Player is loading...

Embed

Copy embed code

Transcriptions

Note: this content has been automatically generated.
00:00:00
Yes so you you all had some cream oh
00:00:09
have double cream so I'm going to try
00:00:12
and stay awake longer than you as as
00:00:15
you digest that and talk to you about
00:00:19
the medical informatics platform that
00:00:23
is one of the brief unanimous future
00:00:26
medicine of the human brain function.
00:00:28
So this is the the sort of response to
00:00:33
Michael stated American centric usable
00:00:37
as the european brain project is the
00:00:39
first big brain project and is that for
00:00:42
euro centric you know like you you
00:00:44
still lot I think I think that the
00:00:50
number of motives or motivations. And
00:00:54
and most of them are very clear to you
00:00:56
the the first is that the populations
00:00:59
of ageing that as you get the age of
00:01:02
eighty one in five you will become
00:01:04
dependent on somebody because of your
00:01:07
cognitive decline and what's not so
00:01:11
well known is that psychiatric diseases
00:01:15
disease of the young is a fatal disease
00:01:19
sometimes. So the side or even homicide
00:01:24
and there's also a disease that takes
00:01:27
people out of active economic life
00:01:31
probably for the lifetime. So as we
00:01:34
think about the generation about
00:01:36
children and our grandchildren. We not
00:01:38
european social democracies paid for
00:01:42
now yeah when we mail I for a pensions
00:01:47
every time. We have to realise that
00:01:50
there is an economics in only coming
00:01:52
along much greater majority and one of
00:01:57
the problems that with all of this is
00:01:59
that we're coming up against and
00:02:03
intellectual brick wall not in
00:02:06
intellectual one but also the medical
00:02:07
prequel. I have practised neurology for
00:02:11
forty years under top three months ago
00:02:13
and at the end of that time I've seen
00:02:16
great things happen usually when the
00:02:20
closest one and not always in multiple
00:02:22
sclerosis we still don't know the
00:02:24
courses but we don't bring patients
00:02:27
impossible six months keep than the
00:02:30
while I have the pencils treated turn
00:02:32
them in to cushion wide individuals.
00:02:34
And further not again after nature's
00:02:37
taken its course and they've recovered
00:02:38
naturally we now chickens out patients
00:02:42
we have active treatments which is
00:02:44
symptomatic and useful the problem is
00:02:47
that with dementia that is not the case
00:02:50
with the mention there are a number of
00:02:54
issue so first of this that you can be
00:02:57
demanded. And still very powerful no
00:03:03
I'm using a I'm using technical terms
00:03:07
so yeah so please don't get excited for
00:03:12
example if you just read some of the
00:03:13
latest speeches if you listen to what
00:03:17
his wife said subsequently and so on in
00:03:20
these second to Ronald Reagan very
00:03:23
powerful figure brought about a change
00:03:25
in the world. And was when giving
00:03:28
speeches talking of the point using the
00:03:30
allegiance on screen a grammatical. And
00:03:33
he when he left office is she pretty
00:03:37
much went into very rapid decline
00:03:38
simply because he's been so well
00:03:40
supported. Now this simply brings
00:03:43
across the fact that with adequate
00:03:45
support cognitive decline can be hidden
00:03:48
what helped with some of these diseases
00:03:55
the veneer of social grace time domain
00:03:58
for much longer than cognitive function
00:04:01
remains. So what do we know about this
00:04:07
what courses alzheimer's disease.
00:04:10
Please just remember alzheimer's
00:04:11
disease is one of probably fifteen plus
00:04:14
courses of dementia which most of you
00:04:17
know the we often get on the main
00:04:20
feature mixed up. And many people talk
00:04:23
about alzheimer's as the those the
00:04:25
whole field of the major what are the
00:04:28
mechanisms were the role jeans other
00:04:31
than and the ten percent where there is
00:04:33
an association without accosted
00:04:36
mechanism. What's the role of abnormal
00:04:39
proteins tombstones vocals is why do
00:04:42
people get annoyed than the cognitive
00:04:44
decline why do people with cognitive
00:04:46
decline an ageing not have I'm alone it
00:04:51
why did we concentrate on that the
00:04:53
mouse is of new transmission project in
00:04:55
relation past okay I mean I'm on the
00:04:58
irreverent sort of guys you've noticed.
