henhaobio
Sign in

bioRxivImmunologyPreprint

Systematic Comparison of 10X Genomics and Parse Single Cell RNA Technologies across PBMC and CD8+ TEMRA Cells

Alejandro EspinozaiD, Bruno Jorge de Andrade SilvaiD, Matteo A. PellegriniiD

PDFDOI
In-depth full-text digest available · Jump to digest ↓

Abstract

Single-cell RNA sequencing technologies provide insights into gene expression at the cellular level, enabling detailed analysis of cellular heterogeneity. In this study, we systematically compared two scRNA-seq platforms,10X Genomics and Parse Biosciences, using human peripheral blood mononuclear cells (PBMCs) and terminally differentiated effector memory CD8+ T cells (TEMRAs). We identified significant differences in gene expression variability and platform-specific biases, such as ribosomal and mitochondrial gene capture. 10X has a bias for shorter genes, while Parse exhibited enhanced detection of longer transcripts. In CD8+ TEMRAs, the expression of key genes related to antimicrobial immune responses were underrepresented in Parse cells compared to 10X cells (e.g. GNLY, PRF1 and GZMB). These findings underscore the need for careful selection of scRNA-seq platforms based on specific research objectives, as platform-specific biases can influence cell type identification and as well as mechanistic insights derived from gene expression data. Our results provide insights for selection of scRNA-seq experimental platforms in immunological studies.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (6 Sept 2026) · CC BY 4.0

Side-by-side PBMC and CD8+ TEMRA profiling shows Parse favors longer transcripts but underdetects cytotoxic genes (GNLY, PRF1, GZMB), so 10X Chromium better resolves TEMRA subsets.

Split-pool ligation scRNA-seq (Parse) is a newer, potentially cheaper alternative to droplet-based 10X, and an earlier PBMC comparison saw lower T cell gene expression counts in Parse but left T cell subpopulations unexplored. This work tests both platforms on heterogeneous PBMCs and on homogeneous cytotoxic CD8+ TEMRA cells, whose co-expression of GZMB, GNLY and PRF1 defines their functional states.

Key results

  1. ShownIn matched PBMCs, average protein-coding gene expression agreed closely between platforms. However, 10X captured far more ribosomal transcripts per cell than Parse, and more mitochondrial transcripts too.

    PBMCs (three donors; 37,809 10X and 37,529 Parse cells after QC): Pearson r = 0.904 for averaged protein-coding UMI counts; ribosomal transcripts averaged 5.58% per cell in 10X vs 0.38% in Parse (p < 0.05). · Quality control analysis of PBMCs; Transcriptomic differences in gene abundance…

  2. ShownTranscript length is associated with a consistent detection bias: Parse preferentially detects longer genes. Genes enriched on each platform map to different functions: translation and mitochondrial processes for 10X; adhesion, epidermal and extracellular matrix terms for Parse.

    In PBMCs, Parse-favored genes were longer than 10X-favored genes by >1000bp on average. In TEMRA cells, Parse-favored genes had longer total exon length (p < 2.22e-16) and 10X-favored genes a slight GC bias (p < 7.37e-9). Mean protein-coding gene size: 4686.52 (bulk) vs 4674.52 (10X) and 4769.76 (Parse). · Transcriptomic differences in gene abundance…; CD8+ TEMRA cells’ immune gene expression differs…

  3. ShownBoth platforms recovered the major PBMC lineages, but T cell and dendritic cell (DC) marker sets agreed poorly. Parse showed weaker CD3 gene expression, yet it resolved basophils, DCs and CD34+ progenitors that were absent from 10X clusters.

    15 (10X) vs 16 (Parse) clusters, annotated as 11 vs 13 cell types. Jaccard overlap of top markers was notably lower for T cells and DCs. Azimuth L1 gave more T cells in 10X and more monocytes in Parse; L2 showed more CD4+ TEM, CD8+ TCM and Tregs in 10X and an enrichment of double-negative T (dnT) cells in Parse. · Unsupervised clustering identifies differences…

  4. ShownIn sorted CD8+ TEMRA cells, Parse showed reduced CD3, CD8 and cytotoxic transcripts. Azimuth labeled most Parse cells as NK cells but most 10X cells as CD8 TEM, a shift the authors attribute, as likely, to CD3 and CD8 dropout in Parse.

