Comparative performance of the BGI and Illumina sequencing technology for single-cell RNA-sequencing

Abstract

The libraries generated by high-throughput single cell RNA-sequencing (scRNA-seq) platforms such as the Chromium from 10× Genomics require considerable amounts of sequencing, typically due to the large number of cells. The ability to use these data to address biological questions is directly impacted by the quality of the sequence data. Here we have compared the performance of the Illumina NextSeq 500 and NovaSeq 6000 against the BGI MGISEQ-2000 platform using identical Single Cell 3' libraries consisting of over 70 000 cells generated on the 10× Genomics Chromium platform. Our results demonstrate a highly comparable performance between the NovaSeq 6000 and MGISEQ-2000 in sequencing quality, and the detection of genes, cell barcodes, Unique Molecular Identifiers. The performance of the NextSeq 500 was also similarly comparable to the MGISEQ-2000 based on the same metrics. Data generated by both sequencing platforms yielded similar analytical outcomes for general single-cell analysis. The performance of the NextSeq 500 and MGISEQ-2000 were also comparable for the deconvolution of multiplexed cell pools via variant calling, and detection of guide RNA (gRNA) from a pooled CRISPR single-cell screen. Our study provides a benchmark for high-capacity sequencing platforms applied to high-throughput scRNA-seq libraries.

Figure 1.

Experimental design. Preparation of single cell libraries and sequencing using Illumina and BGI platforms and subsequent analysis. (A) Human iPSC were generated generated from a human donor and underwent SNP genotyping in addition to scRNA-seq. (B) Primary TMWC were screened with a CRISPR-based molecular screen (CROP-seq). (C) PBMC. Single-cell libraries were prepared from two individual pools of PBMCs.

Figure 2.

Cells and genes detected by platforms. Both technologies demonstrated similar sensitivity in the detection of cells and genes. (A) Capture efficiency of each platform. Efficiency is evaluated based on the number of genes and molecules detected in a cell. (B) Total number of molecules detected in a cell. Histograms on each axis represent the distribution of total UMIs in a cell, while the scatter plot represents the correlation of UMI detection for a cell, between the two platforms. (C) Dropout rate across genes detected by platform. Dropout rates for each gene, per platform were calculated using NBDrop from the M3Drop package.

Figure 3.

Concordance of datasets sequenced by different platforms. (A) Pearson correlation of gene expression between cells identified by both sequencing platforms. (B) PCA representation of each sequencing platform per dataset. (C) Cluster predictions projected on to UMAP plots separated by sequencing platform. (D) AUROC scores measuring the similarity of corresponding clusters across platforms, for each dataset as calculated by MetaNeighbor.

Figure 4.

Experiment-specific metrics. (A) Metrics related to guide RNA assignment in TMWC. This excludes cells that were not affiliated with a guide RNA and cells that with ambiguous assignments. (B) Number of SNPs called per cell in iPSCs. SNPs were called from alignments of cells found in NextSeq 500 and MGISEQ-2000 datasets.