00:05:01
I will tell you why we have
00:05:03
concentrated very great resources on
00:05:06
about it is because the first time to
00:05:09
come into the topological the about
00:05:11
three university we get a slice bread
00:05:14
and a little bit of Congo red and
00:05:15
reported of and we wash it off. And you
00:05:18
don't even need just have a microscope
00:05:20
to see them it's easy to so it's easy
00:05:25
for them ask okay slightly more complex
00:05:28
my cynical conclusions are the elder
00:05:33
that treated parkinson's disease
00:05:35
parkinson's disease. Responded to and
00:05:37
seem to be a mother I mean a jig you
00:05:40
need managing does or the money found
00:05:43
the absence of certain parameters cells
00:05:46
in this instance no not and they could
00:05:48
united cells we made of a metaphorical
00:05:51
john these are now with drugs which are
00:05:54
the most expensive placebos that we
00:05:56
have not according to me according to
00:05:59
nice which is you know is the regulate
00:06:01
reagents united kingdom. So maybe but
00:06:04
so we have major major problems how do
00:06:09
we prevent treat well can be diagnosed
00:06:13
with the answer is we're not very good
00:06:15
at that the latest surveys of the
00:06:19
American pathological technical
00:06:21
literature suggests that if we talk
00:06:24
about alzheimer's disease and we take
00:06:26
as the gold standard the presence of
00:06:28
your finger literate angles and I'm
00:06:30
annoyed in the relevant parts the
00:06:32
campus we make a thirty to forty
00:06:34
percent error in the best possible for
00:06:35
the five days day actually no models
00:06:42
probably thirty to forty percent
00:06:45
outside soft yeah JT is just that
00:06:47
they've compensated you know that in
00:06:51
parkinson's disease you need to do
00:06:52
seventy percent of the one I'm allergic
00:06:55
Nike restraint on your arms with some
00:06:57
about three hundred thousand on each
00:06:58
side an infinitesimal small number
00:07:00
compared total number one before you
00:07:02
get the first symptoms the first
00:07:06
symptoms. So now some disease we sort
00:07:08
of yes fifty percent loss before you
00:07:10
start showing subtle changes in your
00:07:12
cognition a terrible thing for the
00:07:15
doctor but actually a gateway to
00:07:18
potential treatment of the preclinical
00:07:20
disorder no I've seen a doctor say to
00:07:24
me how could you possibly treated
00:07:25
normal person. So I say that if I could
00:07:28
diagnose it as a good but even if I
00:07:31
don't recognise it I've been using I've
00:07:34
been using inoculations against against
00:07:37
and vaccinations against diseases in
00:07:39
healthy children. And you accepted so
00:07:41
why this in such an unusual for do
00:07:45
symptoms matter well it's I've just
00:07:47
described on your little what waits to
00:07:49
this pathologies the gold standard
00:07:51
every brained one point four kilograms
00:07:55
average brain off the death of outside
00:07:58
the patient something between seven and
00:08:00
nine hundred rounds. So it's completely
00:08:03
wasted. It's in stage what the doctors
00:08:08
find when they had any stage little
00:08:10
boxes of the renal doctors on that end
00:08:12
stage renal but is not very much they
00:08:15
found out when they went in there
00:08:16
rather than they found the mechanisms
00:08:19
do we compensate well I've mentioned
00:08:22
about a two facts the brain is a highly
00:08:24
redundant talking the plasticity
00:08:27
depends on that redundancy. So these
00:08:30
are the sorts of issues that we have a
00:08:32
had alright. So let's ask a basic
00:08:36
scientist for how actually don't just
00:08:39
don't do that. But I think they sure
00:08:41
let's just do it today. So across
00:08:45
spatial scales from Paul brain up to
00:08:49
sub molecular resolution that land
00:08:52
something cellular. We have a number of
00:08:55
methodologies something which you one
00:08:57
about prices how do we do a science
00:09:01
which focus we're not worried about
00:09:05
someone else doing something with a
00:09:06
slightly different molecule we wanna
00:09:08
know about our molecule we want to know
00:09:10
about our microscopy we want to know
00:09:13
about our gene we even get to that's
00:09:16
the extent of taking a mouse knocking
00:09:18
out our gene men calling it and someone
00:09:20
else where the behavioural
00:09:24
correspondence between the mouse and
00:09:26
the human is completely on so this is
00:09:31
the way we do in your side it's a good
00:09:33
thing that said something which I
00:09:35
understood what you meant "'cause" I
00:09:37
know that but you said something wrong.
00:09:39
She said we have a lot of knowledge
00:09:41
sides we actually have a lot of facts
00:09:43
seniors and then she went on to say we
00:09:46
had no overriding scaffolding which to
00:09:49
hang these facts something people might
00:09:51
colour theory or blueprint to whatever
00:09:53
you like. And that's true. We don't
00:09:57
have anything that we have vast amounts
00:09:59
of data these are the number of papers
00:10:03
published in neuroscience since
00:10:05
nineteen ninety two twenty twelve when
00:10:07
out about a hundred and fifty thousand
00:10:09
papers they yeah your science becoming
00:10:11
more more popular plenty and plenty of
00:10:14
facts. But do we have an integration
00:10:16
plan do we represent had later in the
00:10:20
same way a little integration to link
00:10:23
across spatial levels is there any plan
00:10:27
to transfer respects the knowledge. And
00:10:30
knowledge in June and in animals to
00:10:32
humans no plan to go beyond the SN five
00:10:37
this in five is purely based on
00:10:40
symptoms and signs one of the greatest
00:10:44
discoveries of the human genome project
00:10:47
is the following that Huntington gene
00:10:50
which is dominant which is completely
00:10:54
protective of the presence of presents
00:10:56
Huntington disease. In the future if
00:10:58
the patient survives long enough can
00:11:02
present in any of five if not six
00:11:04
different ways. So there is no direct
00:11:07
correspondence from gene to disease
00:11:10
manifestation taken the other way
00:11:13
around if you take spine to set about
00:11:15
the generations but you're reasonably
00:11:16
well known neurological condition there
00:11:19
are currently twenty five partially
00:11:23
dominant not fully management mutations
00:11:30
which course exactly that's into twenty
00:11:33
five months and they seem to be
00:11:34
associated in any way along well wait
00:11:37
on the proximity on trying to censor
00:11:39
anything similar. So we have a a major
00:11:42
problem which is that we have to force
00:11:44
ourselves to leave the linear well into
00:11:48
the complex well we have to stop
00:11:51
talking about T tasks bases we really
00:11:54
did. I mean you know my instigator
00:11:57
eleven attending about peace and T
00:11:59
tests at the age when they're doing
00:12:02
that. They going to college actually
00:12:03
already know about about complex
00:12:06
mathematics things that you can not in
00:12:09
that situation. But which are
00:12:11
absolutely vital the fact that if a
00:12:13
causes B in this context it causes this
00:12:16
but that this thing and a which causes
00:12:18
me in another context "'cause" it's
00:12:20
something completely different that is
00:12:22
nonlinearity and that is what we did
00:12:24
with in the highly complex organisation
00:12:27
of this remarkable organ now the other
00:12:31
thing but we are guilty of is ignoring
00:12:34
information technology whether we
00:12:36
medics whether we basic scientists we
00:12:38
have not taken up information
00:12:40
technology in the way that
00:12:41
meteorologists have in the way that
00:12:44
people who could start aircraft have in
00:12:47
the way that the astronomers have
00:12:50
somebody may think the paper and a half
00:12:52
ago nature which predicted the shape.
