One thing we have noticed with cell type annotation using SingleR is that there regardless of the appropriateness of the reference, SingleR will always favor assigning a score rather than not labeling a cell. Although there are metrics that we have found may be helpful in determining if a reference is completely inappropriate (looking at the delta between the top score and median score), we would like to know if other types of cell annotation will be less likely to assign a cell type in general and will fail more easily.
We should do a quick check of one of the marker genes based methods using the same Gawad libraries that we have been using in #210 and check the number of cells assigned to each type compared to using SingleR.
The methods that we know that use marker genes are:
We should do this in an R notebook where we read in the annotated sce file that has been annotated using SingleR, assign cell types using one of the above methods, and then show the distributions of assignments. We can do a quick comparison of labels between the two methods where we are particularly interested in comparing how both methods do when using references that are inappropriate.
One thing we have noticed with cell type annotation using
SingleRis that there regardless of the appropriateness of the reference,SingleRwill always favor assigning a score rather than not labeling a cell. Although there are metrics that we have found may be helpful in determining if a reference is completely inappropriate (looking at the delta between the top score and median score), we would like to know if other types of cell annotation will be less likely to assign a cell type in general and will fail more easily.We should do a quick check of one of the marker genes based methods using the same Gawad libraries that we have been using in #210 and check the number of cells assigned to each type compared to using
SingleR.The methods that we know that use marker genes are:
We should do this in an R notebook where we read in the annotated sce file that has been annotated using SingleR, assign cell types using one of the above methods, and then show the distributions of assignments. We can do a quick comparison of labels between the two methods where we are particularly interested in comparing how both methods do when using references that are inappropriate.