scICE

scICE evaluates clustering consistency in single-cell RNA sequencing (scRNA-seq) data and provides consistent clustering results by narrowing the number of candidate clusters to explore.

Key Features:

• Clustering consistency evaluation: scICE quantifies clustering inconsistency among trials caused by stochastic processes in clustering algorithms.
• Computational efficiency: scICE achieves up to a 30-fold speed improvement compared to conventional consensus clustering-based methods such as multiK and chooseR.
• Candidate cluster reduction: scICE substantially narrows the number of clusters to explore, enabling focus on a narrower set of more reliable candidate clusters.
• Scalability: scICE can be applied to large scRNA-seq datasets including those with over 10,000 cells.

Scientific Applications:

• Reliable clustering of large scRNA-seq datasets: scICE identifies consistent clustering results in real and simulated scRNA-seq datasets, reducing computational burden while generating more robust results.
• Consensus clustering alternative: scICE provides a computationally efficient alternative to conventional consensus clustering-based methods for determining optimal cluster numbers in scRNA-seq analysis.

Methodology:

scICE evaluates clustering consistency through multi-cluster label consistency evaluation across multiple clustering trials.