Simulation-based inference for efficient identification of generative models in computational connectomics.

Area of Science:

• Neuroscience
• Computational Biology
• Systems Biology

Background:

• Connectomics research generates vast neuronal connectivity data.
• Hypothesis testing in connectomics often relies on simulating neuronal networks with manual parameter tuning.
• Existing methods face challenges in efficiently identifying parameters for generative network models.

Purpose of the Study:

• To introduce simulation-based Bayesian inference (SBI) as a method for parameter inference in computational connectomics.
• To enable efficient derivation and testing of hypotheses about neuronal wiring principles.
• To provide a quantitative approach for constraining model parameters with empirical connectivity data.

Main Methods:

• Utilized simulation-based Bayesian inference (SBI) for parameter inference.
• Applied SBI to an in silico model of the rat barrel cortex using in vivo connectivity measurements.
• Employed machine learning to estimate posterior distributions over model parameters.

Main Results:

• SBI successfully identified a wide range of data-compatible wiring rule parameters.
• The posterior distribution revealed biologically plausible parameter interactions.
• The method enabled the generation of experimentally testable predictions and ruled out invalid wiring hypotheses.

Conclusions:

• SBI offers a quantitative and efficient method for parameter inference in connectomics.
• This approach facilitates the analysis of parameter relationships and hypothesis testing.
• SBI is broadly applicable to various generative models in connectomics research.