openaire-coar conference 2014: global collaborative neuroscience - building the brain from big data,...

The Human Brain Project: Building the brain from big data

!Sean Hill

Co-Director, Blue Brain Project Co-Director, Neuroinformatics, Human Brain Project

EPFL, Lausanne, Switzerland !!

Co-Executive Directors: Henry Markram, Brain Mind Institute, EPFL, Switzerland

Karlheinz Meier, Kirchoff Institute, University of Heidelberg, Germany Richard Frackowiak, Department of Clinical Neurosciences, CHUV,

Switzerland !

www.humanbrainproject.eu

Upon this gifted age, in its dark hour, Rains from the sky a meteoric shower Of facts . . . they lie unquestioned, uncombined. Wisdom enough to leech us of our ill Is daily spun; but there exists no loom To weave it into fabric; Edna St. Vincent Millay, 1939

Developmental Disorders• Autism spectrum disorders • ADHD • Learning disorders, conduct

disorders • Strong genetic disorders

(Fragile X, Down’s etc)

Adolescent Disorders• Depression, Suicide • Eating disorders • Bipolar disorder • Conduct disorders and violence • Borderline syndrome • Adjustment disorders • Anxiety, phobias, suicide • Tourette’s syndrome • Epilepsy

• Schizophrenia • Epilepsy • Mood disorders, hysterias,

anxieties and phobias • Obsessive compulsive disorders • Eating disorders, sexual disorders • Sleep disorders, stress disorders • Impulse control disorders • Substance abuse disorders • PTSD/TBI

Adult Disorders

• Depression • Dementia • Neurodegenerative disorders

• Alzheimer’s • Parkinson’s • Huntington’s

• Memory disorders

Aging Disorders

Glutamate !Nutrition !Dopamine !Genes !Sugar !GABA !Myelin !Serotonin !Metals !Dopamine !Toxins !Acetylcholine !Protein misfolding

Microarrays Electron Microscopy

Confocal Microscopy

Single Cell PCR Protein quan8fica8on Magne8c bead

Gene sequencing

Gene silencing Gene over-‐expression Gene8c vectors Two-‐hybrid

system Protein separa8on

Wholecell &Inside-‐Out Patch

Laser micro-‐dissec8on Cell culture Fluorescence

microscopyCellular tracing

Cell sor8ng

In situ hybridiza8on

Rhodopsin vectors

Immuno-‐detec8on amplified by T7 Mass-‐spectroscopy

Organelle transfec8on

Spa8al Proteomics

Immuno-‐staining

Mul8 Electrode Array Extracellular Recording Dye Imaging 2DE proteomics Tissue

transfec8onEnzyma8c-‐ac8vity measurement

Behavioral Studies UltramicroscopyMagnet Resonance Diffusion Imaging fMRI EEG Transgenic lines

What is the Human Brain Project?

7

A 10-year European initiative to launch a global, collaborative effort to understand the human brain, enabling advances in neuroscience, medicine and future computing. !One of the two final projects selected for funding as a FET Flagship from 2013. !Officially launched October 1, 2013. !A consortium of 256 researchers from 135 research groups, 81 partner institutions in 22 countries across Europe, America and Asia. !€ 74M funding for the first 30 months. €1 billion/10 years.

What is a FET Flagship?

8

Future and Emerging Technologies (FET) Flagships are ambitious large-scale, science-driven, research initiativesthat aim to achieve a visionary goal. !The scientific advance should provide a strong and broad basis for future technological innovation and economic exploitation in a variety of areas, as well as novel benefits for society. !Objective is to keep Europe competitive and drive technological innovation

Goal

9

The Human Brain Project should: !•Lay the technical foundations for a new model of ICT-based brain research !

•Drive integration between data and knowledge from different disciplines !

•Catalyze a community effort to achieve a new understanding of the brain, new treatments for brain disease and new brain-like computing technologies.

Research Areas

10

Neuroscience !

Integrate everything we know about the brain into computer models and simulations !!Medicine !

