MARKRAM, Henry, HBP, April 24, 2014

MARKRAM, Henry, Human Brain Project,  April 24, 2014

Henry gave this keynote at TSC; He is PI of HBP.

1B euros to do this by 2023:  whole human brain model;

also want to build human neuromorphic computing chips;

want an integrated understanding across organizational levels;

Will NEVER map the human brain experimentally  (he realized years ago)

we need a data-driven algorithmic reconstruction for the brain:

only a pinhead is what we did in the rat somatosensory cortex;

worked on it for 20 years;  1000s of experiments;

even with 10,000 neurosci working for 10 years: will not map even that (if you do it directly)

hundreds of difft types of neurons;

55 types of neurons in the microcircuit  (many more inhibitory neuron types)

he shows a cortical column with 31,000 cells

layer 6 with 10K cells  etc.

every step of reconstruction must be validated;

do  in silico immunohistology

and compare that to in vitro immunohistology;

now just talking about morphological types  (not  molecular types)

600M possible tangencies in cortical column;

neurons are not growing  toward neurons!!!

it is an accident when they cross one another =>

that is a POSSIBLE location for a synapse...

can trim down from 600M down to 30M synapses

1)  INCIDENTAL synapse location  but get only

2)  fractional conversion into synapses;

then 3) and 4)  but must validate it...

validate against bouton density;  # of inhib synapses, # on soma;

there are 2,258 unique bio viable pathways  btwn  various types of neurons

35M intrinsic synapses;

will get better and better;

now get wiring diagram of this piece of brain;

you get a micro-connectome;

can calculate  how much is coming from outside... abt 75 to 80% of synapses are

coming from the outside of the column.

only about 300 of these 2258  constitute most of these synapses

 (this follows a power law... but must not ignore

 the other 2000 types of synapses cuz

they add up to equal the influence of the 300);

we predicted the detailed synaptic anatomy;

we are predicting elements of biology;

perin, berger, markram, PNAS 2011

common neighbor rule: dense clusters:

but no learning involved  in these clusters: these are innate building blocks.

synaptically clustered neruons emerge from the geometry of the neurons;

master of the universe is the layer 5 pyramidal cells!

morpho-electrical diversity

207  morpho-electrical types in 2 week old rat.

have done many validation tests;

important to challenge these models;

we  go further to simulate the synapses:

depends on ion channels;

can predict synaptic dynamics;

predicts the synaptome of neurons.

on avg receive input from 217 neurons...

35M synapses in size of pinheads;

he shows neocortex

this is in P14  (2 week old rat) activity;

we also build in nissl staining:

mesoconnectome:  from Allen Institute;

(within a brain region)

during the first 5 years: will be mouse brain... (to get a test data set)

building virtual avatar of mouse:

do the Morris water maze and maze with cheese at the end;

want to see function across all levels;

*

in silico brain slice simulation:

how does this piece of tissue work:

what conditions should I put the tissue in;

(eg  excitat and inhib synapses have difft sensitivites to calcium;

inhibitory neurons are much less sensitive to calcium

by moving calcium can change from synchrony to asynchrony;

if you  move to asynchrony then the clusters dissolve;

differential cluster survival along the SA (synach/asynch) spectrum...

with synchrony you have millions of types of clusters;  but with Asynchrony,

most of these clusters die until all you have is one type;

the network lives on  a sync/ asynchron spectrum

synchrony =>  fast reaction rate, spreading activity, all possibly clusters;

a few million neurons are probably the ones that are pivotal in the entire brain

(that are in the clusters)

big consortium:  80 institutions;  150 PIs;

2000 PhDs; 20 European Countries;

includes  USA, Israel, Japan, China, Canada;

he shows a project map with  circular nodes with  all the subprojects of the HBP:

Q  david chalmers:  how do you know if you've got it right?  in terms of behavior

A  its all bottom up  (no edges, etc; no features)  it is NOT:  we want a certain behavior

(the neuromorphic guys will do more of that);

this is a computational microscope:  we are interested in the MECHANISM.

this is a snapshot in time  (not a developmental model)

we reconstruct a model at one point in time;

Q  astrocytes?:  A  yes:  will put in the glia;

MARKRAM, Henry, HBP, April 24, 2014

Attachments