00:12:56
And construction of the universe
00:12:58
currently from the all the information
00:13:01
tended to skirt around the well about
00:13:03
this after the first thirty thousand is
00:13:06
beyond the big bang on the
00:13:08
correspondences absolutely remarkable
00:13:10
the data finding of hypotheses in the
00:13:13
data not thinking with the four I
00:13:15
mention Morgan about a complex machine
00:13:18
that has some millions of dimensions it
00:13:22
would be total serendipity to be able
00:13:24
to think up some you define period the
00:13:27
brand from a human brain the brain
00:13:30
cannot understand the bright and
00:13:32
someone said the brain needs a super
00:13:34
calculated on the the right brain
00:13:36
pacific calculator will understand the
00:13:38
bright. That's the question the fight
00:13:40
like route you on the side just I don't
00:13:42
have played. But I still believe that
00:13:44
to be true at least as my ruling
00:13:47
hypothesis. I don't need to go through
00:13:50
this list just to remind you that most
00:13:53
of these things. Well it certainly in
00:13:57
terms of Google wherein the guy rush
00:13:58
eighteen years ago and I'm now with a
00:14:00
hundred billion things changing
00:14:03
extremely fast that changing so fast
00:14:06
that a lot of people are still talking
00:14:08
about clouds and databases when they're
00:14:10
talking about going to big data just
00:14:16
remember that I'm gonna come back now
00:14:19
in medicine actually when you go to
00:14:22
nonlinearity too complex analysis. And
00:14:25
complex statistics then you can
00:14:28
actually do some very interesting
00:14:29
things. This is a paper published in
00:14:32
two thousand and eight it shows that
00:14:35
we've normal clinical diagnosis of
00:14:37
alzheimer's disease. This particular
00:14:40
group of about eighty patients actually
00:14:43
got correct classification about eighty
00:14:45
percent which is pretty good compared
00:14:47
to the seventy percent that's been
00:14:49
found everywhere else. If we used
00:14:52
postmortem pathologist ago stuff. So
00:14:56
the rest what demons but from all the
00:14:59
courses. And they were normal subjects
00:15:04
if we use a very simple nice scan
00:15:08
anatomical T one way to maximise
00:15:11
contrast between grey and white matter.
00:15:13
And put it through a technique drive
00:15:16
from machine then in called us support
00:15:19
back to machine the machine that
00:15:21
classifies in binary fashion after
00:15:24
having been trained on a group or
00:15:27
cardinal representatives of the two but
00:15:32
I got hold of a lot of my friends in
00:15:34
American Japan in europe. And got hold
00:15:38
of scans of patients who don't oppose
00:15:40
Morton who either had only outside this
00:15:43
disease. Well actually normal not
00:15:46
normal whatever cognitive sense what
00:15:48
these small incompetence sense and did
00:15:50
not have plaques and tangles and all
00:15:52
the relevant things. So we train the
00:15:54
SVM on that and then when we looked at
00:15:57
a group of patients who also went to
00:16:00
put all that stuff was about to
00:16:02
postmortem and had eighty on the melody
00:16:04
confirm. We're up to ninety five
00:16:07
percent with once again. Now that
00:16:09
hasn't been taken up by radiology I'm
00:16:13
not sure why but that's a fact and it
00:16:16
may just reflect that medicine Singers
00:16:20
forward slowly another way which
00:16:23
informatics can help is to federated
00:16:25
data. So an image is worth a thousand
00:16:29
words it is it's probably worth a
00:16:32
million data points you have to think
00:16:35
of these images as sets of pixels where
00:16:37
each pixel can be referred to any other
00:16:40
database to could pick up information
00:16:43
that is been found out about that site
00:16:44
in the right. So these a like
00:16:47
informatics encyclopedias if you like.
00:16:51
You are also techniques this is a
00:16:52
standard human brain where you could
00:16:54
strip of the face and the skyline can
00:16:56
strip of the grey matter you can look
00:16:58
at that connect and you can look at
00:17:00
this you can look at the function you
00:17:01
can look at the structure this is a an
00:17:04
interesting image because it's an image
00:17:06
of the development of the mouse brain
00:17:09
through to the human brain so this is
00:17:11
millions of years. Um also about I know
00:17:14
it does not a linear thing but let's
00:17:16
just not like that way. So this is this
00:17:19
is a different so they this is another
00:17:21
sort of image which uses special
00:17:23
techniques it's a brain that's been put
00:17:24
into an oil which is a rented it
00:17:27
transparent to the particular
00:17:28
wavelength monochromatic light. So that
00:17:31
you could do colleges staining in in in
00:17:34
in a postmortem again but in a very
00:17:38
delicate precise that following the
00:17:40
fibres of the of the so this is a this
00:17:43
is something that was technique that
00:17:45
was discovered in the polytechnic of
00:17:47
yeah and Austria sometimes taken up by
00:17:51
a and cool okay. So if there are ways
00:17:55
in which we can bring data together and
00:17:57
favourite. But the other thing that we
00:17:59
have to do as we go forward is that it
00:18:02
forward from a classical cartesian
00:18:05
model complex systems cannot be
00:18:07
analysed in this way competition model
00:18:11
where a mental feel really is
00:18:12
mathematically expressed confronted
00:18:14
with relevant data nice objectives
00:18:17
thing that and then there's
00:18:18
parameterisation of the model to
00:18:21
optimise it more introduction of the
00:18:25
human bias before five to G that's what
00:18:29
we are asked to do by the NIHOYAMRC
00:18:33
with the ERC when writing a grounds
00:18:36
pure reduction decides you have to have
00:18:37
a good hypothesis have to have a good
00:18:39
design and so on. So but in fact there
00:18:42
are other tell technologies and there
00:18:44
are other ways of thinking about data.
00:18:47
And someone's already talked about
00:18:48
multivariate data a multimodal data.
