moCluster: Identifying Joint Patterns Across Multiple Omics Data Sets.

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• Multiple omics approaches are increasingly used to study complex biological systems.
• Efficient computational methods for integrating diverse omics data are lacking.

Purpose of the Study:

• To introduce moCluster, a novel algorithm for discovering joint patterns across multiple omics data.
• To demonstrate moCluster's efficiency and accuracy compared to existing methods like iCluster/iCluster+.

Main Methods:

• moCluster uses multiblock multivariate analysis to identify latent variables representing joint patterns.
• These latent variables are then used by an ordinary clustering algorithm to find joint clusters.
• The algorithm was tested using simulated data and applied to real-world datasets (NCI-60, colorectal cancer).

Main Results:

• moCluster demonstrates superior performance and speed (100x-1000x faster) over iCluster/iCluster+, without nondeterministic solutions.
• Clustering of NCI-60 proteomic and transcriptomic data revealed distinct cellular phenotypes, including differences in doubling time and drug response.
• Analysis of colorectal cancer data identified four molecular subtypes, including a novel microsatellite instability subtype with high immune-related gene/protein expression (e.g., PDL1).

Conclusions:

• moCluster effectively integrates multiple omics data to uncover complex biological patterns and identify novel molecular subtypes.
• The identified subtypes, particularly in colorectal cancer, highlight the complexity of oncogenesis and the necessity of holistic, multi-data analysis strategies.
• This approach has significant implications for understanding disease mechanisms and developing targeted therapies.