Contribute to understanding, diagnosing and treating diseases of the brain !!Future Computing !Learn from the brain to build the supercomputers of tomorrow

Scientific organization

11

Organisation of HBP Scientific

& Technological Work

Divisions

Mol

ecul

ar a

nd

Cel

lula

r N

euro

scie

nce

Cog

nitiv

e N

euro

scie

nce

Theo

retic

al

Neu

rosc

ienc

e

Neu

roin

form

atic

s

Bra

in S

imul

atio

n

Hig

h P

erfo

rman

ce

Com

putin

g

Med

ical

Info

rmat

ics

Neu

rom

orph

ic

Com

putin

g

Neu

roro

botic

s

Mouse Data

Human Data

Cognition Data

Theory

NIP

BSP

HPCP

MIP

NMCP

NRP

Applications for neuroscience

Applications for medicine

Applications forfuture computing

Sci

entifi

c an

d te

chno

logi

cal r

esea

rch Data

Platform building

Platform use

Data

12

Generate and interpret strategically selected data needed to build multilevel atlases and unifying models of the brain.

54 Scientific American, June 2012

stellation of symptoms, matches what we see in real life, that virtual chain of events becomes a candidate for a disease mecha-nism, and we can even begin to look for potential therapeutic targets along it.

This process is intensely iterative. We integrate all the data we can find and pro-gram the model to obey certain biological rules, then run a simulation and compare the “output,” or resulting behavior of pro-teins, cells and circuits, with relevant ex-perimental data. If they do not match, we go back and check the accuracy of the data and refine the biological rules. If they do match, we bring in more data, adding ever more detail while expanding our model to a larger portion of the brain. As the soft-ware improves, data integration becomes faster and automatic, and the model be-haves more like the actual biology. Model-ing the whole brain, when our knowledge of cells and synapses is still incomplete, no longer seems an impossible dream.

To feed this enterprise, we need data and lots of them. Ethical concerns restrict the experiments that neuroscientists can perform on the human brain, but fortu-nately the brains of all mammals are built according to common rules, with species-specific variations. Most of what we know about the genetics of the mammalian brain comes from mice, while monkeys have giv-en us valuable insights into cognition. We can therefore begin by building a unifying model of a rodent brain and then using it as a starting template from which to de-velop our human brain model—gradually integrating detail after detail. Thus, the models of mouse, rat and human brains will develop in parallel.

The data that neuroscientists generate will help us identify the rules that govern brain organization and verify experimen-tally that our extrapolations—those pre-dicted chains of causation—match the bi-ological truth. At the level of cognition, we know that very young babies have some grasp of the numerical concepts 1, 2 and 3 but not of higher numbers. When we are finally able to model the brain of a new-born, that model must recapitulate both what the baby can do and what it cannot.

A great deal of the data we need al-ready exist, but they are not easily accessi-ble. One major challenge for the HBP will be to pool and organize them. Take the medical arena: those data are going to be immensely valuable to us not only be-cause dysfunction tells us about normal

Illustration by Emily Cooper

Deconstructing the Brain!"#$%&'()$*+(,)$-+./#01�� ##��!�� "�� !��"��!��$�� !��

L AY E R B Y L AY E R

Molecular!"#$%&'()"*+"($,$-(#./"0$12%%2%1""�� '%3$("-"42#(*,#*5$/"6*'73"&(-%,7-&$"2%&*"-"3212&-7"+-#,2427$"&.-&"#*402%$,"#*45*%$%&"4*7$#'7-("5-(&,"&*"-,,$4"07$"-"#$77"&.-&"3$4*%,&(-&$,"&.$"$,,$%&2-7"5(*5$(&2$,"*+"-"%$'(*%/"&.$"&(-%,42,,2*%"*+"$7$#&(2#-7"-%3"#.$42#-7",21%-7,8"

Cellular!"0(-2%92%9-90*:",24'7-&2*%"6277".-;$"&*"#-5&'($"$;$()"3$&-27"*+""%$'(*%,"-%3"%*%%$'(*%-7"172-7""#$77,/"2%#7'32%1"&.$"$:-#&"1$*4$&(2#",.-5$,"*+"&.$2("3$%3(2&$,"-%3"-:*%,"&.-&"($#$2;$"-%3",$%3"2%+*(4-&2*%8""

Circuits!"4*3$7"*+"&.$"%$'(-7"#*%%$#&2*%,"��-4*%1"%$21.0*(2%1"#$77,"4-)"+'(%2,."#7'$,"&*"&.$"*(212%,"*+"#*457$:"0(-2%"32,$-,$,",'#."-,"-'&2,4"-%3",#.2<*5.($%2-8"

