CellSavior: an open-source framework for improved cell calling in targeted single-cell DNA sequencing

Single-cell DNA sequencing (scDNA-seq) has emerged as a key approach for studying complex genomic variation, including tumor heterogeneity. Despite its increasing commercial availability, exemplified by the Tapestri platform, analytical reproducibility remains limited due to reliance on proprietary and non-transparent pipelines. Among preprocessing steps, cell calling is particularly critical, as it determines which barcodes correspond to true cells rather than background noise and directly shapes all downstream analyses. However, existing methods typically rely on empirical thresholding and exhibit inconsistent performance across datasets. These limitations underscore the need for a robust and generalizable cell calling framework for reliable scDNA-seq analysis.

We developed CellSavior, an open-source cell calling algorithm with an integrated downstream pipeline that produces standardized outputs. CellSavior employs an ensemble clustering strategy across multiple distance metrics, combined with confidence-weighted merging, to generate a consensus classification that reduces bias and improves robustness. Across 33 public and in-house single-cell DNA sequencing datasets, CellSavior recapitulated cell annotations generated by commercial software (CellFinder) in 31 cases (93.9%). In the remaining discordant cases, CellSavior yielded more biologically interpretable annotations, rescuing previously undetected clusters enriched for key hotspot genomic alterations.

The source code for CellSavior is freely available at https://github.com/dauer-tgilab/CellSavior; DOI: https://doi.org/10.5281/zenodo.22744146.

Supplementary data are available at Bioinformatics online.