Dartmouth Undergraduate Journal of Science Volume 19 Article 2 Number 3 Influence: Science in Service to Society 2017 Towards Simulating the Human Brain Logan T. Collins [email protected] Follow this and additional works at: https://digitalcommons.dartmouth.edu/dujs Part of the Engineering Commons, Life Sciences Commons, Medicine and Health Sciences Commons, and the Social and Behavioral Sciences Commons Recommended Citation Collins, Logan T. (2017) "Towards Simulating the Human Brain," Dartmouth Undergraduate Journal of Science: Vol. 19 : No. 3 , Article 2. Available at: https://digitalcommons.dartmouth.edu/dujs/vol19/iss3/2 This Review Article is brought to you for free and open access by the Student-led Journals and Magazines at Dartmouth Digital Commons. It has been accepted for inclusion in Dartmouth Undergraduate Journal of Science by an authorized editor of Dartmouth Digital Commons. For more information, please contact [email protected]. COMPUTER SCIENCE Towards Simulating the Human Brain BY LOGAN COLLINS ’19 The Potential of Brain Emulation of implementing this new understanding of Figure 1: This figure depicts a cognition in artificial intelligence and even Blue Gene/P supercomputer. The human brain has been described as developing intelligent machines. Brain- “the most complex object in the universe.” Its computer interfaces (BCIs) may also benefit Source: Wikimedia Commons meshwork of 86 billion neurons, 84 billion glial from WBE since more precise neuronal codes for (Credit: Argonne National cells, and over 150 trillion synapses may seem coding motor actions and sensory information Laboratory) intractable (Azevedo et al., 2009; Pakkenberg, could be uncovered. By studying WBEs, the 2003). Nonetheless, efforts to comprehensively “language” of the brain’s operations could be map, understand, and even computationally revealed and may give rise to a rich array of new reproduce this structure are underway. Large advances. collectives of researchers have come together, “Whole brain working in concert towards these goals. The emulation (WBE), Human Brain Project (HBP) and its precursor, Foundations of Computational the Blue Brain Project, have spearheaded the Neuroscience the computational brain simulation goal (Grillner et al., 2016). Biologically accurate neuronal simulations simulation of the Some other notable organizations include the usually employ conductance-based models. human brain with China Brain Project, the BRAIN Initiative. On a The canonical conductance-based model was synaptic (or higher) scale which parallels the space program and the developed by Alan Hodgkin and Andrew resolution, would Human Genome Project, neuroscience may be Huxley and published in 1952, later winning approaching a revolution. them the Nobel Prize in Physiology or Medicine fundamentally Whole brain emulation (WBE), the (Hodgkin and Huxley, 1952). Current models change medicine, computational simulation of the human brain still use the core principles from the Hodgkin- artificial with synaptic (or higher) resolution, would Huxley model. intelligence, and fundamentally change medicine, artificial This model is a differential equation neurotechnology.” intelligence, and neurotechnology. Modeling which takes into account the conductances the brain with this level of detail could reveal and equilibrium potentials of neuronal sodium new insights about the pathogenesis of mental channels, potassium channels, and channels that illness (Markram et al., 2011). It would provide transport anions. The output of this equation is a virtual environment in which to conduct the total current across the membrane of the experiments, though researchers would need neuron, which can be converted to a voltage by to develop guidelines regarding the ethics of multiplying by the total membrane resistance. these experiments, since such a construct may possess a form of consciousness. This virtual connectome could also vastly accelerate studies of human intelligence, leading to the possibility SPRING 2017 5 The Hodgkin-Huxley equation (above) can constructing virtual neurons and assemblies generate biologically accurate predictions for the of neurons. They have shown remarkable membrane voltage of a neuron at a given time. biological fidelity even in complex simulations. Using these membrane voltages and the firing Multi-compartmental Hodgkin-Huxley models, threshold of the neuron in question, the timing given the proper parameters from experimental of action potentials can be computationally data, can make predictive approximations of predicted. biological activity. Another important concept in “Multi- computational neuroscience is the multi- The Blue Brain Project compartmental compartmental model. Consider an axon which