Comparing Slide-seq and Slide-seqV2 counts

Slide-seq is a promising technology for the study of cellular tissue anatomy.It uses a clever strategy to randomly deposit barcoded RNA-capture beads on slides. To record the location of the beads the barcode is read off using in situ sequencing. Following this, a slice of tissue can be deposited on the slide, where the beads can capture RNA from the cells of the tissue. After this step the beads can be treated as from a Drop-seq experiment, and with high throughput sequencing one can arrive at a count matrix of beads x genes, with the paired information of spatial coordinates of the beads.

Slide-seq was published a year ago, and the dataset is still the largest published in terms of number of observations with 2,522,640 beads. The data ended up being difficult to analyze however, due to extremely low counts per bead. Even with very large numbers of observations it is difficult to learn structure in count data of counts are small.

Recently, an update called ‘Slide-seqV2’ was published. In the paper they describe testing Slide-seq and Slide-seq2 on serial sections of brain tissue and find a ten-fold improvement in average counts.

This is on newly generated data, and I was curious how the previous data compares. Relative improvement within an experiment tells you about how much better the protocol is, but what I wanted to know was how much easier this experiment would be to analyze. For this sequencing depth needs to be considered, which may not necessarily be matched between the old and the new data.

After downloading the data made public by the authors, I converted it, combined it with the old Slide-seq data, and made a plot comparing the number of counts per bead between the protocols. (I also live streamed this process, available here)

A simple analysis shows that for the Slide-seq data there are on average 40 UMI’s per bead, while the Slide-seqV2 data has 200 UMI’s per bead: a 5-fold increase. The Slide-seq2 data has a very wide spread of counts per bead.

Preferably you would have at least a few thousand UMI’s per observation when doing analysis. This is a step in the right direction, but it still might require the supervised style analysis presented in both the papers, which make use of additional scRNA-seq data as a guide.

The analysis is available on Github, and I have made available H5AD files containing the Slide-seq here and Slide-seq2 data here.