In the page Search, there are 4 sections for users to search scRNA-seq dataset, context-specific reference, marker gene and immune cell type by entering letters within cancer names and obtaining a list of possible names or choose data directly.

In the page Tool, users can submit their single-cell RNA-seq data to explore the immune cell types by selecting some interested prediction methods. Through the integrated project, we give all the results belonging to every method and the optimal annotations. Following pages would be loaded in order.

- Data upload. Users should submit a single-cell expression profile(.txt or .csv) and provide E-mail address to create task ID. As some integrated method would take a long time, we will send a message when these computations are done. To get an integrated result, users should choose some interested methods, with specifying cancer types or normal tissue and specifying whether to predict unknown, and submit.

- Quality control. After data uploading and method selection, users can define quality control conditions according to data distribution by adjust the right sliders and GO.

- Analysis results. Results of each method would be displayed in order and users can check them by clicking the button above. Only selected methods and the integrated results are available. Note that if different contexts are chosen, there would be two ensemble results.

Results of Annotation

ImmCluster provided four major analytic modules that allow users to interactively explore the annotations of immune cells.

I) Clustering and annotating the immune cell clusters

Firstly, two immune cell clustering plot, which represent clusters and annotated immune cell types of a certain method separately. Then, the annotating results for each cell can be check and downloaded in the following table and the barplot below gives the number of cells belonging to different cell types.

II) Gene expression of markers across cell clusters

Using COSGR, a method to rapidly identify highly expressed genes in single cell data, we provide a list of top 15 genes within each immune cell type ranked by the COSGR scores, which could also be downloaded. And the expression level of each gene can be selected to be shown in a violinplot. All top 15 genes’ expression were shown in a whole heatmap.

III) Functional assignment of the cell clusters

We calculate function scores of each immune clusters for cancer hallmarks, cell states and immune pathways through ssGSEA as shown in the heatmap plot. Asterisk indicates the significance of overlaps between the top 100 marker genes and the signatures of certain function, which would shown in the table.

IV) Cell-cell communications

We infer cell-cell communication based on the expression of ligands and their coupled receptors in scRNA-seq datasets by using three methods including iTALK, celltalker and ICELLNET. A table shows the communication score of each pair of ligand-receptor interactions occurred in two cell types. An overview of cell-cell communication and strong interactions ranked by the score will be displayed in a circle plot and a river plot, respectively.

Annotations of immune clusters from the page Browse

scRNA-seq data collection

We used a list of keywords ((scRNA seq OR single-cell RNA-seq) AND (cancer OR carcinoma OR adenocarcinoma)) to search datasets from GEO, 10X Genomics, and NGDC platform, and obtained 60 datasets of human single cell RNA-seq transcriptomes. In addition, for the datasets which were already annotated in the original publications, only cells identified as immune cell types were included. If the original publications not described, we only retain cells expressing PTPRC, a general marker of immune cells. Through literature review, the immune cell types in brain cancer are different from those of other solid tumors. After removing the data of glioma and following strict quality control, a total of 986,589 immune cells were remained, including 25 immune cell types from 17 cancer types and 49 immune cell types from 9 normal tissues.

scRNA-seq data filtering

After quality control of each scRNA-seq data, samples with the number of cells that were less than 50 or datasets with the number of cells that were less than 1000 were deleted. Besides, we filtered datasets or samples that seriously affect the effectiveness during data integration and batch effect correction.

Marker atlas developing *

1.Gene is detected in at least 3 cells at last 3 counts across all cells.

2.Gene in statistically significantly higher in expression in this cluster compared to the complement set. To establish significance, we used a two-tailed Mann-Whitney U test with multiple hypothesis correction,FDR < 5%.

3.Gene has maximal average expression in this cluster.

4.Gene satisfies the max-to-second-max ratio is at least 1.1x.

ensemble integration

We develop an integrating project to annotate immune cell types with a label matrix based on voting mechanism.

For each cell, the selected methods would give annotation results which are gathered and labeled. According to the frequency of immune cell types, we record the one which has the highest credibility with a 'real' label. When more than two label a cell hold, we define it as 'unknown'. Similarly, the cells with different immune lineages are also labeled as 'unknown'.

* : Zilionis R, Engblom C, Pfirschke C, Savova V, Zemmour D, Saatcioglu HD,et al. Single-Cell Transcriptomics of Human and Mouse Lung Cancers Reveals Conserved Myeloid Populations Across Individuals and Species. Immunity (2019) 50(5):1317–1334.e10. doi: 10.1016/j.immuni.2019.03.009

1.Normal tissues from healthy donor

1.1 No. of immune cells

1.2 No. of samples

1.3 No. of marker genes

2.Cancer types

2.1 No. of immune cells

2.2 No. of samples

3.3 No. of marker genes