Multi-dimensional classification of GABAergic interneurons with Bayesian network-modeled label uncertainty.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning in Biology

Background:

• Interneuron classification is a complex and debated topic in neuroscience.
• Existing classification schemes rely on subjective, categorical features of axonal morphology.
• Previous studies lacked a robust method for multi-dimensional interneuron classification.

Purpose of the Study:

• To develop a computational model for the multi-dimensional classification of interneurons.
• To utilize axonal morphometric parameters for objective interneuron classification.
• To address the challenge of interneuron classification with probabilistic labels due to expert disagreement.

Main Methods:

• Developed a method to predict label Bayesian networks (LBNs) representing probabilistic interneuron classifications.
• Employed a probabilistic consensus approach using LBNs of similar interneurons.
• Introduced 13 new quantitative axonal morphometric parameters as predictors.

Main Results:

• Accurately predicted interneuronal LBNs using axonal morphometric parameters.
• The developed method demonstrated superior performance compared to related work in crisp interneuron classification.
• The new morphometric parameters effectively predict interneuron type and axonal morphology features.

Conclusions:

• The proposed method is suitable for multi-dimensional interneuron classification with probabilistic labels.
• Axonal morphometric parameters provide objective and effective features for interneuron classification.
• This approach offers a data-driven and objective framework for understanding neuronal diversity.