There are different methods to automatically classify cell type based on biomarkers. There are two methods available in Connected Multiomics

• ScType

• Garnett

Garnett[1] automated cell type classification method should be invoked from a non-normalized raw count matrix data node. It use the specified classifier to generate cell type annotation on the input data

Select a raw count data node, choose Classification > Classify cell type task on the pop-up menu

If using the Managed classifiers tool for the first time, you will be asked to create a new classifier file. Users may select either the Download Garnett classifier that matches the species and tissue type with their dataset or the Import Garnett classifier that is trained previously.

Next, click the Create button to create the classifier file. Once the correct classifier file has been created, select Finish to start running the task.

If you would like to classify cell type with the classifier obtained from the same analysis, choose Project classifiers, and click on ‘Select data node’ to choose the classifier in the analysis. Details on how to train a classifier can be found here. Then click the Finish button.

The output data node named Classify result, it contains the cell type annotation attributes, "cell_type" is the cell type assignments directly from Garnett model. While "cluster_ext_type" is the cell type that's determined by expanding cell type assignments to nearby cells using Louvain clustering. Downstream analysis tasks such as normalization and PCA can be performed on the Classify result data node.

References

1. https://www.nature.com/articles/s41592-019-0535-3

2. https://cole-trapnell-lab.github.io/garnett/classifiers/

3. https://cole-trapnell-lab.github.io/garnett/docs/#constructing-a-marker-file

To train a classifier with a list of biomarkers with your own dataset:

Select any non-normalized single cell data node, choose the Train classifier task in the Classification section.

Similar to previously described steps, first time users will be asked to create the Marker file, users may add them as library files. Marker files should be a .txt file with the marker information in correct format. The same example in the Garnett tutorial of a simple valid Marker file is provided here. Here is an example:

Note:

1. Cell type name line starts with ">", followed by the name of the cell type, cell type name can NOT include hyphen (-).

2. Definition line starts with a keyword e.g. "expressed", "not expressed" e.g. it is case sensitive. It followed by a ":" and space. Each gene name is followed by a comma and space, gene name is case sensitive. The line cannot be ended with comma or space.

For more details on how to construct a Marker file, please refer to Garnett tutorial[3]. Choose a Local file option and browse to the file location, click the Finish button to start running the task as default.

Once the task has finished, click the Classifier data node and choose the Task report in the Task results section.

There are two parts in the task report: the marker evaluation plot and the classification gene table. The marker evaluation plot provides some key information about whether the chosen markers are optimal. Ambiguity scores are calculated for each of the markers which indicates how many cells receive ambiguous labels when this marker is included. The classification gene table may give a hint to which genes are chosen as the relevant genes for distinguishing between different cell types.

Other adjustable parameters in this task include:

Number of Unknown: this tells Garnett how many outgroup cells it should compare against; the default is 500. For a dataset with fewer cells, the number should be smaller.

ScType is "a fully-automated and ultra-fast cell-type identification based solely on a given scRNA-seq data, along with a comprehensive cell marker database as background information."[1]. It allows accurate cell typing in scRNAseq and spatial transcriptome data[2]. Our implementation builds on the original tool, but allows for improved performance on larger datasets.

ScType cell classification

The task can be called from any non-normalised counts node, in the Classification menu on the right.

• Click on the Counts node.

• Click on Classification> ScType cell classification in the toolbox.

• Select the marker database from the SC Type database drop-down.

• Use the Tissue selector to the select the appropriate tissue or tissues by checking the boxes. The options are automatically identified from the marker database provided.

• You can choose Classify all samples together if you want the task to consider all your observations (cells) as one sample, for example if the clustering attribute was computed on all samples at once.

• The task can use an existing attribute for the classification, select the node contains the attribute to use for cell clustering. Choose the specific attribute from the Choose attribute drop-down list.

• You can choose to recompute the clustering attribute within the task by selecting Recompute.

• Select the appropriate resolution parameter for the Leiden clustering.

• Click Finish to run the task.

The task outputs an annotation for each cell in the dataset. If a cell does not pass the significance threshold it will be classified as 'unknown'. The annotation is saved as 'sctype_classification', and is available from the node. You can use Annotation/Metadata>Publish cell attributes to project in the toolbox to make the attribute available at the Analysis level.

Our implementation of ScType allows the user to provide their own marker databases for classification. Please ensure that the gene names used in the marker database match the names in the annotation used for the dataset. In principle, this can be used to classify cells from any species. The marker database present by default supports human and mouse only.

The task expects a marker database file in either .txt or .tsv format. If you are using the original ScType marker database as a template, please save the resulting file in either of these formats.

The file should have 5 columns: tissueType, cellName, geneSymbolmore1, geneSymbolmore2, shortName.

• tissueType: Contains tissues being typed. This column will be parsed te determine the Tissue options in the task.

• cellName: Contains the name of each cell type in each tissue. If a cell type is found in more than one tissue, it will be found more than once in this column.

• geneSymbolmore1: Contains the positive markers for the cell type in that tissue, comma separated.

To add a ScType marker database, first upload it to your ICA project as described . After creating your Study and linking it to the correct ICA project:

• Click on the Add Data button.

• Then Select from ICA project.

• Select Library file > ScType from the drop down menu.

• Navigate to the file location and select the checkbox to the left of the file name.

• Click Add selected data at the bottom right of the page.

The newly added gene marker database will now be available form the dropdown menu in the task setup page.

Overall, the task works by assigning cell type labels (based on marker genes) to clusters. The max value of the typescore for cells within a cluster will define the assigned cell type. Scores for all cell type markers will be computed for each cluster, with the highest scoring cell type assigned to the cluster. Due to this, multiple clusters may be assigned the same cell type (eg. if the data was overclustered).

It is possible to get a cluster with an unknown cell type (e.g. using a custom sctype database where the gene names are not canonical such that the gene names do not match the dataset). Or if the cluster does not meet the significance threshold.

The scType algorithm task results are consistent with the python implementation of scType given the same input and defaults described. The scType webapp has different default methods and parameters, so comparisons may not be consistent.

A table of results for the cell typing scores can be viewed in the task report page. For each cluster, the top 3 cell types are shown ranked by score. Score is the sum of enrichment scores across all cells in the cluster, normalized by cell count. Confidence is the percentage of total positive score attributed to that cell type. The "Other" column summarizes all remaining cell types not in the top 3.

[1] Ianevski, A., Giri, A. K., & Aittokallio, T. (2022). Fully-automated and ultra-fast cell-type identification using specific marker combinations from single-cell transcriptomic data. Nature communications, 13(1), 1246.

[2] Nader, K., Tasci, M., Ianevski, A., Erickson, A., Verschuren, E. W., Aittokallio, T., & Miihkinen, M. (2024). ScType enables fast and accurate cell type identification from spatial transcriptomics data. Bioinformatics, 40(7), btae426.