dropkick

dropkick identifies and filters high-confidence cellular barcodes from ambient barcodes in droplet-based single-cell RNA sequencing (scRNA-seq) datasets to improve quality control and downstream analyses.

Key Features:

• Automated Quality Control: Automates filtering of ambient barcodes while recovering real cells that are near quality thresholds.
• Dataset-Specific Training Labels: Automatically determines training labels per dataset using predictive global heuristics and learns a gene-based representation of cells and ambient noise.
• Cell Probability Scoring: Computes a cell probability score for every barcode to distinguish true cellular signals from background noise.
• Compatibility with AnnData/scanpy and Common Formats: Operates on AnnData objects (scanpy) and accepts .h5ad as well as flat formats such as .csv and .tsv, and outputs .h5ad.
• Benchmarking and Performance: Evaluated on simulated and real-world scRNA-seq datasets and compared to conventional thresholding methods and EmptyDrops, showing improved recovery of rare cell types and exclusion of empty droplets and noisy barcodes.
• Robustness to Dataset Variation: Uses a weakly supervised model to identify genes enriched in ambient barcodes and establish a multidimensional boundary adaptable to varying background noise levels.

Scientific Applications:

• Preprocessing for clustering: Provides filtered cell calls that improve input quality for clustering analyses.
• Differential expression analysis: Produces more reliable cell assignments for downstream differential expression testing.
• Trajectory inference: Delivers cleaner cell inventories for trajectory and lineage inference workflows.
• Rare cell recovery and empty-droplet exclusion: Enhances recovery of rare cell types while excluding empty droplets and noisy barcodes from datasets.

Methodology:

Implements a weakly supervised model that automatically derives dataset-specific training labels from predictive global heuristics, learns a gene-based representation, computes per-barcode cell probability scores, establishes a multidimensional decision boundary and identifies genes enriched in ambient barcodes; operates on AnnData/scanpy inputs (.h5ad, .csv, .tsv) and was benchmarked on simulated and real-world scRNA-seq datasets with comparisons to EmptyDrops.