SCHNEL

SCHNEL performs scalable, automated graph-based clustering of high-dimensional single-cell, cytometry, and single-cell RNA sequencing data to identify distinct cell populations.

Key Features:

• Scalability: Handles datasets containing millions of cells, demonstrated on datasets of 3.5 million and 17.2 million cells.
• Automated Clustering: Automates clustering without requiring manual estimation of the number of clusters.
• Hierarchical Stochastic Neighbour Embedding (HSNE): Transforms large high-dimensional datasets into hierarchies of smaller subsets that preserve the original data manifold.
• Integration with Graph Clustering: Combines HSNE hierarchical representations with graph clustering using Louvain community detection.
• Performance: Outperformed three other cytometry clustering tools across seven cytometry datasets in comparative studies.
• Versatility: Validated on cytometry, single-cell RNA sequencing, and the MNIST benchmark dataset as a general-purpose clustering approach.

Scientific Applications:

• Cell population identification: Identification of distinct cell populations in large-scale cytometry and single-cell RNA-seq datasets.
• Cellular heterogeneity analysis: Characterization of cellular heterogeneity in genomics and systems biology studies.
• Benchmarking and validation: Application to benchmarking datasets such as MNIST to validate clustering performance.

Methodology:

Transforms datasets into hierarchies using Hierarchical Stochastic Neighbour Embedding (HSNE) to preserve the data manifold and applies graph clustering with Louvain community detection, with automated cluster-number estimation.