A HORSESHOE MIXTURE MODEL FOR BAYESIAN SCREENING WITH AN APPLICATION TO LIGHT SHEET FLUORESCENCE MICROSCOPY IN BRAIN...

Area of Science:

• Neuroscience
• Biostatistics
• Microscopy

Background:

• Light sheet fluorescence microscopy enables whole-brain imaging.
• Current statistical methods for brain imaging analysis often use binary classification (significant/non-significant activation).
• Binary classification may oversimplify findings by discarding weak but potentially important signals masked by noise.

Purpose of the Study:

• To develop a new Bayesian approach for classifying brain regions into multiple tiers of relevance.
• To overcome the limitations of binary classification in analyzing complex brain imaging data.
• To provide more biologically meaningful and interpretable results in brain region activation studies.

Main Methods:

• A novel Bayesian approach combining shrinkage priors and mixture models.
• Utilizes a discrete mixture of continuous scale mixtures for priors.
• Develops a cluster shrinkage version of the horseshoe prior.

Main Results:

• The proposed method allows for tiered classification of brain regions based on activation relevance.
• It offers a more general framework for Bayesian sparse estimation.
• Reduces the number of necessary shrinkage parameters and facilitates information sharing across brain regions.

Conclusions:

• The new Bayesian approach provides more nuanced and interpretable results for brain imaging studies.
• It effectively discriminates between active and inactive regions while ranking discoveries by importance.
• This method enhances the analysis of whole-brain imaging data obtained via light sheet fluorescence microscopy.