Update Cell Type Classifier
The Update Cell Type Classifier tool takes one Cell Type Classifier () element as input and outputs a new Cell Type Classifier () element containing updated training data.
The tool can be found in the Toolbox here:
Toolbox | Single Cell Analysis () | Gene Expression () | Cell Type Classification () | Update Cell Type Classifier ()
In the first wizard step `Update classifier', the following options can be adjusted:
- Remove cell types. The cells annotated with the selected cell types are removed from the training data. The classifier will be re-trained.
- Remove samples. The cells from the selected samples are removed from the training data. Cell types for which all training cells are from the removed samples will be subsequently removed. The classifier will be re-trained.
- Rename cell types. One or more cell types are renamed. The Add button can be used to add additional cell types to be renamed.
- Map cell types to QIAGEN Cell Ontology. When this is checked, cell types will be translated, if possible, to the QIAGEN Cell Ontology (see The QIAGEN Cell Ontology). The translation attempts to match each cell type with a QIAGEN cell type based on the name and known synonyms. For example, `alveolar epithelial cells' are also called `pneumocytes'. If this option is selected, the `alveolar epithelial cells' cell type, if present, will be named `pneumocytes'. This option can be useful when standardizing cell types from different sources.
Cell types with the same name will be merged into one and the classifier will be re-trained. Cell types are merged when:
- A cell type is renamed to an existing name.
- Two cell types are renamed to the same name.
- Two cell type names are synonyms for the same QIAGEN cell type and cell types are mapped to the ontology.
In the next wizard step `Validation', the tool can be configured to validate the resulting classifier, as done for the Train Cell Type Classifier tool (see Train Cell Type Classifier). Regressions are calculated using the input classifier.
The options above can be mixed and matched to obtain the desired output. For example, a cell type can be first renamed and then the new name can be mapped to the ontology.
Renaming and/or mapping cell types to the ontology does not require updates to the underlying classification model. However, the classifier will be re-trained when the training data is changed:
- Cells are removed, by removing cell types and/or samples.
- Cell types are merged. Note that the classifier will be trained with approximately 50 cells per cell type, see Train Cell Type Classifier. When cell types are merged, some cells might be discarded during training.
The Update Cell Type Classifier tool produces a report summarizing the performed updates and, if validation data was provided, the performance of the updated classifier as described in The output of Train Cell Type Classifier.
The Update Cell Type Classifier tool can also be started from the table view of the Cell Type Classifier () element (see The Cell Type Classifier element), where the dialog is automatically filled in with relevant selections:
- The selected cell types can be removed or renamed.
- The sample over which the mouse hovers can be removed.