00:18:52
And large amounts of data. And the
00:18:54
algorithms to the used in order to go
00:18:56
through those large datasets in order
00:18:59
to demonstrate mathematical structures
00:19:02
by correlational by classification of
00:19:04
by both which produce hypotheses
00:19:07
essentially which also mathematically
00:19:09
express just the same as yet and which
00:19:12
can then be taken to the doctors to the
00:19:14
physiologist to the therapist to the
00:19:17
pharmacology ecologist whether they can
00:19:18
be allowed to characterise in in their
00:19:20
domains. And here the hypothesis comes
00:19:23
not from the imagination but it comes
00:19:26
from the data themselves and we live in
00:19:28
an extremely data rich environment. So
00:19:31
this is what the human brain project is
00:19:34
based on simulation modelling from
00:19:36
bottom out looking five policies. And
00:19:40
generating rules which determine the
00:19:43
relationship between each left one of
00:19:46
the great criticisms at the beginning
00:19:47
of the time right project was this is a
00:19:50
ridiculous thing to spend so much money
00:19:52
on because there's an infinite number
00:19:55
of degrees of freedom that underlie
00:19:57
this problem presented in this way. But
00:20:00
it just takes a little thought to say
00:20:02
well hang on them moment. It'll start
00:20:04
with the in a full bass says the
00:20:07
degrees of freedom car like that. And
00:20:10
then it builds up in complexity. And at
00:20:12
each level the rules which should we
00:20:14
know the rules for the NA two are they
00:20:16
or denied the messenger and it running
00:20:18
to proteins we don't know the rules
00:20:20
about how the distributed amongst the
00:20:22
different cells we don't know how they
00:20:24
then go on to form the microseconds of
00:20:26
the circuits well the systems if there
00:20:28
is a if there's exist. They are at the
00:20:31
moment figment of our imagination but
00:20:34
we can work out the rules but after
00:20:36
rolling fifty seven o'clock actually
00:20:39
and hodgkin's got noble problems for
00:20:41
looking up one of those rules something
00:20:43
truly conduction better plenty of rules
00:20:46
and all of them can be expressed
00:20:47
mathematically and it's a question of
00:20:49
working about because at each level as
00:20:52
you provide more more rules. So you
00:20:54
decrease the degrees of freedom until
00:20:56
competition will come out of it like a
00:20:58
little founded even different and this
00:21:03
is something that is very important
00:21:04
that clearly you can have rules
00:21:06
developed at different levels and try
00:21:08
and integrate them through each other
00:21:10
but the principle drive has to be from
00:21:12
bottom up. And that is exactly what
00:21:14
we're on about sorry in the medical
00:21:16
informatics platform sub project but
00:21:19
what we decided to do is definitely
00:21:23
federated all the data in your
00:21:25
hospitals not while you're because
00:21:27
because you're hospitals also actually
00:21:29
found it and for every piece of data on
00:21:32
me is mine but it's also a little bit
00:21:34
of my countrymen state and my european
00:21:39
colleagues data because I pay for it as
00:21:40
well. "'cause" the but I by taxes. So
00:21:45
we thought People will probably accept
00:21:47
that that actually used for medical
00:21:49
research and that's a probability this
00:21:51
we can funding sources signs doctors
00:21:54
are just I think it's because the
00:21:57
conservative but other people think
00:21:58
it's with that could be we were taken
00:22:02
to from behaviour on your psychology
00:22:05
brain image England physiology lot ugly
00:22:08
too medics anything you like and then
00:22:10
we'll meet into those data and that
00:22:13
would be an interesting thing two and
00:22:15
then we would die try to mindless data
00:22:18
to find call of models which we will
00:22:21
simulate and obtain knowledge fall in
00:22:24
order to give us biological signatures
00:22:26
of disease which will include a
00:22:28
biological information that will give
00:22:31
us ideas about new targets which come
00:22:35
out of the biology new targets trials
00:22:40
where we will have to find the
00:22:42
diagnosis you will have a definite
00:22:45
diagnosis whatever it is however its
00:22:47
name. So that you can enter draws on
00:22:50
ten verses ten instead of thirty
00:22:52
thousand versus the pounds you all have
00:22:56
a way of giving patients some sort of
00:22:59
idea about pro basis because of
00:23:02
homogeneous group of people who
00:23:04
identical in terms of that genetics the
00:23:08
physiology and anatomy. And all the
00:23:11
other aspects of brain disease will
00:23:13
constitute the disease. So this is the
00:23:15
transformation DS descent five to
00:23:18
descend six but we think should happen
00:23:21
the integration of biological
00:23:23
information into the clinical
00:23:24
information now what sort of data do we
00:23:28
have you have a truthful quality which
00:23:32
is kinda cool it's terribly corporate
00:23:35
quality but a solution to that is to
00:23:37
get millions of people state because we
00:23:41
know that then solves the problem. I
00:23:43
think the whole its buttons on that so
00:23:45
we did that that crap data. But they
00:23:48
collected two three four months first
00:23:50
before useful signal came out of the of
00:23:53
the noise which you prints scroll down
00:23:55
but not the amount of data we also have
00:23:58
research data which is a very high
00:23:59
quality both of which we have very
00:24:00
little volume. So the first there is
00:24:03
that we we use this to validate. I was
00:24:06
also this or at least to validator
00:24:08
methods So this I've just said it's
00:24:12
protection of this data is the big
00:24:14
problem. You want to be I'm sure that
00:24:18
no one's going a cover up the results
00:24:20
of your test when they get a fiddling
00:24:22
around with the data stalls in the
00:24:25
hospitals you also multidimensional
00:24:27
that you're protected for privacy
00:24:29
vertically if you've got some funny
00:24:30
disease. That currently is not up to
00:24:33
acceptance by the rest of the
00:24:35
population research databases are also
00:24:39
protected I got healthily but I
00:24:40
actually for identical culturally this
00:24:42
is this is the victorian site is is put
00:24:45
is data into a database and then
00:24:47
surrounded with three five Miles
00:24:49
somewhere that anyone else look at it
00:24:51
is if you have data equals no hope for
00:24:53
something else to remember that this
00:24:54
brain we'll have to work a little bit
00:24:56
before you can get a noble price
00:24:58
actually this is being tackled by the
00:25:01
NIH and now by the yeah see I'm by the
00:25:05
MFC seventy by the Wellcome trust you
00:25:07
don't get it man and a simple you
00:25:08
dangerous. And and I recently that's
00:25:12
the crest of course when he told us
00:25:14
that they put all the raw data out on
00:25:16
ice about how often if you think is up
00:25:18
