Spectrum

Spectrum performs spectral clustering of single- and multi-omic data to identify disease subtypes and uncover shared structures across omic datasets while reducing noise.

Key Features:

• Adaptive Density-Aware Kernel: Spectrum employs a self-tuning density-aware kernel that enhances similarity between points sharing common nearest neighbors and strengthens connections in the data graph to improve clustering accuracy.
• Tensor Product Graph Integration: Spectrum uses a tensor product graph data integration and diffusion procedure to integrate multiple omic data sources, reduce noise, and reveal underlying structures.
• Optimal Cluster Determination: Spectrum determines the optimal number of clusters (K) via eigenvector distribution analysis, automatically identifying K for both Gaussian and non-Gaussian data structures.
• Performance and Flexibility: Spectrum demonstrated improved runtimes and superior clustering results across 21 real expression datasets and can handle a wide range of data structures.

Scientific Applications:

• Single-cell RNA-seq clustering: Clusters individual cell transcriptomes to uncover cellular heterogeneity.
• Multi-omic data integration: Integrates multiple omic layers to uncover shared structures across datasets while minimizing noise.
• Disease subtype identification and precision medicine: Identifies disease subtypes from patient omic data to support precision medicine and the discovery of potential therapeutic targets.

Methodology:

Spectral clustering using a self-tuning density-aware kernel, tensor product graph integration with diffusion, and eigenvector distribution analysis for automatic K selection.