Two popular kits for reading single cells count different things

In Short. A new analysis finds that kits from 10x Genomics and Parse Biosciences give different gene counts from the same cells. Researchers who combine data made with the two kits will need to correct for the kit first.

EVERY CELL in a blood sample carries the same set of genes, but each one makes RNA copies of only some of them, and which genes it copies tells a scientist what kind of cell it is. A method called single-cell RNA sequencing counts those copies one cell at a time. Each cell’s RNA is tagged with a genetic barcode, so every copy can be traced to its cell. Most such work uses kits from two firms.

The two kits work in different ways. The kit from 10x Genomics seals each cell in a tiny oily droplet together with a bead that carries the barcode. The kit from Parse Biosciences first fixes whole cells, soaking them in chemicals that preserve them and open holes in their membranes. It then tags the RNA by pooling the cells and splitting them across the wells of a plate in several rounds, adding part of a barcode each time. A study published in 2025 by researchers at Genentech, a drugmaker that makes neither kit, compared nine commercial kits and found that the ones that fix cells, as a side effect, recorded proportionally more RNA from the nucleus. That means that a low count for a gene that is copied mainly outside the nucleus may say more about what a kit misses than about the cell.

On October 11th Maya Caskey and Ingileif Hallgrimsdottir of the California Institute of Technology posted a comparison of the two kits. The paper is a preprint and has not yet been reviewed by other scientists. It also reanalyzes experiments that other groups had already run, seven in all from five published studies, in which the same cells were put through both kits, so its independence lies in the analysis rather than in the cells.

In Parse data, RNA that had not yet been fully processed, which is found mostly in the nucleus, made up between 35.9% and 59.9% of the counts in each cell, against between 13.8% and 40.7% in 10x data. Genes that make the proteins of ribosomes, the structures that build a cell’s proteins, accounted for between 0.44% and 2.15% of counts in Parse data and between 9.00% and 25.96% in 10x data. Standard markers that researchers use to recognize T cells and natural killer cells, two kinds of immune cell, were recorded at roughly 10 to 50 times lower levels in Parse data. One of them, NKG7, was about 49 times lower.

The paper also tested ambient RNA, stray RNA from outside a cell that lands on that cell’s barcode and makes it look as if the cell had copied genes it did not. The cleanest test mixes human and mouse cells, because any mouse RNA on a human cell’s barcode must be stray. It misses stray RNA from the same species, so its figures are a lower bound. In two Parse datasets from such mixtures, a median of 1.2% of the counts on human barcodes came from mouse, and 2.1% of the counts on mouse barcodes came from human. In a 10x dataset the comparable figures for the middle half of barcodes ranged from 0.5% to 0.8% and from 1% to 2%. The authors call the two kits broadly similar on this measure.

That finding matters for a claim Parse makes about its own product. A comparison on the firm’s website reports reduced ambient RNA and fewer ribosomal reads in its kit. But that claim concerns its newest kit, version 4, while the preprint used older versions of both products. Its older-kit data do not match the claim, but they do not show that the claim is wrong for the newer kit.

In matched blood samples, however, both kits sorted cells into the same types, and the share of cells of each type differed by no more than 3 percentage points. That is a weaker test than agreeing on counts, because a cell type can be recognized from many genes even when a few are poorly measured. An earlier peer-reviewed comparison points the same way. A team at the University of Tartu in Estonia found that ribosomal genes made up 12.5% of counts in 10x data from mouse thymus cells against 0.6% in Parse data, and that only 364 of the 1,000 most-counted genes in each dataset were shared.

Labs that pool public data made with the two kits will need to correct for the kit before comparing cells.