bio-machine-learning-atlas-mapping

You can map query scRNA-seq data to a reference atlas by using transfer learning with scArches to transfer latent representations and cell type labels from pre-trained scVI references.

The best way to annotate new single-cell datasets is by transferring latent representations and cell type labels via scANVI, which provides confidence scores for each transferred label.

Yes, you must match genes between your query and reference data before performing scArches surgery to ensure accurate latent representation transfer during the mapping process.

Yes, you can use scvi-tools and scanpy to visualize integration quality between datasets after mapping your query data to the reference atlas using transfer learning.

ScArches uses surgical fine-tuning with controlled weight decay to preserve the reference structure while adapting the model to the new query scRNA-seq dataset.

Label transfer limitations depend on confidence scores generated by scANVI, as preserving the reference structure during surgical fine-tuning may restrict mapping novel cell types absent from the reference.