This shiny application allows the user to perform: i) cell type identification with their own query data using pre-trained models,
and ii) sample size learning to determine the number of cells required for accurately discriminating between two cell subtypes anywhere in a cell type hierarchy.
To predict cell types with your own single cell query dataset:
Step 1: select an appropriate pre-trained model.
Step 2: upload your single cell dataset. (or use the demo dataset we provide by ticking the demo dataset box)
Step 3: click the start prediction button.
Note: When the progress bar is finished, a t-SNE plot will appear on the screen. This indicates the prediction result is available to download.
We provide i) predicted cell types at the finest level and ii) predicted cell types at all levels of the cell-type hierarchy.
You can download each of these prediction results by clicking on the respective download button.
Please upload the expression matrix and cell type labels.
The expression matrix should be a csv file, with first column being gene symbols as defined in NCBI Gene database and the first row being sample names.
The cell type labels should be a csv file, with first column being cell type labels (as shown in the example).
You could also explore the tool using the demo dataset we provide, by ticking the demo dataset box.
A hierarchical cell type tree will be automatically generated once both datasets are uploaded.
Please select the level you would like distinguish at (1 being the level immediately under root node, the most coarse level) and click the button to fit the sample size curve.
Once the curve is fitted and plotted, you can type in an accuracy value to obtain an estimation of the number of samples needed to achieve this accuracy at the level you have chosen.