Whole Organ!%"2%",272#*"0(-2%"421.&",'0,&2&'&$"+*("&.$"-#&'-7"*(1-%8"=)"($4*;2%1"&.$"#*45'&$("#*3$"+*("-">1$%$/?""&.$";2(&'-7",),&$4"#-%/"+*("2%,&-%#$/"��"-,",#2$%&2,&,"3*"&*3-)"0)">@%*#@2%1"*'&?"-"1$%$"2%"42#$8"A.$"&**7"6*'73"-;*23"&.$"7$%1&.)"0($$32%1"5(*#$,,"-%3"#*'73",24'7-&$"-"4'7&2&'3$""*+"$:5$(24$%&-7"#*%32&2*%,8""

RegionsB-C*("%$'(-7",'0,&('#&'($,D"&.$"-4)13-7-"E$4*&2*%,F/"&.$".255*#-45',"E4$4*()F/"&.$""+(*%&-7"7*0$,"E$:$#'&2;$"#*%&(*7FD#-%"0$"2%,5$#&$3"-7*%$"*("-,"&.$)"2%&$(-#&"62&."*%$"-%*&.$(8""

A Data Ladder to the Human Brain

13

However, !

HBP is NOT primarily a data generation project !

It IS a data integration project.

What I cannot create, I do not understand.!-Richard Feynman

Build, Simulate and Validate Unifying Brain Models

16

208.17

0 100 0 100 0 100 0 100 0 100 200

L6L5L4L2/3L1/2

L6

L5

L4

L2L1

L3

Stimulation Site

Firing Cell Count

Experimental Data Gathering

Supercomputing

10 ms

50 mV

Vm

Istim

4 nA

Simulation

Model Building

Model Validation

Refinement of

Models and Experiments

Worldwide Published Data,

Models and Literature

Analysis and

Visualization

Mouse atlas data courtesy of Allen Ins8tute for Brain Science

Theory

17

FIGURE 13

The Integrative Role of Theory in the HBP

Theories of computing

Theories of information processing

Theories of cognition

Principles of Neural Computation

Bridging layers of biology

Statistical validation

Mathematical formulations

Numerical methods

Statistical predictive models

Data Analysis

Conceptual Predictions

Molecular, Biophysical & Phenomenological Models

Neuroinformatics

Neuroscience

Brain Tissue Learning & memory

Cognition& Behavior

Neurons Synapses Circuits

Multiscale models

Simplified models

Neuromorphic implemetation

Robotics Implementation

ICT Platforms

18

An integrated network of ICT platforms of the HBP

Figure 48: The HBP Platforms – physical infrastructure. Neuromorphic Computing: Heidelberg, Germany & Manchester, United Kingdom; Neurorobotics: Munich, Germany; High Performance Computing: Julich, Germany & Lugano, Switzerland, Barcelona, Spain & Bologna, Italy; Brain Simulation: Lausanne, Switzerland; Theory Institute: Paris, France

Six new open access ICT platforms: !1. Neuroinformatics

2. Brain Simulation

3. Medical Informatics

4. High Performance Computing

5. Neuromorphic Computing 6. Neurorobotics

!For the entire research community.

Neuroinformatics Platform

19

Provide technical capabilities to federate neuroscience data, analyze structural and functional brain data and to build and navigate multi-level brain atlases. This involves: !•spatial and temporal data registration •ontology development and semantic annotation •predictive neuroscience •machine learning, data mining •track provenance, build workflows. !Goal: enable an integrated view of the neuroscience data. Prepare data for modeling pipelines

Multiscale and Multimodal Brain Atlases

20

Data registered to standard coordinate spaces:- collections of spatially and semantically registered and searchable data, models and literature

Mouse atlas data courtesy of Allen Ins8tute for Brain Science

• Share data: Secure access, authorization, collaborative • Curate data: Data annotation, quality control, consistency check • Publish data: Data citation, community rating and reviews • Preserve data: Versioning, archive, mirror, replicate • Federate data: Tens to hundreds of thousands of members of the

federation • Federated search: Search over all data, free text, ontology-based or spatial

search • Ease of use: Lightweight! Dropbox-like functionality, easy metadata

annotation, portal for search • Access data: Object-based view of data, semantic annotations, data

models, flexible metadata • Analyze and integrate data: Support data analysis, workflow building,

provenance tracking

21

Requirements: Global Neuroinformatics Infrastructure

22

Global shared data (INCF DataSpace)

