Benchmarking and optimizing organism wide single-cell RNA alignment methods

Analysis of human scRNA-seq data collected from public respositories and standardized by Diaz-Mejia JJ et al. (2025) for the paper "Benchmarking and optimizing organism wide single-cell RNA alignment methods" presented at the LMRL Workshop at the International Conference on Learning Representations (2025).

File scref_h5ad_ICLR_2025.tar.bz2 contains 46 h5ad files, one for each study analysed that we call scREF. File names contains {first author, last name}_{journal}_{year}_{Pubmed ID}. Files contain adata.obs['included_scref_train'] annotations indicating if the cell was included in downsampled training and benchmark analyses.

File BA-scVI_ICLR_2025.tar.bz2 contains the analysis of the 46 scREF datasets (i.e. normal tissue) and the 12 studies included in scMark v2.0 (i.e. normal and cancer tissues), which is available here: https://zenodo.org/records/7795653

Code is available in the scref_h5ad_ICLR_2025.tar.bz2 file for each method evaluated and for our newly developed method called Batch Adversarially trained single-cell Variational Inference (BA-scVI)at https://github.com/PhenomicAI/bascvi