmarking process by testing a combination of publicly available, and simulated 10X Genomics’ Visium datasets generated from cardiac tissue. I will be assessing the performance of SpatialDE, SPARK, BOOST-GP, scGCO and Seurat. The simulated data will be modelled after the distributions of gene expression most common in various Visium data sets.
Results
Each package uses a different mathematical model to label SVGs and preliminary results from publicly available data demonstrate that each package identifies SVGs independent of other results from other packages. Of the 4 packages tested on publicly available data, scGCO is the most conservative while SPARK finds the most SVGs and only 52 SVGs are identified by all packages.
Conclusion
Various studies have established the ability of SRT to uncover novel genetic determinants in different biological tissues and conditions; therefore, ensuring the most accurate analysis is critical. Once the benchmarking is completed, it will offer a clear workflow to be implemented in future SRT experiments and a better understanding of mechanisms underlying disease and tissue development.