Periodic Reporting for period 1 - HBP SGA1 (Human Brain Project Specific Grant Agreement 1)
Reporting period: 2016-04-01 to 2018-03-31
Understanding the world inside our head is at least as important as understanding the one outside it. Given the brain’s complexity, a huge effort is required to unlock the brain’s secrets, but it promises major scientific, social and economic benefits. One is improved diagnosis and treatement for brain-related diseases, which are a growing health burden in our ageing population. A second is neuroscience’s potential to contribute to approaches for future ICT, including extreme and neuromorphic computing. Ultimately, the HBP will contribute to a more efficient, more intuitive, biologically based approach to Artificial Intelligence.
The HBP advances understanding of the human brain at all levels, from genomics to higher-level brain functions; all practical benefits depend on this. The HBP is also building an ICT-based research infrastructure to facilitate research collaboration, via the sharing of software tools, data and models. The HBP’s scientists and engineers are well-placed to ensure that our infrastructure meets real research needs. A third aim is to accelerate medical research, by facilitating researcher’s secure access to broader data sets of patient data, as well as HBP tools and models. The HBP also educates young scientists to work across disciplinary boundaries and addresses the ethical implications of its work. Finally, it helps to integrate global brain research efforts and leads Europe’s contribution.
Comparative studies determined which animal data can be used for understanding human brain organization. A mouse model of stroke provided insights valid in human brains. Deep learning helped us to compare mouse and human brain in visual feedback experiments. We examined connectivity between human brain regions, using multiple analytical techniques on a single subject for more robust results. Data and modelling interact: physiological mouse data fed modelling and simulation of the brain’s slow-wave activity, which in turn suggested future research. Our data is now accessible via our Brain Atlases.
HBP cognitive neuroscientists (SP3) enabled a robotics implementation of their spatial memory model and contributed to Brain Simulation Platform models. Our theoretical neuroscientists (SP4) developed models to bridge spatial scales and allow large-scale simulation of the visual cortex. Their work also helped to understand and replicate plasticity, the mechanism whereby the brain reconfigures itself to store memories. Diseased brain features were also modelled using data from the Medical Informatics Platform. Interdisciplinary collaboration has already had a clinical impact, helping to improve assessment of consciousness levels in coma patients.
Our six Platforms are now in routine operation with regular updates and a growing user community, which helps to co-design Platform refinements. Cross-platform integration is reflected in publications, and we laid the foundations for a large-scale infrastructure for data-sharing and management.
The KnowledgeGraph (SP5) established a community standard for managing metadata, which makes HBP data searchable. The Brain Simulation Platform (SP6) has released novel modular, use case-oriented tools for integrating simulations with the KnowledgeGraph. The High-Performance Computing and Analytics Platform (SP7) deployed its services to the first FENIX site, exploiting its virtual machines and object storage: a crucial infrastructure integration milestone. After in-depth review, the Medical Informatics Platform (SP8) is implementing local and federated installations for interrogating medical data according to a new deployment plan. The Neuromorphic Computing Platform (SP9) has two current-generation brain-inspired systems, routinely accessed by a range of users for tasks such as faster-than-real-time brain simulation, for exploring how the brain re-organises itself. We developed next-generation neuromorphic chips, with biologically inspired learning capabilities driven by new neuroscience insights. The Neurorobotics Platform demonstrated an in-silico reconstruction of a stroke recovery experiment, confirming that a large-scale, data-driven scaffold model of the whole-mouse brain can be used in a realistic closed-loop robotics experiment.
Integration of ethics has improved, helped by a revived Ethics Rapporteur network and the Ethics Advisory Board. An Ombudsperson and a Data Protection Officer were appointed.
A broad range of data types and experimental setups were used to populate atlases as precursors for future components. IEEG data from epilepsy patients served as a start point for integrating dynamic physiological data, preparing the ground for medical informatics work in SGA2, for increased impact on brain diseases. Acquired data sets represent high-quality, often unique inputs for modelling, feeding biologically realistic information into simulations of the neocortex, hippocampus and basal ganglia brain regions, whole brain models and highly detailed molecular-level simulations. This basic research facilitates study of drug-receptor interactions, a prerequisite for improved understanding of pharmacodynamics of drug candidates, and helps to identify new drug targets. Finally, neuroscience insights are increasingly informing new technologies such as brain-inspired computing, deep learning and robotics.
The HBP’s state-of-the-art ICT to advance modern neuroscience research is unrivalled. The HBP infrastructure complements growing global efforts to advance our understanding of the brain, cure brain diseases and develop brain-inspired technologies, and is a very competitive product. It builds on interdisciplinary collaboration, the use of modern ICT methods and their refinement. The interaction between neuroscience and modern computing is crucial. Only modern ICT infrastructures like the HBP’s can enable the efficient use of neuroscience data for model building and hence a multilevel understanding of the brain. Equally, only brain science can provide the urgently needed inputs to go from traditional deep network approaches to truly biologically inspired artificial intelligence.