Data access, query and provenance services

Portals, Search, Applications

Analytic Services

Neuroinformatics Infrastructure

Digital Atlasing services

Private LabSpace(Curated data)

Global Public KnowledgeSpace (Published data)

Data management strategy

23

DATASPACE• Don’t centralize - federate

• INCF Global datasharing infrastructure

• Federated data management

• Dropbox-like Ease of Use

• Highly scalable - index petabytes of data

• Robust Infrastructure

dataspace.incf.org

24

MIRROR METADATA ACROSS FOUR AMAZON ZONES

US-‐WEST US-‐EAST EU-‐WEST AP-‐EAST

Cloud Server8 GB memory; 4 cores850 GB instance storage64-bit platformI/O Performance: 500 Mbps

Dataspace metadata

Cloud Server A

Dataspacemetadata

Cloud Server B

Dataspacemetadata

Cloud Server C

Dataspacemetadata

Cloud Server D

Knowledge integration strategy

25neurolex.org

•INCF Community encyclopedia •Living review articles •Curated data •Build and maintain working ontologies •Links to data, models and literature •Define all vocabulary, terms, protocols, brain structures, diseases, etc •Semantic organization, search, analysis and integration •Global directory of all shared vocabularies, CDEs, etc

Semantic search of data

Search for data through semantic relationships

•Producer •Laboratory •Experimenter

•Protocol •Solutions •Equipment •Measurements

•Specimen •Species •Strain •Age

•Location •Brain region name •X,Y,Z location

•Data sample •Format

•Community rating •Extracted features •Analysis

Semantically annotated digital data publications


Brain volume descriptions


Methods


Data with extracted features


Overview statistics and classifications


Cell densities


Gene expression


Neuron morphologies


Electrophysiology


Synaptic connectivity

Automated Analysis

38

Morphology Analysis and Classification

39

Brain Simulation Platform

40

Provide technical capabilities to build and simulate multi-scale brain models at different levels of detail.

•internet portal for neuroscientists •modeling tools •workflows •simulation •virtual instruments (EM, LFP, fMRI, etc) •link to virtual body and environment •in silico experiments

Goal: Integrate large volumes of heterogeneous data in multi-scale models of the mouse and human brains, and to simulate their dynamics.

Enable neuroscientists to ask new questions and prioritize experiments.

Integration of laboratory data

41

S1L1 Neocortical Microcircuit

30’000 neurons

400 Electrical BehaviorsGene profiles

Synap.c profiles

Morphology profiles

Connec.vity profiles

Electrical profiles

Circuit profiles

Protein profiles


42

Data-driven biophysical single cell models

Hay et al., PLOS Comp. Biology, 2012

Building a cortical microcircuit

43

44

Predictive Neuroscience: One example

0&ï3&��Q ��

3&ï1%&��Q ��

%7&ï3&��Q ��

6%&ï3&��Q ��

1%&ï3&��Q ��

3&ï/%&��Q ��

3&�0&��Q ��

3&�%7&��Q ��

�

��

��

��

��

$[RQ

+,� ��KS*

�

��

��

��

��

Soma

%DVDO

ASLFDO

7XIt

+,� ��KS*

�

��

��

��

��

$[RQ

+,� ��

�

��

��

��

��

Soma

%DVDO

ASLFDO

7XIW

+,� ��KS*

�

��

��

��

��

$[RQ

+,� ��KS*

�

��

��

��

��

Soma

%DVDO

ASLFDO

7XIt

+,� ��KS*

�

��

��

��

��

$[RQ

+,� ��KS*

�

��

��

��

��

Soma

%DVDO

ASLFDO

7XIW

+,� ��

�

��

��

��

��

$[RQ

+,� ��

�

��

��

��

��

Soma

%DVDO

ASLFDO

7XIt

+,� ��

�

��

��

��

��

$[RQ

+,� ��

�

��

��

��

��

Soma

%DVDO

ASLFDO

7XIt

+,� ��

�

��

��

��

��

$[RQ

+,� ��KS*

�

��

��

��

��

Soma

%DVDO

ASLFDO

7XIt

+,� ��

� ��

��

��

��

��

$[RQ

+,� ��

�

��

��

��

��

Soma

%DVDO

ASLFDO

7XIt

+,� ��

� ��

Hill et al. PNAS 2012

2,970 possible synaptic pathwaysin a cortical microcircuit alone.

!22 have been characterized.

!Can we identify principles

to predict the rest?


45

Cortical microcircuitry


46

Cortical microcircuitry and local field potential

Reimann et al., Neuron, 2013 in collaboration with the Allen Institute, Seattle, WA

230μm thickA virtual brain slice

Spontaneous activity in a virtual brain slice230μm thick

Virtual slice: evoked activity, 20Hz stimulation

Virtual slice: evoked activity, 40Hz stimulation

51

UNCO

RRECTED P

RO

OF

311 variance between different lines (C57BL/6 and Kunming) of mouse312 seems greater than the variance among different individuals in the313 same line (C57BL/6).314 To compare the laminar distribution in different cortical regions, we315 extracted the image stacks of the other two cortical regions in the left316 hemisphere of a C57 mouse brain, the posteromedial barrel subfield317 (PMBSF, AP: Bregma, −0.22 mm, ML: Bregma, +3.25 mm, DV:318 Bregma, +2 mm) and the primary visual cortex (V1, AP: Bregma,319 −3.88 mm, ML: Bregma, +2.5 mm, DV: Bregma, +1.25 mm). The320 inner cells and blood vessels were vectorized and reconstructed (Sup-321 plementary Figs. 5a–e). In the barrel cortex, the vascular diameters322 were estimated both in vivo and in vitro for evaluation (Supplementary323 Fig. 6e). The estimated diameter histogram showed that the diameter324 distribution peaked in 3–6 μm both in vivo and in vitro, which is also325 in good agreement with the histograms from Tsai et al. (2009). Note326 that a penetrating vessel had a sinuous path in the middle layer of the327 barrel cortex (Supplementary Fig. 5b), which is consistent with an ob-328 servation in the human brain (Duvernoy et al., 1981). The cellular and329 vascular distribution parameters were computed along the cortical330 axial direction (Fig. 5). As expected, the cell number density varied331 greatly among different cortical regions, especially near layer IV332 (Fig. 5a). The densities in the granular (PMBSF), and partially granular

333cortex (V1), were relatively higher than in the agranular cortex (M1).334The smallest distances from cells to the nearest blood vessels were335near the middle layers (Fig. 5b). For the vascular length densities336(Fig. 5c) andmicrovascular fractional volumes (Fig. 5d), the peak values337were also located near layer IV. Interestingly, we noticed that the vascu-338lar density values in layers I, V and VI were similar (Figs. 5c, d). For339quantitative evaluation, we computed the standard deviation of three340cortical regions in different layers and found that the variation in vascu-341lar density was significantly higher in the middle layers (20%–50% cor-342tical depth, near layer II/III and IV) than in the upper and lower layers343(p b 0.01).

344Areal configurations of cells and blood vessels

345Mainly limited by imaging depth, most of the 3D optical imaging346and quantitative analysis techniques was restricted in the part of the347cerebral cortex (about 1 mm3), leaving the deeper brain areas largely348uncharted. Our customized protocol enabled us to analyze the detailed349cellular and vascular configurations on a brain-wide scale. We350extracted three additional image stacks in the mouse brain: frontal351association cortex (FrA, AP: Bregma,+2.58 mm,ML: Bregma,+1.5 mm,352DV: Bregma,+1.5 mm), striatum (AP: Bregma,+0.02 mm,ML: Bregma,

Fig. 4. Reconstruction and colocalization of vectorized cells and blood vessels inM1. (a) 3D reconstruction of the cytoarchitecture inM1. The green points represent detected cellular cen-troids. (b) The surface rendering of thevasculature showshowsome large vessels penetrate into the cerebral cortex. The curved tubes represent the traced blood vessels, and the diametersof the blood vessels are encoded as indicated by the color bar. The white arrow indicates a large penetrating vessel, the diameter of which exceeds 60 μm. The pial vascular network wasmanual labeled and rendered to give an anatomical context. The size of the gray bounding box is 600 μm × 600 μm × 920 μm. (c) An enlarged view of the white box in (a, b) showscolocalization of the cells (green) and blood vessels (red). The size of the white bounding box is 200 μm × 200 μm × 100 μm.

