Markus Diesmann

Overcoming the complexity barrier of brain modeling by digitization and collaboration

(1) Institute of Neuroscience and Medicine (INM-6) and Institute for Advanced Simulation (IAS-6) and JARA Institute Brain Structure-Function Relationships (INM-10), Jülich Research Centre. (2) Department of Psychiatry, Psychotherapy and Psychosomatics, School of Medicine, RWTH Aachen University. (3) Department of Physics, Faculty 1, RWTH Aachen University

Future brain models will not be created by individuals but by increasingly larger teams of researchers because of the complexity of the undertaking. At the outset of the Human Brain Project (HBP) the reproducibility of published network models and data analysis in computational neuroscience was limited. Thus, the community was ill-prepared for the new era, both technologically and sociologically. In this contribution we demonstrate, on the example of a multi-scale network model of one hemisphere of macaque vision-related cortex, the progress we made in the HBP in technology and in transforming the way computational neuroscience is done.

In the initial phase of the HBP, as a first step, we published as open-source code the formal executable description of a model of a cubic millimeter of cortical tissue [1] containing all its roughly 100,000 neurons and 300 million synapses between them. At this scale neurons reach their natural number of 10,000 synapses. The model is based on the integration of knowledge of some 50 experimental studies and reproduces a number of prominent features of neuronal activity, like the distribution of spike rates across cortical layers and the comparatively high activity of inhibitory neurons. For network theory this model of natural size is relevant because properties derivable only in the infinite size limit can be validated.

The microcircuit model found the resonance in the community we hoped for. The neuroscientific results were quickly reproduced by independent groups, and the model is now in use as a building block for more advanced models and as a testbed for studies on brain function and the validation of theoretical work.

As a second step we started to investigate a model of multiple cortical areas. The cortical architecture, i.e. the area-specific cellular and laminar composition of cortex, is related to the connectivity between areas, which forms a hierarchical and recurrent network at the brain scale. Using the vision-related cortex of the macaque as an example, our model therefore integrates data on the microscopic cortical architecture with macroscopic axonal tracing data into a multi-scale framework.

Reaching out to the brain scale required a new approach to model development, not because of the increased model size in terms of neurons and synapses, but because of the amount of data that needs to be integrated due to the heterogeneity of the brain. Therefore, we adopted methods of software development like version control, collaborative development, and code review to our needs using the GitHub platform.

The total network contains about 1 billion neurons, the limit of what supercomputers can simulate today [3] but too costly for routine laboratory work. Therefore, the initial version of the model [4] represents each of the 32 areas by just one microcircuit, reducing network size to a few million neurons. While the isolated microcircuit model has deficits with respect to the longer time scales of neuronal activity, the multi-area model is compatible with in-vivo resting-state data. Furthermore, the matrix of correlations between the activities of areas is more similar to the experimentally measured functional connectivity of resting-state fMRI than the anatomical matrix. This relates, in one model, the single-neuron level to the level of common brain imaging data.

When working on making the model accessible for our colleagues we noticed that the executable model description is not sufficient. The experimental data entering the model span multiple scales and come from different sources. Algorithms are required to collocate the data and derive the final model parameters. In several instances data are only partially available, such that predictive connectomics is needed to formulate quantitative hypotheses bridging the gaps. As a consequence, researchers can only add new data to the model or modify assumptions if they have access to the construction process. Therefore, the workflow of data integration also needs to be documented in an executable format. Borrowing techniques from computer science and systems biology [5] we demonstrate the development of a digitized workflow of model construction reproducing all figures of our respective publications.

Further increasing model size, for example to faithfully represent the vastly different relative extents of cortical areas, requires progress in simulation technology. The HBP therefore is a driver of the development of the next generation of supercomputers, so-called exascale systems, and researches corresponding simulation technology. The most recent code [6] makes the memory consumption of the individual compute nodes of a supercomputer fully independent of total network size and reduces simulation time on the current petascale systems from half an hour to five minutes. Still, solving the equations for microscopically parallel systems on conventional computers with their rather coarse-grained parallelism consumes considerable energy and reaching real time or accelerated speeds is difficult. Therefore, we also explore neuromorphic computing as a component of future modular supercomputers. A recent breakthrough is the simulation of the microcircuit model on the SpiNNaker hardware system [7]. All larger cortical models are less densely connected and in this sense easier to simulate.

[1] Potjans TC, Diesmann M (2014) Cerebral Cortex 24:785–806.

[2] Schmidt M, Bakker R, Hilgetag CC, Diesmann M, van Albada SJ. Brain (2018) Brain Struct Func 223(3):1409-1435

[3] Kunkel S, Schmidt M, Eppler JM, Plesser HE, Masumoto G, Igarashi J, Ishii S, Fukai T, Morrison A, Diesmann M, Helias M (2014) Front Neuroinform 8:78

[4] Schmidt M, Bakker R, Shen K, Bezgin G, Diesmann M, van Albada SJ (2018) PLOS Comput Biol 14(9): e1006359, in press

[5] Köster J, Rahmann S (2012) Bioinform 28:2520-2522

[6] Jordan J, Ippen T, Helias M, Kitayama I, Mitsuhisa S, Igarashi J, Diesmann M, Kunkel S (2018) Front Neuroinform 12:2

[7] van Albada SJ, Rowley AG, Senk J, Hopkins M, Schmidt M, Stokes AB, Lester DR, Diesmann M, Furber SB (2018) Front Neurosci 12:291

Short CV

Prof. Dr. Markus Diesmann is director of the Institute of Neuroscience and Medicine (INM-6, Computational and Systems Neuroscience), director of the Institute for Advanced Simulation (IAS-6, Theoretical Neuroscience) and director of the JARA-Institute Brain structure-function relationships (INM-10) at Jülich Research Centre, Germany. He is also full professor in Computational Neuroscience at the School of Medicine, RWTH University Aachen, Germany and affiliated with the Department of Physics of the same university. Prof. Diesmann studied physics at Ruhr University Bochum with a year of Cognitive Science at University of Sussex, UK. He carried out his PhD studies at Weizmann Institute of Science, Rehovot, Israel, and Albert-Ludwigs-University Freiburg. In 2002 he received his PhD degree from the Faculty of Physics, Ruhr-University Bochum, Germany. From 1999 Prof. Markus Diesmann worked as senior staff at Department of Nonlinear Dynamics, Max-Planck-Institute for Dynamics and Self-Organization, Göttingen, Germany. In 2003 he became assistant professor of Computational Neurophysics at Albert-Ludwigs-University, Freiburg, Germany before in 2006 joining the RIKEN Brain Science Institute, Wako City, Japan as a unit leader and later team leader. In 2011 Markus Diesmann moved to Jülich. His main scientific interests include the correlation structure of neuronal networks, models of cortical networks, simulation technology and supercomputing. He is one of the original authors of the NEST simulation code and a member of the steering committee of the NEST Initiative.