Cell-level annotation doesn't need to come after integration. Additionally, the workflow could benefit from annotating cells, then assigning cell types to clusters, integrating, and then checking how well the pre-integration annotated clusters align with each other post-integration.
The new workflow would be something like:
- QC + doublet detection + clustering
- Annotation
- Integration
- Re-clustering
- Check alignment of pre-integration and post-integration clusters
- Re-assign cell types to post-integration clusters
- Pseudo + DE + FEA
Cell-level annotation doesn't need to come after integration. Additionally, the workflow could benefit from annotating cells, then assigning cell types to clusters, integrating, and then checking how well the pre-integration annotated clusters align with each other post-integration.
The new workflow would be something like: