Uncertainpy: A Python Toolbox for Uncertainty Quantification and Sensitivity Analysis in Computational Neuroscience.

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Computational Biology

Background:

• Computational neuroscience models often have numerous parameters with limited experimental constraints.
• Quantifying parameter uncertainty and its impact on model outputs is crucial but not standard practice.
• Existing methods for uncertainty quantification and sensitivity analysis can be computationally intensive.

Purpose of the Study:

• To introduce Uncertainpy, an open-source Python toolbox for uncertainty quantification (UQ) and sensitivity analysis (SA) in neuroscience.
• To provide an accessible tool for neuroscientists to analyze parameter uncertainty in their models without code modification.
• To demonstrate the application and broad utility of Uncertainpy across diverse neuroscience modeling scenarios.

Main Methods:

• Utilizes polynomial chaos expansions for efficient UQ and SA, outperforming standard Monte Carlo methods.
• Enables analysis of existing models without altering their equations or implementation.
• Incorporates built-in capabilities for analyzing salient neural features (e.g., spike timing, action potential width) beyond raw output.

Main Results:

• Uncertainpy successfully performed UQ and SA on three distinct neuroscience models: Hodgkin-Huxley, a multi-compartmental interneuron (NEURON), and a recurrent network (NEST).
• The toolbox efficiently quantifies uncertainty and sensitivity of both raw model outputs and specific neural features.
• Demonstrates ease of use and integration for both common and custom neuroscience models and features.

Conclusions:

• Uncertainpy significantly lowers the barrier to entry for performing rigorous UQ and SA in computational neuroscience.
• The toolbox enhances the reliability and interpretability of computational neuroscience models by characterizing parameter influences.
• Facilitates broader adoption of robust quantitative analysis methods within the neuroscience research community.