6 J. Wu et al. / NeuroImage xxx (2013) xxx–xxx

Please cite this article as: Wu, J., et al., 3D BrainCV: Simultaneous visualization and analysis of cells and capillaries in a whole mouse brain withone-micron voxel resoluti..., NeuroImage (2013), http://dx.doi.org/10.1016/j.neuroimage.2013.10.036

Wu et al, Neuroimage, 2013

UNCO

RRECTED P

RO

OF

274 and previous definitions (Blinder et al., 2013; Heinzer et al., 2006; Tsai275 et al., 2009).

276 Results

277 High resolution 3D datasets of whole mouse brains

278 Four 3D datasets of the whole mouse brains were obtained at a sec-279 tioning thickness of 1 μm with a lateral pixel size of 0.4 μm × 0.35 μm280 (Figs. 1, 3a–c, Supplementary Video 2). Cells and blood vessels can be281 simultaneously distinguished by voxel intensity (Figs. 1f–j, 3f–h), and282 brain-wide background intensity is uniform after preprocessing283 (Figs. 1a–e, 3a–c). In addition, little data was lost while sectioning, and284 the morphology of the cells and blood vessels that spanned the section285 boundaries was well maintained (Figs. 1f–j, Supplementary Figs. 2c–f).286 Furthermore, the cytoarchitecture and vascular architecture could be vi-287 sualized using minimum (Fig. 2d) and maximum intensity projections288 respectively (Fig. 2g), and some brain structures, such as the olfactory289 bulb, hippocampus, and striatum can be clearly identified (Figs. 1a–e,290 3b–d).

291Laminar configurations of cells and blood vessels

292In order to obtain a normative anatomy of laminar configurations of293cells and blood vessels, we computed the cellular and vascular configu-294rations in M1 of three mouse brains (the two hemispheres in two C57295mice, left hemisphere in a Kunmingmouse brain). We cropped the cor-296onal images in the primary motor cortex (M1, AP: Bregma, +1.18 mm,297ML: Bregma, +1.75 mm, DV: Bregma, +1.75 mm) according to a298mouse brain atlas (Paxinos and Franklin, 2001). The image stacks299were virtually resliced parallel to the pial surface and were cropped300(600 μm × 600 μm) to cover Q5the entire cortical depth. The inner cells301and blood vessels were vectorized and reconstructed (Fig. 4). Due to302the curved nature of the pial surface and the cortical layers, the absolute303cortical range was hard to precisely localize in a large area of cortical304mantle. Therefore, 600 μm × 600 μm sections of the pial surface were305divided into 200 μm × 200 μm squares, and the cortical depth in each306square was normalized to allow direct comparison of different layers.307The laminar densities of the cells and blood vessels were successfully vi-308sualized, and the results show amismatch between cellular density and309vascular density (Supplementary Fig. 7), which is in agreement with310previous results (Tsai et al., 2009). The result suggested that the

Fig. 3. 3Ddataset of awholemouse brain used to simultaneously visualize the cells and blood vessels. (a) 3D reconstruction of thewholemouse brainwith a corner-cut view. (b, c) Sagittal andcoronal sections in (a) (thickness: 1 μm) show the inner structures of the whole mouse brain. Note that the background intensity is uniform. (d) Minimum intensity projection of (c) (thick-ness: 30 μm,posterior direction) shows the cytoarchitecture. (e)Maximum intensity projection of (c) (thickness: 200 μm,posterior direction) shows vascular architecture. (f) Enlarged viewofthe cortical region indicated by the white box in (c) shows simultaneous cross-sections of cells (black) and blood vessels (white). (g) The enlarged view of the cortical region indicated by thewhite box in (d) shows the laminar cytoarchitecture in the neocortex. Layers I to VI can be clearly distinguished. (h) Enlarged viewof the cortical region indicated by thewhite box in (e) showsthe vascular architecture, including capillaries, in the neocortex. The diameters of the large penetrating vessels are approximately 23 μmand thediameter of small capillaries are less than 3 μm.Note that the capillaries are interconnected with few gaps, suggesting complete labeling of the vascular network.

5J. Wu et al. / NeuroImage xxx (2013) xxx–xxx


UNCO

RRECTED P

RO

OF

274 and previous definitions (Blinder et al., 2013; Heinzer et al., 2006; Tsai275 et al., 2009).