next you did you have to know what
00:25:21
you're looking for you you have to have
00:25:22
some ideas all you have to be about the
00:25:24
algorithms or you have to be able to
00:25:26
generate hypothesis from the big data
00:25:28
and then we have pharmaceutical
00:25:30
databases and this approach tactical
00:25:32
actually the interesting thing is that
00:25:34
pharmaceutical companies are very rich
00:25:35
which means they got a lot of
00:25:36
intelligent people in them. So if you
00:25:39
go to them and say actual not
00:25:40
interested you know all your data how
00:25:43
Justin traded interest in your field
00:25:46
trials. Well the quite a number of
00:25:48
those on the that pages percent then to
00:25:51
some button. And that you're not
00:25:53
interested only file files at all. I'm
00:25:55
interest in the placebo once again felt
00:25:57
ross. Well one company or about mention
00:26:00
its name it begins with this has give a
00:26:03
stop and patient bang like that. So I'm
00:26:07
sure the pharmaceutical industry but
00:26:08
it's very structured data will be very
00:26:11
very helpful in the sort of and that if
00:26:15
we could show that we can the was
00:26:17
protecting privacy. And not corrupt. So
00:26:21
this is very important. And we made a
00:26:23
decision right up from the release all
00:26:26
straight away. So the moment you you
00:26:29
personalise but there's all of you know
00:26:31
know you but right now you know and no
00:26:35
sex then you could be found in even a
00:26:37
very large database. So in addition to
00:26:40
the personalisation we decided we had
00:26:42
to work with aggregate data at least
00:26:46
ten patients together so you can't do
00:26:48
that. Once you have an aggregate then
00:26:52
you're and don't non anonymous when
00:26:54
you're anonymous in in your parental
00:26:56
you're not subject to any of the rules
00:26:59
because you can't get back. And indeed
00:27:01
we do as you'll see in a moment double
00:27:03
aggregation then people say well how do
00:27:06
you know about statistics actually
00:27:08
statistics like I'll find a little bit
00:27:11
of bodies in this a little bit of
00:27:13
fiddling with them that equation like
00:27:14
that. And the statistics actually
00:27:16
easier to do than anywhere else. I can
00:27:19
say this is what social sciences
00:27:22
colleagues of very very helpful lots
00:27:25
and lots of people people how effective
00:27:27
repetitive broken sent to the shape of
00:27:28
it. There's a project going forward now
00:27:31
taking genetic date implied data from
00:27:33
everyone who comes in they opted in but
00:27:37
a boss whether they want to opt out
00:27:39
normal up to about ten percent of
00:27:41
remote and that we can live with that
00:27:43
and the question of retrospective a
00:27:46
prospective data disappears if you can
00:27:48
guarantee privacy and then you have a
00:27:52
management of the ethics. Well
00:27:54
basically want to leave with the local
00:27:55
ethics committees because they have
00:27:57
functions are remarkably well breathing
00:28:00
found in the pharmaceutical industry
00:28:01
the that have just one or two little
00:28:03
problems recently but in a very large
00:28:06
extent pharmaceutical industry and set
00:28:09
in the hostels and so on this is then
00:28:11
the bedrock of why people crosstalk
00:28:13
does essentially. And then the value
00:28:16
and credibility of science which of
00:28:17
last fifty is is going enormously as
00:28:19
people have begun to use things like
00:28:22
computers oh so that's an interesting
00:28:24
thing that happens years but we just go
00:28:27
back the ticket oh yes because this is
00:28:33
so so this is what we propose this red
00:28:42
line represents a hospital or
00:28:44
pharmaceutical database or research
00:28:46
database it's your five you're in
00:28:48
charge of it you have to guarantee that
00:28:52
it is not corrupted so it must be
00:28:54
touched. And but if it is destroyed you
00:28:58
have copies of side the cold archives
00:29:04
right. So that's what that's that's
00:29:06
what you're interested in and you're
00:29:07
interested in deciding whether data
00:29:10
comes out of doesn't count we ask for
00:29:13
one archive copy where we can make all
00:29:16
the files that that yeah universal.
00:29:19
That's easy down in these days we want
00:29:22
any data to come real time and it will
00:29:25
be de personalised using your de
00:29:27
personalisation algorithms industry
00:29:29
standard top industry standard. So we
00:29:33
never move or corrupt the original
00:29:35
database so that there's a lot of
00:29:37
problems. We line three process on
00:29:40
hardware in your possible we the noise
00:29:44
we've been doing that in imaging for a
00:29:46
long time that's moving we standardise
00:29:49
an anatomically normalise mean
00:29:51
centralisation plus one minus one and
00:29:54
at a local normalisation and we can do
00:29:56
the sub millimetre you know putting
00:29:57
different brains at the same show some
00:29:59
millimetre very quickly about
00:30:01
completely automatically all sorts of
00:30:03
numerical normalisation and then we
00:30:05
have our dataset in your possible which
00:30:09
still you control from the outside are
00:30:14
accredited researchers all the hospital
00:30:18
administrator or the pharmaceutical
00:30:21
epidemiologist can ask a question like
00:30:25
you ask a question of Google the
00:30:27
question is split up. And goes to this
00:30:32
data set which is all prepared. And
00:30:34
looks on the raw data for data that is
00:30:37
relevant to that question take it out.
00:30:40
It's aggregated encrypted filtered once
00:30:44
what make sure there's the personal
00:30:45
better left unsaid back where multiple
00:30:48
aggregates from all the hospitals in
00:30:50
the federation come and and the second
00:30:52
reactor system. And then the statistics
00:30:55
the visualisation the testing the
00:30:57
hypothesis the public health survey the
00:31:01
time I putting my resources into the
00:31:03
right to disease is at this time of the
00:31:05
years of books about time at the end of
00:31:07
the possible management's so can be
00:31:08
done. And that gives the result and
00:31:12
then what's the result of the time
00:31:13
because we store all the problems
00:31:16
because this is continuously being
00:31:18
refined than added to we get rid of any
00:31:21
data aggregated or what because we run
00:31:24
the whole thing with the same technique
00:31:27
in the future. That's the plan. And
00:31:31
that's what hospitals are exactly. So
00:31:33
the shoes is online three bogus online
00:31:37
the the the lab hospital is coming
00:31:42
online now revisiting the subject re an
00:31:45
make small. And we resenting the
00:31:47
national health service and in
00:31:48
particular universities of Oxford
00:31:50
university college London and Kings
00:31:52
college to that seems to denounce in to
00:31:54
discussion at the beginning of December
00:31:57
we have in reserve the hospital bother
00:32:01
universe possible often possible.
