In SydneyBioX/scMerge: scMerge: Merging multiple batches of scRNA-seq data

Introduction

This is a human pancreas single-cell data comprising of 4 experiments.

Integration challenge

Prior to integration, there is a strong separation effect by experiments.
In addition, there are 3 different protocols, the sequencing depths poses an extra challenge.
With 4,566 cells, this data poses a moderate dimensionality challenge to data integration.

Data description

Data source:

| Type of merge | | Name | ID | Author | DOI or URL | Protocol | Organism | Tissue | # of cell types | # of cells | # of batches | |----------------------------------------------------|---|-----------|-------------|-------------|----------------------------|------------|----------|----------|-----------------|------------|--------------| | Across platforms with significant depth difference | | Pancreas4 | GSE81608 | Xin | 10.1016/j.cmet.2016.08.018 | SMARTer/C1 | Human | Pancreas | 6 | 4566 | NA | | | | | GSE83139 | Wang | 10.2337/db16-0405 | Smart-Seq | | | | | | | | | | GSE86469 | Lawlor | 10.1101/gr.212720.116 | SMARTer/C1 | | | | | | | | | | E-MTAB-5061 | Segerstolpe | 10.1016/j.cmet.2016.08.020 | Smart-Seq2 | | | | | |

Relation to the scMerge article: Supplementary Figure 4 and 5.

Data visualisation

PCA plots by cell types and batch

{width=100%}

RLE plots and percentage of explained variance plots

{width=100%}

Integrated `scMerge` data

Data availability: Pancreas4 (in RData format)
scMerge parameters for integration:
- Unsupervised scMerge
- kmeans K = (4,6,6,6)
- Negative controls are human scSEG