Spectrum: fast density-aware spectral clustering for single and multi-omic data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Clustering omic data is crucial for identifying disease subtypes in precision medicine.
• Integrating multi-omic data presents challenges in identifying shared structures and reducing noise.
• Spectral clustering is increasingly applied to single-cell RNA sequencing data for transcriptome analysis.

Purpose of the Study:

• To develop a flexible and effective spectral clustering tool for both single and multi-omic data.
• To address the challenge of integrating complex omic datasets for robust subtype identification.

Main Methods:

• Introduced Spectrum, a novel spectral clustering method utilizing a self-tuning density-aware kernel.
• Employed tensor product graph data integration and diffusion to reduce noise and reveal underlying structures.
• Developed a new method for determining the optimal number of clusters (K) through eigenvector distribution analysis, applicable to Gaussian and non-Gaussian data.

Main Results:

• Spectrum demonstrated improved runtimes and superior clustering results across 21 real-world expression datasets compared to existing methods.
• The self-tuning kernel enhances similarity between points with common nearest neighbors.
• The method effectively reduces noise and identifies underlying structures in complex omic data.

Conclusions:

• Spectrum offers an effective and efficient solution for clustering omic data, facilitating precision medicine.
• The tool's ability to handle both single and multi-omic data and automatically determine the number of clusters enhances its utility.
• Spectrum provides a valuable resource for researchers in bioinformatics and computational biology for disease subtype discovery.