Omada: robust clustering of transcriptomes through multiple testing.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Cohort studies increasingly collect biosamples for molecular profiling, revealing significant molecular heterogeneity.
• High-throughput RNA sequencing generates large datasets crucial for understanding disease mechanisms.
• Analyzing complex transcriptomic data requires expertise in machine learning and extensive computational experimentation.

Purpose of the Study:

• To develop Omada, a suite of tools designed to automate unsupervised clustering of transcriptomic data.
• To make robust transcriptomic data analysis more accessible through automated machine learning functions.
• To assist researchers without extensive machine learning expertise in performing exploratory clustering analysis.

Main Methods:

• Developed Omada, a toolkit with automated machine learning-based functions for unsupervised clustering.
• Tested Omada's efficiency using 7 diverse RNA sequencing datasets with varying expression signal strengths.
• Evaluated the toolkit's ability to identify stable partitions and biological distinctions in transcriptomic datasets.

Main Results:

• Omada accurately reflected the number of stable partitions in datasets with discernible subgroups.
• In datasets with less clear biological distinctions, Omada identified stable subgroups with distinct expression profiles and clinical associations.
• The toolkit also detected signs of problematic data, such as biased measurements, in challenging datasets.

Conclusions:

• Omada successfully automates robust unsupervised clustering of transcriptomic data.
• The toolkit enhances the accessibility and reliability of advanced transcriptomic analysis for researchers lacking extensive machine learning expertise.
• Omada is available for implementation at http://bioconductor.org/packages/omada/.