276 Results

277 High resolution 3D datasets of whole mouse brains

278 Four 3D datasets of the whole mouse brains were obtained at a sec-279 tioning thickness of 1 μm with a lateral pixel size of 0.4 μm × 0.35 μm280 (Figs. 1, 3a–c, Supplementary Video 2). Cells and blood vessels can be281 simultaneously distinguished by voxel intensity (Figs. 1f–j, 3f–h), and282 brain-wide background intensity is uniform after preprocessing283 (Figs. 1a–e, 3a–c). In addition, little data was lost while sectioning, and284 the morphology of the cells and blood vessels that spanned the section285 boundaries was well maintained (Figs. 1f–j, Supplementary Figs. 2c–f).286 Furthermore, the cytoarchitecture and vascular architecture could be vi-287 sualized using minimum (Fig. 2d) and maximum intensity projections288 respectively (Fig. 2g), and some brain structures, such as the olfactory289 bulb, hippocampus, and striatum can be clearly identified (Figs. 1a–e,290 3b–d).

291Laminar configurations of cells and blood vessels

292In order to obtain a normative anatomy of laminar configurations of293cells and blood vessels, we computed the cellular and vascular configu-294rations in M1 of three mouse brains (the two hemispheres in two C57295mice, left hemisphere in a Kunmingmouse brain). We cropped the cor-296onal images in the primary motor cortex (M1, AP: Bregma, +1.18 mm,297ML: Bregma, +1.75 mm, DV: Bregma, +1.75 mm) according to a298mouse brain atlas (Paxinos and Franklin, 2001). The image stacks299were virtually resliced parallel to the pial surface and were cropped300(600 μm × 600 μm) to cover Q5the entire cortical depth. The inner cells301and blood vessels were vectorized and reconstructed (Fig. 4). Due to302the curved nature of the pial surface and the cortical layers, the absolute303cortical range was hard to precisely localize in a large area of cortical304mantle. Therefore, 600 μm × 600 μm sections of the pial surface were305divided into 200 μm × 200 μm squares, and the cortical depth in each306square was normalized to allow direct comparison of different layers.307The laminar densities of the cells and blood vessels were successfully vi-308sualized, and the results show amismatch between cellular density and309vascular density (Supplementary Fig. 7), which is in agreement with310previous results (Tsai et al., 2009). The result suggested that the

Fig. 3. 3Ddataset of awholemouse brain used to simultaneously visualize the cells and blood vessels. (a) 3D reconstruction of thewholemouse brainwith a corner-cut view. (b, c) Sagittal andcoronal sections in (a) (thickness: 1 μm) show the inner structures of the whole mouse brain. Note that the background intensity is uniform. (d) Minimum intensity projection of (c) (thick-ness: 30 μm,posterior direction) shows the cytoarchitecture. (e)Maximum intensity projection of (c) (thickness: 200 μm,posterior direction) shows vascular architecture. (f) Enlarged viewofthe cortical region indicated by the white box in (c) shows simultaneous cross-sections of cells (black) and blood vessels (white). (g) The enlarged view of the cortical region indicated by thewhite box in (d) shows the laminar cytoarchitecture in the neocortex. Layers I to VI can be clearly distinguished. (h) Enlarged viewof the cortical region indicated by thewhite box in (e) showsthe vascular architecture, including capillaries, in the neocortex. The diameters of the large penetrating vessels are approximately 23 μmand thediameter of small capillaries are less than 3 μm.Note that the capillaries are interconnected with few gaps, suggesting complete labeling of the vascular network.

5J. Wu et al. / NeuroImage xxx (2013) xxx–xxx


In progress: Modeling Neural Glial Vasculature Coupling

Bruno Weber, University of Zürich

52

In progress: Scaling infrastructure to the whole rodent brain

High Performance Computing Platform

53

Provide the project and the community with: •The computing power necessary to build and simulate models of the brain. •Develop new supercomputing technology, up to the exascale •Drive new capabilities for interactive computing and visualization.

