VICatMix: variational Bayesian clustering and variable selection for discrete biomedical data.

Area of Science:

• Computational biology
• Bioinformatics
• Statistical genetics

Background:

• Clustering biomedical data is vital for patient stratification in precision medicine.
• High-dimensional categorical data, like 'omics data, require computationally efficient algorithms.
• Existing methods struggle with scalability and accuracy for complex datasets.

Purpose of the Study:

• To introduce VICatMix, a variational Bayesian finite mixture model for categorical data clustering.
• To enhance computational efficiency and scalability in clustering high-dimensional biomedical data.
• To enable accurate patient stratification and discovery of disease subtypes.

Main Methods:

• Developed VICatMix, a variational Bayesian finite mixture model for categorical data.
• Implemented variational inference for computationally efficient training and scalability.
• Incorporated variable selection, summarization, and model averaging for improved performance.

Main Results:

• VICatMix outperforms existing methods in computational time and scalability while maintaining accuracy.
• The model effectively performs variable selection on high-dimensional, noisy data.
• Demonstrated utility in cancer subtyping and driver gene discovery using The Cancer Genome Atlas data.

Conclusions:

• VICatMix offers a computationally efficient and accurate solution for clustering high-dimensional categorical biomedical data.
• The model facilitates precise patient stratification and the discovery of novel disease subtypes through integrative analysis.
• VICatMix is available as an R package for broader research application.