Predict Cell Types
The Predict Cell Types tool uses a Cell Type Classifier (![Image cell_type_classifier_16_n_p](./cell_type_classifier_16_n_p.png)
![Image expression_matrix_track_16_n_p](./expression_matrix_track_16_n_p.png)
![Image expr_matrix_spliced_unspliced_16_n_p](./expr_matrix_spliced_unspliced_16_n_p.png)
It can be found in the Toolbox here:
Gene Expression () | Cell Type Classification (
) | Predict Cell Types (
)
The following options can be adjusted (figure 6.2):
Figure 6.2: The options in the dialog of the Predict Cell Types tool. A Cell Type Classifier for human data downloaded from the Reference Data Manager has been selected.
- Cell type classifier. A classifier downloaded from the Reference Data Manager (see The Reference Data Manager) or produced by the Train Cell Type Classifier tool (see Train Cell Type Classifier). Note that the features in the input matrix and those used for training the classifier should be matching, see Features used for training and prediction.
- Tissue type. Many cell types from the QIAGEN Cell Ontology (see The QIAGEN Cell Ontology) are associated with specific tissues. When one or more tissues are selected, cell types associated with other tissues will no longer be predicted. For example, hepatocytes are associated with the liver. If Tissue type = Heart, then no cells will be predicted as "hepatocytes". A list of cell types that would have been predicted had no tissue type been specified can be found on the History view (
) of the outputs.
Note that for this option to have an effect, it must be possible to link cell types from the classifier with cell types from the QIAGEN Cell Ontology. This linking works by looking at whether the classifier cell type has the same name as, or is a known synonym of a cell type from the QIAGEN Cell Ontology. Linking is always possible for classifiers downloaded from the Reference Data Manager.
The tool outputs:
- A Cell Clusters (
) element containing two categories:
- "Cell type (all)" containing the predicted cell types, for each cell in the input matrix.
- "Cell type (high confidence)" containing the same predicted cell types, but with predictions with low probability being replaced with "Unknown". This can be useful for detecting novel cell types.
- Optionally, a Cell Annotations (
) element with the probabilities assigned for a subset of relevant cell types, for each cell in the input matrix. A cell type is considered relevant if:
- The cell type is the predicted cell type for at least one cell in the "Cell type (all)" category, or
- There is at least one cell with a probability of at least
for the cell type.
Using the outputs, the cells can be colored in a Dimensionality Reduction Plot (see Dimensionality reduction) by the predicted cell type (using the Cell Clusters) or the probability of having a specific cell type (using the Cell Annotations). Guided by the probabilities and other information, such as the clusters produced by the Cluster Single Cell Data tool (see Cluster Single Cell Data), the predicted cell types can be manually refined in the Dimensionality Reduction Plot (see Manual Annotation).
For details on how cell types are predicted, see SVMs for cell type classification.