SHARE-Topic: Bayesian interpretable modeling of single-cell multi-omic data.

Area of Science:

• Genomics
• Epigenetics
• Computational Biology

Background:

• Multi-omic single-cell technologies offer deep insights into gene regulation.
• Noisy and sparse data present significant statistical challenges for analysis.
• Extracting biological knowledge requires advanced computational methods.

Purpose of the Study:

• To develop a Bayesian generative model, SHARE-Topic, for analyzing multi-omic single-cell data.
• To address statistical challenges posed by noisy and sparse multi-omic datasets.
• To identify co-variation patterns across omic layers and interpret data complexity.

Main Methods:

• Utilized Bayesian inference and topic modeling within the SHARE-Topic framework.
• Applied the model to single-cell data from diverse technological platforms.
• Developed methods for low-dimensional data representation and association analysis.

Main Results:

• SHARE-Topic successfully identified common patterns of co-variation between different omic layers.
• The model generated low-dimensional representations that recapitulated known biological information.
• Established associations between genes and distal regulatory elements within individual cells.

Conclusions:

• SHARE-Topic provides an interpretable framework for understanding complex multi-omic single-cell data.
• The model effectively addresses data noise and sparsity, enabling robust biological discovery.
• Facilitates the study of epigenetic mechanisms and gene regulation at a single-cell level.