Algorithm 1: Reference data selection
Input: $N\times M$ data matrix X; parameter vector a.
1. Filter out cells with library size greater than a1-th percentile.
2. Remove genes with mean expression less than a2-th percentile.
3. Remove genes with less than a3-th percentile nonzero cells.
4. Keep cells with library size greater than the a4-th percentile.
5. Keep genes with nonzero proportion greater than a5-th percentile.
The default parameter values are$a_1=95,a_2=25,a_3=15,a_4=5,a_5=50$.
Output: $N\prime \times M\prime$ reference data matrix Y.