High Performance Computing Challenges

54

Computa.onal Complexity

Memory Requirements

O(1 MB)

O(10 GB)

O(1 TB)

O(100 TB)

O(100 PB)

Single Cell Model

Cellular Neocor.cal Column O(10,000x cell)

Cellular MesocircuitO(100x NCC)

Cellular Rat BrainO(100x Mesocircuit)

Cellular Human Brain O(1,000x Rat Brain)

O(Gigaflops) O(10 Teraflops) O(100 Teraflops) O(1 Petaflops) O(1 Exaflops)

EPFL 4-‐rack BlueGene/L

CADMOS 4-‐rack BlueGene/P

Reac.on-‐Diffusion O(1,000x memory)

Reac.on-‐Diffusion O(1,000x memory)

Reac.on-‐Diffusion O(1,000x memory) Glia-‐Cell Integra.on O(10x memory) Vasculature O(1x memory)

Par.cles O(10-‐100x memory) Reac.on-‐Diffusion O(1,000x memory) Glia-‐Cell Integra.on O(10x memory) Vasculature O(1x memory)

Reac.on-‐Diffusion O(100-‐1,000x performance) Plas.city O(1-‐10x performance)

Local Field Poten.als (1x performance)

Reac.on-‐Diffusion O(100-‐1,000x performance) Glia-‐Cell Integra.on O(1-‐10x performance)

Vasculature O(1x performance) Electric Field Interac.on O(1-‐10x performance)

Plas.city O(1-‐10x performance)

Reac.on-‐Diffusion O(100-‐1,000x performance) Glia-‐Cell Integra.on O(1-‐10x performance)

EEG (1-‐10x performance) Vasculature O(1x performance) Plas.city O(1-‐10x performance)

Behavior O(10-‐100x performance)

Reac.on-‐Diffusion O(100-‐1,000x performance)

Medical Informatics

55

•Federate clinical data from hospital archives and proprietary databases, while providing strong protection for patient data. !

•Enable researchers to develop data-driven “biological signatures” of diseases. !

•Develop new approaches to understanding the causes of disease and identifying effective treatments

Disease signatures

56

Disease signatures

57

Neuromorphic Computing Platform

58

Simulate models on low power hardware chips. Build off of BrainScaleS and Spinnaker projects to provide ability to run large-scale simulations at or beyond real time with low power consumption.

Neurorobotics Platform

59

Enable closed-loop simulations of behavior. Virtual bodies, sensory input and environments to couple with the simulations. This platform is key to providing sensory input to the simulations and depicting the motor outputs.

Applications: Understanding principles of cognition

60

Figure 28: Use of the HBP platforms to study mechanisms and principles of cognition: comparing simulated and biological humans and mice on the same cognitive task (touchscreen approach)

Use Case 1: Tracing causal mechanisms of cognition

Benchmark species

comparison

In silico species

comparison

Simulation- robot

closed loop

Simulation- robot

closed loop

Virtual- biological

comparison

Virtual- biological

comparison

Applications: Developing new drugs for brain disorders

61

Figure 29: Use of the HBP platforms to accelerate drug development. Optimizing drug discovery and clinical trials

Use Case 2: Developing new drugs for brain disorders

Model of healthy brain

Model of diseased brain

Target identification

New drug

Clinical trials

Side effects

In silico search for treatments

Computational chemistry &Molecular Dynamics simulations

Preclinical trials

Binding kinetics

Multi-level biological signature of disease

Applications: Developing neuromorphic controllers

62

Use Case 3: Developing neuromorphic controllers for car engines

Figure 30: Use of the HBP platforms to accelerate the development of future computing technologies. Four steps: brain simulation; simplification; rapid prototyping; low-cost, low-power chips

Biologically detailed brain model

SELECTthe most suitable models of cognitive circuitry

EXPORTsimplified brain model and learning rules to generic neuromorphic system

Simplified brain model

TRAINneuromorphic circuit rapidly in closed loop with simulated robot in simulated environment

EXTRACTcustom, compact,

low-power, low-cost

neuromorphic designs

Cross-cutting Activities

•Explore ethical, social and philosophical implications •Engage public in dialogue

63

•Provide young scientists with transdisciplinary knowledge and skills

•Generate widespread awareness of results, open access policy (publications, data, software tools)

•Ensure a cohesive and integrative effort •Collaborate with other European and international efforts

Ethics

Education

Dissemination

Governance

Participating Scientists

64

Figure 38: Scientists identified to participate in the HBP. The scientists listed have agreed to participate in the HBP, if it is approved as a FET Flagship

The HBP Competitive Calls Programme

65

~20% of funding (~200M€/10 years) allocated to open calls

The Initial HBP Consortium

Figure 40: Research institutions identified to participate in the HBP, if approved as a FET Flagship Project

Initial HBP Consortium

Europe

United Statesand Canada

Argentina

Israel

Japan

China

66

For more information and a full list of leaders, partners and collaborators !please visit:

www.humanbrainproject.eu