00:32:03
There's a whole group of possible as
00:32:05
implemented quite like the idea of the
00:32:07
reasons which might not be entirely
00:32:09
clear to the queries run directly over
00:32:14
the files so this is a state of Israel
00:32:16
at the birch realisation technique. And
00:32:19
the question is is it fast enough it's
00:32:21
faster than using a database on the
00:32:23
data file it loads faster it processes
00:32:26
fast. It's being developed by the
00:32:29
computer science department understands
00:32:31
yeah and a mark in charge at BPFL as
00:32:35
part of a spear. And what it allows is
00:32:39
quite apart from this which is primary
00:32:40
important last onto the but work on
00:32:43
large collections of files because it's
00:32:45
a federation or multiple data format so
00:32:48
audio legacy data can also be queried
00:32:50
it's not just prospective but
00:32:52
retrospective as well if the lower. And
00:32:55
it integrates with what existing tools
00:32:57
so nothing has to be changed you have
00:32:59
to pay fifty thousand put in the
00:33:00
hardware we bring along these software
00:33:03
and away you go. We will work with only
00:33:06
five hospitals in the phase from may
00:33:09
two thousand sixteen through two and a
00:33:11
half years of the next stage and at
00:33:13
that point. We will put it out tender
00:33:16
as a commercial issue. So this is
00:33:20
wealth generation in Europe as well.
00:33:23
You jobs new ways of looking at things
00:33:25
which is exactly what the that program
00:33:27
is designed to do. And what sort of
00:33:29
questions we will be looking at we want
00:33:33
to data mining realtime how this is
00:33:35
getting away. Um I'm not sure what I
00:33:38
can do about this other than switchers
00:33:41
oh but that's very sorry. So we will
00:33:47
there's think which is we want data
00:33:49
mining in realtime continuously
00:33:51
interested like we want to populate a
00:33:53
whole brain disease space which is
00:33:56
multi design a functional with these
00:33:59
disease signatures. So it's absolutely
00:34:02
essential that we don't have precise
00:34:04
diagnosis before we start is absolute
00:34:06
essentially done just do without
00:34:08
sinuses but you do it brain diseases
00:34:10
because that's way we tease apart the
00:34:13
different components. And that's the
00:34:15
way we maximise these differences
00:34:17
between the real diseases and that's
00:34:20
the way we can iteratively improve that
00:34:23
sort of diagnostic procedure
00:34:26
transforming distance five and the
00:34:28
distance six E over the years is more
00:34:31
more data collected. And what this will
00:34:34
do is allow one to see where they're
00:34:36
outdated missing so it might direct
00:34:38
research on the right into the right it
00:34:41
is it might ask where other too many
00:34:45
data where data seem complete yeah kill
00:34:47
two with the rest. So there is a
00:34:49
possibility that that this sort of as a
00:34:51
background activity maybe give a lot of
00:34:54
indications we have other clinical
00:34:55
research is required the data
00:34:57
visualisation will be with a hypothesis
00:35:01
and then it may be a demon logical
00:35:04
health services research. And then that
00:35:07
will be standard research hypothesis
00:35:10
testing and clinical trials which will
00:35:13
be carried out in a similar fashion we
00:35:16
have to do all sorts of things to the
00:35:18
data a one sitcoms what sorry when it's
00:35:21
being prepared we have to extract
00:35:23
various components because a date or a
00:35:25
very complex for example would be quite
00:35:28
nice to have four hundred thousand
00:35:29
pixels a brain anatomy in there. But it
00:35:33
would be extremely yeah costly in terms
00:35:36
of computer power so getting out
00:35:38
functional rather brain anatomical
00:35:39
there is in the way that was described
00:35:41
earlier on would be useful so we have
00:35:43
some preliminary data I put in a
00:35:45
carried out the for an additional five
00:35:47
thousand patients. So it's not big data
00:35:49
yet. So it might all be wrong. But it's
00:35:52
interesting. We got we data mostly from
00:35:55
fright which researchers just show you
00:35:58
this is really possible you know how
00:36:01
yeah okay once and lots of clinical
00:36:04
scales the management's tie I I'm a
00:36:06
rice can not complete Freddy patient
00:36:09
every every every person by any means.