The quest to simulate the human brain has synapses onto another neuron’s dendrite. When largely emerged from The Blue Brain Project, models use the axon terminal releases neurotransmitters complicated partial a collaboration headed by Henry Markram which depolarize the other neuron’s dendritic (Figures 1 and 2). In 2007, Markram announced differential equation membrane, the depolarization will need to travel the completion of the Blue Brain Project’s first extensions of the down the dendrite, past the soma, and onto the phase, the detailed simulation of a rat neocortical Hodgkin-Huxley axon of this post-synaptic neuron to contribute column in IBM’s Blue Gene supercomputer. to the initiation of an action potential. Since equation. When This achievement required a powerful the density of voltage-gated channels outside engineering strategy that integrated the many solved, the resulting of the axon is relatively low, the depolarization components of the simulation (Markram, 2006). multivariable decreases as it moves along this path. In order To start, gene expression data were used to functions take both to accurately recreate a biological neuron in a determine the ion channel distributions in the computer, this process must also be modeled. membranes of various types of cortical neuron. the location along To accomplish this, the model segments each the dendrite and the Over twenty different types of ion channel portion of the neuronal membrane into multiple were considered in this analysis. These ion time as inputs.” “compartments.” In general, the higher the channels were incorporated into extensions number of compartments, the more accurate of the Hodgkin-Huxley model with additional the model, but also the more computational terms for the many types of channel. Three- resources the model will require. Multi- dimensional neuron morphologies were paired compartmental models use complicated partial with appropriate ion channel distributions differential equation extensions of the Hodgkin- and a database of virtual neuron subtypes was Huxley equation. When solved, the resulting assembled. Experimental data on axon and multivariable functions take both the location dendrite locations was collected to recreate the along the dendrite and the time as inputs. synaptic connections in the neocortical column. These methods form the fundamentals for A collision detection algorithm was employed Figure 2: Henry Markram explains neuronal function Source: Wikimedia Commons (Credit: Steve Jurvetson) 6 DARTMOUTH UNDERGRADUATE JOURNAL OF SCIENCE Figure 3: Cajal Blue Brain research is ongoing in Spain Source: Wikimedia Commons (Credit: Cesvima) to adjust the three-dimensional arrangement of the HBP was selected as a European Union axons and dendrites by “jittering” them until they Flagship project and granted over 1 billion euros matched the experimental data. Physiological in funding (equivalent to slightly over 1 billion recordings provided membrane conductances, USD). probabilities of synaptic release, and other Criticisms inevitably arose. Some scientists biophysical parameters necessary to model each feared that it would divert funding from neuronal subtype. In addition, plasticity rules other areas of research (Frégnac and Laurent, mirroring those found in biological neurons 2014). A major complaint was that the HBP were applied to the virtual neurons to allow was ignoring experimental neuroscience in them to perform learning. Through a series of favor of simulations. The simulation approach iterative tests and corrections, parameters were would need to be complemented by further optimized. data collection efforts since connectomic and By 2012, the project reached another functional mapping information is lacking. As milestone, the simulation of a cortical a result, the HBP restructured to broaden its mesocircuit consisting of over 31,000 neurons focus. “Criticisms [of in several hundred “minicolumns” (Markram et An array of neuroscience platforms The Human Brain al., 2015). In this simulation, some of the details with varying levels of experimental and of synaptic connectivity were algorithmically computational focus were developed (Amunts Project] inevitably predicted based on experimental data rather et al., 2016). The Mouse Brain Organization arose. Some than strictly adhering to experimentally and Human Brain Organization Platforms were scientists feared generated maps. Nevertheless, valuable insights initiated to further knowledge of brain structure that it would divert were produced from the emulated mesocortical and function through more experimentally circuit. In vivo systems have shown puzzling centered approaches. The Systems and funding from other bursts of uncorrelated activity. The simulation