Annotation

39. Annotation#

39.1. Motivation#

Similar to scRNA-seq data, it is possible to annotate the ADT data based on surface protein markers. This can be very beneficial for the annotation of immune cells, since they are difficult to annotate in the RNA space and they are well described by their surface proteins. scRNA-seq data can suffer from dropouts, that is, although a gene is expressed in a cell population, the gene is not detected in some cells due to limitations of the sequencing procedure. Instead, ADT data does not suffer so much from dropouts due to using antibodies to quantify surface proteins. Therefore individual surface proteins show a stronger signal in the ADT data than in the RNA data.

For example, although sequenced immune cells usually include CD45 cells, the CD45 gene is not always highly expressed in the RNA data. This can be mitigated by annotating (additionally) on the ADT level.

The general annotation workflow makes use of the same functions as for RNA data and no ADT-specific functions are required.

39.2. Environment setup#

import warnings

import muon as mu
import scanpy as sc

warnings.filterwarnings("ignore")
mu.set_options(pull_on_update=False)
sc.settings.verbosity = 0
sc.set_figure_params(
    dpi=80,
    facecolor="white",
    frameon=False,
)

import lamindb as ln

ln.track()

→ loaded Transform('BnpvfJLWjHuE0005', key='annotation.ipynb'), re-started Run('QEKKxvVQBGXwIHKf') at 2026-04-10 17:37:56 UTC
→ notebook imports: lamindb-core==2.3.1 muon==0.1.7 scanpy==1.12
• recommendation: to identify the notebook across renames, pass the uid: ln.track("BnpvfJLWjHuE")

39.3. Loading the data#

We load the MuData object we saved at the end of the previous chapter Batch correction:

af = ln.Artifact.connect("theislab/sc-best-practices").get(
    key="surface-protein/cite_batch_correction.h5mu", is_latest=True
)
mdata = af.load()
mdata

MuData object with n_obs × n_vars = 117951 × 36737
  var:	'gene_ids', 'feature_types'
  2 modalities
    rna:	117951 x 36601
      obs:	'donor', 'batch'
      var:	'gene_ids', 'feature_types'
    prot:	117951 x 136
      obs:	'donor', 'batch', 'n_genes_by_counts', 'log1p_n_genes_by_counts', 'total_counts', 'log1p_total_counts', 'n_counts', 'outliers', 'doublets_markers'
      var:	'gene_ids', 'feature_types', 'n_cells_by_counts', 'mean_counts', 'log1p_mean_counts', 'pct_dropout_by_counts', 'total_counts', 'log1p_total_counts'
      uns:	'batch_colors', 'donor_colors', 'doublets_markers_colors', 'neighbors', 'pca', 'umap'
      obsm:	'X_pca', 'X_pca_harmony', 'X_umap'
      varm:	'PCs'
      obsp:	'connectivities', 'distances'

39.4. Manual annotation#

First, we check the expression of CD45. CD45 is one of the most abundant proteins in the T-cell plasma membrane and required for TCR signaling. It activates Lck, which in turn is required to phosphorylate the TCR complex [Courtney et al., 2019]. Therefore, CD45 should be broadly expressed in our dataset, and even more highly expressed in T cells.

sc.pl.umap(mdata["prot"], frameon=False, color="CD45")

../_images/4f7000e2cd1b5b04cae78d0768c78eda2f08d8ef0cf4c4cb5055d33f711657e6.png

The measured ADTs use a slightly different nomenclature due to name clashes with RNA genes. The var_names_make_unique function was used to separate gene names from protein names and the proteins might have -1 suffixes. We look up an example gene name (CD38) to exemplarily find the exact nomenclature in our variable names:

mdata["prot"].var[mdata["prot"].var.gene_ids.str.contains("CD38")]

	gene_ids	feature_types	n_cells_by_counts	mean_counts	log1p_mean_counts	pct_dropout_by_counts	total_counts	log1p_total_counts
CD38-1	CD38	Antibody Capture	116434	52.892694	3.986995	4.57481	6453755.0	15.680173

We cluster the cells with a relatively low resolution. Similarly to scRNA-seq data annotation, it is possible to increase the resolution for more fine-grained annotations.

sc.tl.leiden(
    mdata["prot"],
    resolution=0.1,
    flavor="igraph",
    n_iterations=2,
    directed=False,
    random_state=0,
)

To check which surface markers are present in which cell type, we use the scanpy rank_genes_groups() function and then we create a dotplot. The dotplot will indicate the 3 most differentially expressed genes in each cluster.

sc.tl.rank_genes_groups(mdata["prot"], groupby="leiden")
sc.tl.dendrogram(mdata["prot"], groupby="leiden")
sc.pl.rank_genes_groups_dotplot(
    mdata["prot"], n_genes=3, values_to_plot="logfoldchanges"
)

../_images/6c843e093ed726898596af661a05b520b03876a6d077138f279fffbd5c26e5d5.png

We can already identify clusters 0 and 2 as T cell populations by CD3 expression, and cluster 7 as B cells by CD19 expression. We next plot the UMAP that we calculated in our previous chapters and color it by cluster.

sc.pl.umap(mdata["prot"], color="leiden")

../_images/bfc3a60fcf9a3025a68047f82d5396c8a25f2d298e2f92a992848dfbbec804a0.png

We’ll check a few known markers of major immune cell types in order to identify which cell type is each cluster.

