Label Transfer
Reference-atlas annotation with HLCA and related notebooks
What it does
This workflow groups together atlas-based label transfer notebooks for annotating new scRNA-seq datasets. The committed materials emphasize HLCA-based lung annotation and a more general reference-to-query transfer pattern using Seurat- or SCANVI-style tooling.
When to use it
Use this workflow when clustering is complete but biological labels still need to be transferred from a curated reference atlas. It is most useful for lung datasets or for projects that already have a trusted annotated reference object and want a branch notebook focused on transferring labels rather than redoing preprocessing from scratch.
Prerequisites
- Source folder:
scRNAseq_label_transfer_branch - Main files:
- Expected inputs:
- a query dataset
- a pre-annotated reference atlas
- metadata with sample and batch information for the HLCA-oriented path
- File formats explicitly mentioned in the committed materials include
.h5ad,.loom, and.rds
Steps
Choose the notebook that matches your reference setup
This folder contains two parallel entry points rather than one monolithic tutorial:
HLCA_atlas_annotation.ipynbfor Human Lung Cell Atlas-based annotationscRNAseq_label_transfer.ipynbfor a more general reference-to-query transfer pattern
Both expect a processed query dataset plus a curated annotated reference. The HLCA notes also call out metadata with sample and batch information.
Use the HLCA notebook for lung-specific annotation
The HLCA branch is intended for lung datasets and the committed README summarizes its workflow as:
load and preprocess the query dataset
align query data to the HLCA reference with PCA and Harmony integration
transfer labels with scANVI or Symphony
generate visualization plots
The short HLCA introduction file reinforces the same scope: query scRNA-seq input, a pre-annotated HLCA reference, supporting metadata, and per-cell cell type annotations as output.
Use the general notebook for non-HLCA references
The second notebook covers a broader label-transfer pattern that is not specific to lung atlases. The committed README describes it as:
normalize and log-transform the query data
integrate query and reference with PCA/CCA (Seurat) or variational inference (SCANVI)
transfer labels and assess confidence
visualize the combined embedding
Review the transferred labels, confidence summaries, and embeddings
Across both notebooks, the intended deliverables are the same: predicted cell type labels on the query dataset plus visualization outputs that let you inspect whether the reference mapping looks plausible. This page stays at the routing-and-rationale level because the implementation details live inside the committed Jupyter notebooks rather than in rendered markdown.
Gotchas / notes
- The committed source material for this workflow is thinner than some other
scrnafolders: the README is high-level, and most implementation detail lives inside Jupyter notebooks. - There are no committed local figure assets in this folder.
- The HLCA note is lung-focused, so projects outside that assay/tissue context will likely rely more heavily on the general notebook.
- The HLCA introduction file is very brief and partially incomplete, so the notebooks themselves remain the main source for execution detail.