00:36:12
And they're not just patients have a
00:36:14
lot of normal people here as well aged
00:36:15
normal people five hundred thousand two
00:36:19
million that's also foresaw subgroup of
00:36:22
them somehow the proteins somehow I'm
00:36:25
my lord I'm we're the blog ramble again
00:36:27
but service five that's ever filed for
00:36:30
by kind of on and we have to drive them
00:36:33
into five types we did this five
00:36:36
thousand on a simple one million excel
00:36:38
spreadsheet you will need for I mean
00:36:41
here's all sorts of different types of
00:36:43
data mining which we're exploring
00:36:45
company because we don't know what the
00:36:47
best data mining techniques will be and
00:36:50
here are the results using the same
00:36:52
colour colour for people to appear
00:36:56
normal one for people who appear hard
00:36:58
time and this is an interesting one
00:37:02
because this one is associated with the
00:37:04
pattern of atrophy which is typical
00:37:07
about simon's apostle that I talked to
00:37:08
previously. It's associated a perry for
00:37:11
a a is just the project is just just
00:37:15
and then there's all sorts of other
00:37:17
interesting jeans and things associated
00:37:22
with it it's largest one it looks like
00:37:24
it's gonna be out fine sees it needs to
00:37:27
go to the doctors who then need to look
00:37:29
at these patients and tell us what they
00:37:31
look like that a lot of blue circles
00:37:33
you think how could you have lots of
00:37:35
moments when you have a normal so
00:37:37
arrange quickly what was normal so it's
00:37:39
slowly you can have normal so
00:37:40
compensating for outside this is normal
00:37:42
so compensate from temporal dimension
00:37:45
normals exactly expected we we were
00:37:50
extremely which by the fact that with
00:37:52
different snow. And you see that many
00:37:54
investments are associated with
00:37:57
different patterns. So you get families
00:38:00
of the thing is with different roles
00:38:02
this guy. And then you get little
00:38:05
outliers all their which is the prime
00:38:08
diseases alzheimer's disease or just
00:38:10
individual things like you're so cross
00:38:14
or dark. Here's a nice of the one a
00:38:16
particular type of pipe a particular
00:38:19
type of of appetites associated
00:38:22
particular types of so the these are
00:38:24
the sorts of visualisations get out of
00:38:26
visualisations related to integrating
00:38:29
the medical side. So you have to
00:38:31
understand categories which are in the
00:38:34
sense that yeah this is from the from
00:38:35
possible and then sub categories and
00:38:37
sub categories. And then there were
00:38:39
nine thousand hundred forty two forty
00:38:42
thousand eight of whatever that
00:38:45
Haitians came through the doors in this
00:38:48
yeah and you can then begin looking at
00:38:51
various other aspects of is how very
00:38:53
symptoms associated with each other you
00:38:55
can look at how they're expressed. Um
00:39:00
what the sex distribution is what the
00:39:02
component of asking the is in their
00:39:05
what the different names used for the
00:39:07
same diseases how many there are
00:39:09
English the same label we have to get
00:39:12
these labels organise get the
00:39:13
ontologies of there's there's an awful
00:39:15
lot of work to be done it. But not make
00:39:17
Google did that that's why you get
00:39:19
sensible results we mostly Google
00:39:21
question. They have all those
00:39:23
ontologies like that in fact there are
00:39:24
many ontologies already out there which
00:39:27
we can look at and then they the
00:39:28
symptoms which are associated with each
00:39:30
other there's one very interesting one
00:39:31
in relation to T alzheimer's disease.
00:39:34
It's urinary tract infection now you
00:39:38
know entrance comments and use the LO
00:39:40
that's easy then you're attractive
00:39:41
function calls itself sinus disease
00:39:44
because there's an association actually
00:39:47
the fact of the matter is if you
00:39:48
declare should you will know that if
00:39:50
you have a doubt sense these and get a
00:39:52
urinary tract infection your
00:39:53
competition the client you actually
00:39:56
infection your condition comes back to
00:39:58
what it was before the client that's
00:40:00
why they're so stupid another example
00:40:03
of like thinking linearly is really not
00:40:06
the thing to do. So I'm coming to the
00:40:09
the integrated view of the medical
00:40:13
informatics platform is that we have to
00:40:17
acquire and federated data. So we have
00:40:20
to capture it and then create work out
00:40:22
the ontologies we've just been talking
00:40:24
about we have to mind the data so we
00:40:26
need medical intelligence tools got
00:40:28
categorisation we need work flyers
00:40:30
really duration we don't know how much
00:40:33
preparation we need because we haven't
00:40:34
tested it on very very large samples
00:40:37
yeah we need integration an operation
00:40:40
uses community outreach hospitals the
00:40:43
people most interested to this at the
00:40:45
moment are aged between twenty five and
00:40:46
thirty five so for all of your office.
00:40:50
So the end of this talk about have a
00:40:52
crowd in front of me wanting to know
00:40:54
"'cause" that's what I get is that also
00:40:58
hospital administrators. They can see
00:41:01
how this can happen with it by with
00:41:05
their way they put the resources. So
00:41:09
that lots of people and the most
00:41:11
interesting people are applications I
00:41:13
recently spoke to a patient group
00:41:16
associated with with traumatic brain
00:41:19
disease as you know we had a fight
00:41:21
inspiration taylor's than last year.
00:41:23
And so they thought they would get me
00:41:24
that because I had looked after a but
00:41:27
they were absolutely fascinated by this
00:41:28
an immediate united or contracts for
00:41:30
websites they want to know how they
00:41:32
could help anything we're prepared to
00:41:34
say about you I some of this work is
00:41:36
extremely boring but you could do it
00:41:38
because it's like drawing lines between
00:41:40
different names there in the they want
00:41:43
to be involved the same as the HIV
00:41:46
patients were involved there is some
00:41:48
involve they to cover the agenda you
00:41:50
guys remember. So this is the platform
00:41:54
and it's going to work for another two
00:41:56
and a half years well with another of
00:41:58
it really is and then we have the
00:42:00
double become something that is taken
00:42:03
off my hands by industrial aren't the
00:42:05
white people service by people create
00:42:08
at this well have an actual so doing
00:42:12
particular types of analyses hospitals
00:42:14
and so we'll be able to work on this in
00:42:16
their own possible or as part of the
00:42:18
federation and if at all possible
00:42:20
directly comes in and says I'm scared
00:42:22
is IT has to be able to switch off and
00:42:24
the whole thing went full yeah because
00:42:26
it'll be distribute it'll potentially
00:42:30
and the cup oracle potentially undercut
00:42:34
data warehousing I'm now fantasising
00:42:36
but but just just just let is dreams
00:42:40
role because that's the way things are
00:42:43
future neuroscience I'm remote
00:42:47
magnificent paper and sell a week and a
00:42:49
half ago. So in some way the arguments
00:42:53
against simulation looking for rules
00:42:56
and trying to produce the first
00:42:58
blueprint of how the brain is organised
00:42:59
different spatial stuff to my potential
00:43:04
noble prize winning by future mats and
00:43:06
that's what I've just shown you win I
00:43:08
mean they don't noble prize but if we
00:43:10
can make a life the people back and
00:43:12
have fun with these or any you don't
00:43:13
know about that already psychiatric
00:43:16
disease. Where my we don't have many
00:43:18
jokes and we still have Sigmund Freud
00:43:20
charter. And future computing where the
00:43:23
computer people but look at the brain
00:43:26
one point four kilograms and the colour
00:43:29
equivalent of two bananas a day and
00:43:32
then they look at one of these enormous
00:43:33
data warehouses I'm sure you've got one
00:43:35
actually which can you know you need a
00:43:38
little mini atomic power station too
00:43:40
and then they say well I know this is
00:43:43
meant to and this is not a so the laws
00:43:46
of physics and the laws of chemistry
00:43:49
applied to both there has to be a
00:43:51
correspondence of some sort. So the
00:43:53
first correspondences to go from Von
00:43:55
Neumann binary computing to what's
00:43:58
called new remote fit computer and
00:44:00
indeed in Heidelberg solvents my leader
00:44:03
that is constructing is is
00:44:06
manufacturing your remote chips which
00:44:08
give you some some stochastic real
00:44:10
response to these signals coming in and
00:44:12
out. So I finished. Thank you very
00:44:15
much. Indeed. And I don't have anything

Share this talk: 


Conference Program

Introduction to the 12th Nestlé International Nutrition Symposium
Thomas Beck, NRC Director
Oct. 22, 2015 · 8:57 a.m.
