Question: How does high background from unbound CMOs affect Cell Multiplexing metrics in 3’ CellPlex experiments?
Answer: If background noise is too high in a CellPlex dataset, Cell Ranger tag assignment algorithms will be unable to distinguish true signal from noise. If Cell Ranger cannot assign one or more CMO tags to a cell with at least 90% confidence, the cell will remain “unassigned.” Therefore, high background noise can lead to a low percentage of cells assigned to a sample.
Below, we highlight several key metrics and plots from CellPlex Web Summary files and provide examples of how these metrics & plots are affected by high background noise.
High background noise has several possible causes, including poor sample quality, high debris, insufficient washing post CMO labeling, and letting cells sit at room temperature after CMO labeling. For further recommendations on reducing background noise, see this article: How can I reduce background from unbound CMOs using the 3’ CellPlex kit for Cell Multiplexing? Also refer to our Technical Note on Cell Multiplexing.
1) Cells assigned to a sample
We expect to see a high fraction of cells assigned to a sample and a high singlet capture ratio. The exact percentage of cells assigned to a sample will vary, given that the expected cell multiplet rate varies with cell load.
Example - adequate washing & high-quality sample
Insufficient washing or poor sample quality can lead to a low fraction of cells assigned to a sample and a low singlet capture ratio.
Example - insufficient washing and/or poor-quality sample:
2) Histogram of CMO Count
We expect to see a clear separation between the background peak (noise) and the foreground peak (signal). The foreground peak represents cells that are positive for a given CMO.
Example - adequate washing & high-quality sample:
In some heterogeneous samples, such as PBMCs, the foreground peak may be multimodal. However, we still expect to observe a clear separation between the background peak and the foreground peaks.
Example - adequate washing (heterogeneous sample):
Insufficient washing or poor sample quality can lead to a lack of clear separation between the foreground and background peaks, as shown in the examples below. This makes it challenging to distinguish a true signal from background.
Examples - insufficient washing and/or poor-quality sample:
3) Biplots of CMO Count
In a CellPlex experiment performed with two CMO labeled samples, we expect to see five populations in the Biplot:
- Dark gray = cells with high UMI counts for CMO301, and low UMI counts for CMO302. These cells are assigned to sample CMO301.
- Light gray = cells with high UMI counts for CMO302, and low UMI counts for CMO301. These cells are assigned to sample CMO302.
- Red = Cells with high UMI counts for both CMO301 and CMO302. These cells are assigned as cell multiplets.
- Green-gray: Cells with low UMI counts for both CMO301 and CMO302. These cells are assigned as blanks.
- Yellow: Cells with intermediate UMI counts. Cell Ranger cannot confidently assign these cells to one of the above categories and leaves these cells as “unassigned.”
We expect to see well-separated clusters corresponding to categories 1, 2, and 3. We expect to see very few cells in categories 4 and 5.
Example - adequate washing & high-quality sample:
Insufficient washing or poor sample quality can lead to a lack of clear separation between populations, as shown in the example below. This makes it challenging to confidently assign cells to a CMO, leading to many cells in the “unassigned” category.
Example - insufficient washing and/or poor-quality sample:
4) t-SNE Projection of Cells by CMO
We expect each CMO labeled sample to form a distinct cluster in the t-SNE projection. We also expect to see smaller clusters that lie in between the larger clusters. These smaller clusters represent cell multiplets.
Example - adequate washing & high-quality sample:
Insufficient washing or poor sample quality can lead to a lack of clear separation between clusters, as shown in the example below. This makes it challenging to make accurate CMO tag assignments.
Example - insufficient washing and/or poor-quality sample:
5) Fraction CMO reads in cell-associated barcodes
In the CellPlex library, we expect to see a high fraction of reads in cell-associated barcodes for each CMO.
Example - adequate washing & high-quality sample:
Insufficient washing or poor sample quality can lead to a low fraction of CMO reads in cell-associated barcodes, due to the presence of unbound CMOs that get partitioned into GEMs that do not contain any cells. This is a sign of high background in the data, which will make it difficult to distinguish between true signal and noise.
Example - insufficient washing and/or poor-quality sample:
6) CMO Barcode Rank Plot
We expect to see a steep cliff that separates GEMs containing CMO labeled cells from empty GEMs that do not contain any cells.
Example - adequate washing & high-quality sample:
Insufficient washing or poor sample quality can lead to a lack of clear separation between cell-containing GEMs and empty GEMs.
Example - insufficient washing and/or poor-quality sample:
Products: Single Cell Gene Expression, CellPlex