Choosing a clustering resolution is one of the most consequential decisions in single-cell analysis, and one of the most subjective. clustOpt takes the judgment call out of it.

It combines subject-wise cross-validation, principal component splitting, random forests, and silhouette-based cluster quality scoring to recommend a resolution that is reproducible across subjects and free of data leakage.

What it does

• Data-driven selection of the Seurat FindClusters resolution parameter, with no manual marker-gene inspection
• Avoids double dipping by training and predicting in independent PC subspaces
• Preprocessing tuned for both scRNA-seq and CyTOF
• Scales past 200,000 cells using Seurat leverage-score sketching and on-disk BPCells matrices
• Available as an R package and as a Docker image

Gill, Shin, Agrawal, Thomas (2025). Optimizing Clustering Resolution for Multi-subject Single Cell Studies. Presented at ISMB 2025.

Repository · Documentation