Emma Huhtala, Jugoslava Acimovic, Mikko Lehtimäki, Marja-Leena Linne
Pyramidal neurons are abundant in the neocortex and known to contribute to diverse and complex cognitive functions. To better understand the role of individual pyramidal cell types we need neuron models that can be efficiently and reliably simulated and analyzed when embedded into circuit-level models yet that maintain the necessary structural and functional complexity. In this study we contribute to this goal the following ways: **1)** We review and compare over 50 published mammalian cortical pyramidal neuron models available in the literature and public repositories (with the focus on models implemented in the NEURON simulation environment). **2)** We test two recently published tools that tackle critical issues for detailed neuron modeling, the role of model complexity and size and the sensitivity to numerous and occasionally unreliable model parameters.
In goal 1, we compared the models based on the following criteria: the brain area and layer, number of compartments, biophysical properties, software used to simulate the model, and the amount of experimental data used to construct and fine-tune the model. Based on this work, we chose a layer 2/3 pyramidal cell model from the rat somatosensory cortex, acquired from the Blue Brain Project data portal [1], as our test case. We used two recently published toolboxes, Neuron_Reduce and Uncertainpy, to analyze the model. Neuron_Reduce [2] is designed for simplifying the morphology of the model while replicating the model dynamics and accelerating the simulations considerably. Uncertainpy [3] allows the user to examine the impact of uncertainty and sensitivity of the model parameters.
The analyses carried out in this work show that Neuron_Reduce toolbox is an effective tool for simplifying the morphological structure of neuron models. With moderate reduction of the model (preserving 10-25% of the original model compartments) the Neuron_Reduce toolbox simplifies the dendritic structure of the cell while replicating the behavior of the original model. However, a dramatic reduction (preserving 3% of the original model compartments) led to changes in the shape of action potentials. The results obtained with Uncertainpy suggest that the reduced model shows sensitivity to only a subset of model parameters among those that affect the original model. For example, while the original model depends on several somatic conductances, the reduced model is sensitive mainly to sodium and potassium channel conductancesBased on our testing, both toolboxes will be useful tools for analyzing models in neuroscience. In addition, they can help the re-use of compartmental models in new modeling initiatives, particularly when modeling multiple spatiotemporal scales of the brain phenomena.
Acknowledgments
This work was supported by the Academy of Finland (decision Nos 297893 and 318879).
References
[1] “Blue Brain Portal – Digitally reconstructing and simulating the brain.” [Online]. Available: https://portal.bluebrain.epfl.ch/. [Accessed: 12-May-2020]
[2] Amsalem, O., Eyal, G., Rogozinski, N. et al. An efficient analytical reduction of detailed nonlinear neuron models. Nat Commun 11, 288 (2020).
[3] S. Tennøe, G. Halnes, and G. T. Einevoll, “Uncertainpy: A Python Toolbox for Uncertainty Quantification and Sensitivity Analysis in Computational Neuroscience,” Front. Neuroinform., vol. 12, p. 49, Aug. 2018.
