Multitask Learning of a Biophysically-detailed Neuron Model
Jonas Verhellen, Kosio Beshkov, Sebastian Amundsen, Torbjørn V. Ness, Gaute T. Einevoll
Abstract:
The human brain operates at multiple levels, from molecules to circuits, and understanding these complex processes requires integrated research efforts. Simulating biophysically-detailed neuron models is a computationally expensive but effective method for studying local neural circuits. Recent innovations have shown that artificial neural networks (ANNs) can accurately predict the behavior of these detailed models in terms of spikes, electrical potentials, and optical readouts. While these methods have the potential to accelerate large network simulations by several orders of magnitude compared to conventional differential equation based modelling, they currently only predict voltage outputs for the soma or a select few neuron compartments.
Introduction:
In the seven decades since Hodgkin and Huxley first described the action potential in terms of ion channel gating, while the scientific community is gaining a comprehensive understanding of how individual neurons process information, the behavior of large networks of neurons remains comparatively poorly understood. Experimental studies provide qualitative insights through statistical correlations between recorded neural activity and sensory stimulation or animal behavior, but statistical modelling offers little information on how networks perform neural computation or give rise to neural representations.
Materials and Methods:
The subthreshold dynamics of the membrane potential in a compartment have small variations and as a result one would expect them to be easy to predict. However, suprathreshold deviations generated by action potentials propagating through the dendrites can be more problematic to predict. To address this issue, we implemented a form of data balancing. We first identified the time points at which somatic spikes occurred and afterwards we used them to create a dataset in which one third of the targets included a spiking event. Additionally, we standardised the membrane potential through z-scoring.
Discussion:
In our investigation, we evaluated three advanced multi-task learning (MTL) neural network architectures, including MMoE and MMoEx which were augmented with a novel compressor module, to tackle the challenging task of capturing intricate membrane potential dynamics in a complex, multi-compartment, biophysically-detailed model of a layer 5 pyramidal neuron. Given the considerable complexity of the model comprising 639 compartments, each generating distinct timeseries data potentially containing spikes, this MTL problem presented an exceptionally difficult distillation task.
Acknowledgments:
It has been made with Biorender.com through an academic license.
Citation: Verhellen J, Beshkov K, Amundsen S, Ness TV, Einevoll GT (2024) Multitask learning of a biophysically-detailed neuron model. PLoS Comput Biol 20(7): e1011728. https://doi.org/10.1371/journal.pcbi.1011728
Editor: Hugues Berry, Inria, FRANCE
Received: December 4, 2023; Accepted: June 28, 2024; Published: July 31, 2024.
Copyright: © 2024 Verhellen et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The LFDeep: Multitask Learning of Biophysically-Detailed Neuron Models project code is hosted on GitHub. The repository includes the implementation of the paper’s three multitask learning architectures for the distillation of biophysically-detailed neuron models. To access the latest version of the code and contribute to the project, please visit the official GitHub repository at https://github.com/Jonas-Verhellen/LFDeep. All data along with checkpoints with the optimal parameters of the MMoE model are available at https://www.kaggle.com/datasets/kosiobeshkov/data-for-mmoe-prediction.
Funding: This research was funded by UiO:Life Science through the 4MENT convergence environment to JV, the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement N° 945371 to KB and the European Union Horizon 2020 Research and Innovation Programme under Grant Agreement No. 945539 Human Brain Project(HBP) SGA3 to JV. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
