A scalable, multi-resolution consensus clustering approach for prioritizing robust signals from high-throughput...

Area of Science:

• Computational biology
• Bioinformatics
• Data science

Background:

• Modern biology generates high-dimensional datasets from large-scale screening.
• Analyzing these complex datasets is challenging due to hierarchical structures, unknown group numbers, and high-dimensional noise.

Purpose of the Study:

• To present Untangled, an unsupervised consensus clustering tool.
• To address challenges in analyzing high-dimensional biological screening data.

Main Methods:

• Untangled aggregates clustering solutions across granularities.
• It constructs a stability-based representation.
• Cluster number optimization and robustness evaluation are performed.

Main Results:

• Untangled reliably recovers underlying biological relationships.
• It resolves stable, meaningful substructure within datasets.
• It effectively prioritizes robust clusters with shared biological mechanisms and conserved phenotypic responses.

Conclusions:

• Untangled is a scalable framework for cluster discovery.
• It guides efficient follow-up investigations from high-dimensional biological datasets.
• It outperforms alternative clustering approaches in benchmarking studies.