789 views
Introduction to Session I - Cognitive & Brain Development
Susan Gasser, Friedrich Miescher Institute, Basel, Switzerland
Oct. 22, 2015 · 9:04 a.m.
161 views
The development of a healthy brain
Michael Gazzaniga, University of California, Santa Barbara, USA
Oct. 22, 2015 · 9:16 a.m.
398 views
Q&A - The development of a healthy brain
Michael Gazzaniga, University of California, Santa Barbara, USA
Oct. 22, 2015 · 9:56 a.m.
Early influences on brain development and epigenetics
Stephen G. Matthews, University of Toronto, Canada
Oct. 22, 2015 · 10:49 a.m.
154 views
Q&A - Early influences on brain development and epigenetics
Stephen G. Matthews, University of Toronto, Canada
Oct. 22, 2015 · 11:29 a.m.
Building the physiology of thought
Rebecca Saxe, Massachusetts Institute of Technology, Cambridge, USA
Oct. 22, 2015 · 11:38 a.m.
226 views
Q&A - Building the physiology of thought
Rebecca Saxe, Massachusetts Institute of Technology, Cambridge, USA
Oct. 22, 2015 · 12:10 p.m.
Introduction to Session II - Cognitive Decline
Kathinka Evers
Oct. 22, 2015 · 2:02 p.m.
Brain health & brain diseases - future perspectives
Richard Frackowiak, CHUV University Hospital, Lausanne, Switzerland
Oct. 22, 2015 · 2:11 p.m.
120 views
Alzheimer's disease: genome-wide clues for novel therapies
Rudolph E. Tanzi, Massachusetts General Hospital, Charlestown, USA
Oct. 22, 2015 · 3:15 p.m.
Q&A - Alzheimer's disease: genome-wide clues for novel therapies
Rudolph E. Tanzi, Massachusetts General Hospital, Charlestown, USA
Oct. 22, 2015 · 3:59 p.m.
Immunometabolic regulators of age-related inflammation
Vishwa D. Dixit, Yale School of Medicine, New Haven, USA
Oct. 22, 2015 · 4:21 p.m.
159 views
Q&A - Immunometabolic regulators of age-related inflammation
Vishwa D. Dixit, Yale School of Medicine, New Haven, USA
Oct. 22, 2015 · 4:59 p.m.
Introduction to Session III - Nutrition & Cognitive Development
Pierre Magistretti, KAUST, Thuwal, Saudi Arabia and EPFL, Lausanne, Switzerland
Oct. 23, 2015 · 9 a.m.
Energy metabolism in long-term memory formation and enhancement
Cristina M. Alberini, The Center for Neural Science, New York University, USA
Oct. 23, 2015 · 9:16 a.m.
412 views
Q&A - Energy metabolism in long-term memory formation and enhancement
Cristina M. Alberini, The Center for Neural Science, New York University, USA
Oct. 23, 2015 · 9:53 a.m.
Building the costly human brain: implications for the evolution of slow childhood growth and the origins of diabetes
Christopher Kuzawa, Northwestern University, Evanston, USA
Oct. 23, 2015 · 10:29 a.m.
Nutrition, growth and the developing brain
Prof. Maureen Black, University of Maryland, School of Medicine, Baltimore, USA
Oct. 23, 2015 · 11:09 a.m.
152 views
Q&A - Nutrition, growth and the developing brain
Prof. Maureen Black, University of Maryland, School of Medicine, Baltimore, USA
Oct. 23, 2015 · 11:49 a.m.
Introduction to Session IV - Decline & Nutritional Intervention
Tamas Bartfai, The Scripps Research Institute, La Jolla, USA
Oct. 23, 2015 · 12:48 p.m.
176 views
On multi-domain approaches for prevention trials
Miia Kivipelto, Karolinska Institutet, Stockholm, Sweden
Oct. 23, 2015 · 1:04 p.m.
215 views
Q&A - On multi-domain approaches for prevention trials
Miia Kivipelto, MD, PhD, Karolinska Institutet
Oct. 23, 2015 · 1:39 p.m.
Methodological challenges in Alzheimer clinical development
Lon S. Schneider, Keck School of Medicine of USC, Los Angeles, USA
Oct. 23, 2015 · 1:49 p.m.
124 views
Q&A - Methodological challenges in Alzheimer clinical development
Lon S. Schneider, Keck School of Medicine of USC, Los Angeles, USA
Oct. 23, 2015 · 2:32 p.m.
We are what we remember: memory and age related memory disorders
Eric R. Kandel, Columbia University, New York, USA
Oct. 23, 2015 · 3:03 p.m.
228 views
Concluding Remarks
Stefan Catsicas, Chief Technology Officer, Nestlé SA
Oct. 23, 2015 · 3:50 p.m.
168 views

Recommended talks

Perspectives for nutrition and the gut microbiome
Nicholas Schork, J. Craig Venter Institute, La Jolla, USA
Oct. 24, 2014 · 3:02 p.m.
1297 views