# B cells
sc.pl.umap(mdata["prot"], frameon=False, color=["CD19-1"])

../_images/8cbd5e96e01762c3365ca6d6e6cae11fea5197573439b2db60c627f35f755e8b.png

As could be seen in the dotplot, cluster 7 expresses CD19 which is a B cell marker.

Let’s look into the T cells in more detail and separate them into CD4 and CD8 cells.

# T cells
sc.pl.umap(mdata["prot"], color=["CD3", "CD4-1", "CD8"])

../_images/b90710ae256e8056d74c950e1d6ace8925c2cff1183a2c1356d34c2e9a0861b5.png

In the following few plots, we continue looking for known markers of other cell types: NK cells, monocytes and dendritic cells.

# NKT cells are CD3+ and CD56+
# NK cells are CD3- and CD56+
sc.pl.umap(mdata["prot"], color=["CD56"], frameon=False)

../_images/5ba2df062c9783df805c5547261f531158094d6f5e874f37ad57f92075d83b13.png

# Monocytes
sc.pl.umap(mdata["prot"], color=["CD11b", "CD14-1"], frameon=False)

../_images/cbc3e41c7bb837814f8dfa2c98b90f49e18ea041c56aac21d392c51f183d9a6d.png

# Dendritic
sc.pl.umap(mdata["prot"], color=["CD123", "CD11c", "CD303"], frameon=False)

../_images/cc99922e51ab8a2836abf1c808f263cfcef134f64f72aad4af6b2368e89548db.png

# CD16 is expressed in NK cells and in CD16 monocytes, which are CD14-, CD16+ and CD11c+
sc.pl.umap(mdata["prot"], color="CD16", frameon=False)

../_images/71c4b3f5b00bceb8d04a425b20c6be372a73dbb11c5b9ff10788309d95615dcc.png

Now that we know what cell type each cluster is, let’s replace the 0-8 cluster numbers with the actual cell type names:

mdata["prot"].obs["celltype"] = mdata["prot"].obs.leiden.copy()
mdata["prot"].obs.celltype.replace(
    {
        "0": "CD4 T",
        "1": "Cytotoxic T",
        "2": "CD8 T",
        "3": "CD14 Mono",
        "4": "Proliferating cells",
        "5": "DC",
        "6": "CD16 Mono",
        "7": "B",
        "8": "NK",
    },
    inplace=True,
)

sc.pl.umap(
    mdata["prot"],
    color="celltype",
    legend_loc="on data",
    legend_fontsize=11,
    legend_fontoutline=2,
)

../_images/4e17cdab8bdcdb7e7d8115d78c3eb5ead88008d8bbb748f4ca0f5b283e97dfe5.png

We have uncovered and annotated the main cell types in the data. Now we could perform a more fine-grained annotation by increasing the resolution of clustering and annotating the resulting fine-grained clusters.

In this chapter we describe how to annotate cell types based on the ADT data of CITE-seq. Another interesting way forward is to annotate cell types based on combined information from ADT and from RNA data. We refer to the Paired integration chapter for that.

af_annotation = ln.Artifact.from_mudata(
    mdata,
    key="surface-protein/cite_annotation.h5mu",
    description="CITE-seq data after annotation",
)
af_annotation.save()

ln.finish()

39.5. Automated annotation#

It is technically possible to use cell type classifiers trained on ADT data and to map against ADT reference datasets. However, ADT-specific methods are sparse if not non-existent, and we refer to the RNA annotation chapter for methodological details Annotation.

39.6. References#

[CSL+19]

Adam H Courtney, Alexey A Shvets, Wen Lu, Gloria Griffante, Marianne Mollenauer, Veronika Horkova, Wan-Lin Lo, Steven Yu, Ondrej Stepanek, Arup K Chakraborty, and Arthur Weiss. Cd45 functions as a signaling gatekeeper in t cells. Sci. Signal., 12(604):eaaw8151, October 2019.

39.7. Contributors#

We gratefully acknowledge the contributions of:

39.7.1. Authors#

Javier Marchena-Hurtado
Daniel Strobl
Ciro Ramírez-Suástegui

39.7.2. Reviewers#

Lukas Heumos
Anna Schaar