Consensus clustering applied to multi-omics disease subtyping.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Omics data analysis presents challenges due to diversity and method variability.
• Consensus strategies offer a path to reconcile multiple analytical inputs for robust results.

Purpose of the Study:

• Introduce ClustOmics, a novel consensus clustering tool.
• Apply ClustOmics for cancer subtyping using multi-omics data.

Main Methods:

• ClustOmics utilizes a non-relational graph database for integrating omics data and clustering results.
• Employs an evidence accumulation strategy based on sample co-occurrence in clusters.
• Reorganizes data into consensus clusters using co-occurrence scores as a similarity measure.

Main Results:

• ClustOmics successfully applied to multi-omics cancer data from TCGA.
• Demonstrated robustness in reconciling heterogeneous input partitions into high-quality consensus clusters.
• Outperformed or complemented state-of-the-art tools like COCA in cancer subtyping.

Conclusions:

• ClustOmics provides a valuable approach for integrating diverse omics data and clustering methods.
• Enhances the reliability of cancer subtyping by generating robust consensus clusters.
• Offers a flexible solution for leveraging multiple predictions when definitive metrics are absent.