    Donor 302, 2800 10X vs 2915 Parse cells; mean log-normalized expression R² = 0.804; Jaccard overlap of the top 100 DEGs per matched cluster was below 50% in all cases. Parse was lower for GNLY, GZMB, PRF1, NKG7, CD3D, CD3E, CD8A and IL7R; the Discussion cites an average log2 fold change > 2 for GNLY, PRF1 and GZMB in 10X. · CD8+ TEMRA cells’ immune gene expression differs…; DISCUSSION

  5. ShownCytotoxic module scores built from bulk RNA-seq gene sets were higher in 10X TEMRA cells, and far more 10X cells were classified as tri-cytotoxic T lymphocytes (TCTL). Gene Ontology terms for T cell activation, differentiation and T cell receptor signaling were enriched only among 10X-favored genes.

    Mean module scores in 10X: MCTL1 0.022, TCTL1 0.0437, MAIT1 0.00007. Parse was negative for all subtypes (MCTL1 −0.0581, TCTL1 −0.0340). TCTL calls: 53% of 10X cells vs 19% of Parse cells. · CD8+ TEMRA cells’ immune gene expression differs…; DISCUSSION

Methods, data and reproducibility

Matched PBMCs from three donors and FACS-sorted CD8+ TEMRA cells from one donor were profiled with 10X Chromium 3' v3.1 and Parse Evercode WT v2, then downsampled to equal depth. The platforms were compared on gene-level correlation, gene length and GC content, clustering, Jaccard marker overlap, Azimuth and SingleR annotation, and module scores built from bulk RNA-seq gene sets.

  • Matched PBMC scRNA-seq libraries (10X 3' v3.1 and Parse Evercode WT v2) from healthy donors D000, D252 and D394 · human · n = three donors; 37,809 10X and 37,529 Parse cells after QC · GSE285843
  • FACS-sorted CD8+ CD45RA+ CCR7- TEMRA cells (≥99% purity) profiled on 10X and on Parse Evercode WT Mini v2 · human · n = donor 302; 2800 10X and 2915 Parse cells · GSE285843
  • Bulk RNA-seq of healthy-donor PBMCs, used to compare gene length and expression profiles with both platforms · human · n = seven healthy individuals · GSE159337
  • Azimuth human PBMC reference ('pbmcref'), used for automated hierarchical cell type annotation · human
Methods & tools
10x Genomics Chromium Next GEM Single Cell 3' v3.1, Parse Biosciences Evercode WT v2 and WT Mini v2, Cell Ranger v7, Parse Biosciences analysis pipeline v1.1.2, Seurat v5 (Louvain clustering, ModuleScore), scRecover (read downsampling), DoubletFinder, Harmony, Azimuth (pbmcref), SingleR (MonacoImmuneData), clusterProfiler, FACS (BD FACSAria II)
Reagents & environment
R v4 with Seurat v5; Cell Ranger v7; Parse Biosciences analysis pipeline v1.1.2; reads aligned to GRCh38; gene lengths taken from the gencode.v32 primary assembly annotation.
Reproducibility gaps
No analysis code is released. The single-cell data are given only as GEO accession GSE285843, with no URL. The curated marker lists and the module gene sets built from bulk RNA-seq are not provided in the text.
Limitations (5)
  • The Azimuth PBMC references were built from 10X-derived cells, which may bias reference-based annotation of Parse data. The authors call for curated references that suit newer transcriptome methods. · stated by the authors
  • Only 10X 3' v3.1 and Parse Evercode WT v2 chemistries were tested. The authors note that the updated kits (Parse Evercode WT v3 and 10X GEM-X v4) could lessen the biases seen here, and that future benchmarks should cover them. · stated by the authors
  • The analysis focused on T cells. Discrepancies in other lineages, such as monocytes, were noted but not investigated further. · stated by the authors
  • The TEMRA comparison rests on cells from a single sorted donor (donor 302). Subset calls were not checked against protein-level granzyme B, perforin or granulysin staining, so the claim that 10X better captures TEMRA diversity relies on bulk-derived gene signatures. · noted in this summary
  • Platform differences are confounded with fixation and -80°C storage (Parse only), cell input and doublet-rate settings (9.5% vs 3%). Reporting also conflicts: the mitochondrial cutoff is 20% in Results vs 15% in Methods, and Pearson r = 0.904 is later restated as R² > 0.9. · noted in this summary

Read this if you are choosing between droplet-based (10X) and split-pool (Parse) scRNA-seq for immune profiling, especially of cytotoxic CD8+ T cells, or if you annotate Parse data with references built from 10X data.

Citations
1

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISCaskey '26Xie '24Simone '24Xiong '26Colino‐Sang '24King '26
70% semantic matchGenomics
Maya Caskey, Ingileif B. Hallgrímsdóttir·bioRxiv·Oct 11, 2026