single-cell-rna-qc

Single-cell RNA-seq QC applies MAD outlier detection to flag low-quality cells by counts, detected genes, and mitochondrial fractions. A hard mitochondrial percentage cutoff removes stressed cells, while customizable thresholds preserve rare populations.

Filtering low-quality cells from an h5ad file is done by running a provided Python script using scanpy and anndata. It calculates scverse-style QC metrics, detects outliers, filters poor-quality cells, and outputs before/after visualizations plus a filtered h5ad file.

Yes, the QC workflow supports both standard .h5ad files and 10X .h5 inputs. It calculates counts, detected genes, and mitochondrial/ribosomal/hemoglobin fractions across both formats using scanpy and anndata.

To keep rare cell populations during scRNA-seq quality filtering, apply MAD-based outlier detection rather than simple percentage cutoffs. This identifies poor-quality cells by deviation from median metrics while applying a hard mitochondrial percentage cutoff to remove dying cells.

Calculated QC metrics for single-cell RNA-seq data include total counts, detected genes, and mitochondrial, ribosomal, and hemoglobin fractions. These metrics are computed using scanpy and anndata to evaluate cell quality before applying MAD outlier filters.

MAD-based outlier detection adapts to dataset-specific distributions, making it more robust than fixed thresholds for scRNA-seq QC. It identifies outliers based on median absolute deviation across QC metrics, while an additional hard mitochondrial cutoff targets stressed or dying cells.