Player is loading...
5 minutest questions: Data Science for Precision Oncology
Prof. Olivier Michielin, Head of the Center of Precision Oncology, CHUV, and Group Leader at the Swiss Institute of Bioinformatics
Wednesday, Jan. 25, 2023 · 11:54 a.m. · 12m 43s
Embed
Transcriptions
Note: this content has been automatically generated.
00:00:02
in fact i'm sorry this was a great presentation i question respect to your um
00:00:09
you know the straw curve when you stratified by i. p. p. i. and you know these data
00:00:15
published on an h. by allowing and also the association response to something you'd be able to look at
00:00:21
that's a good that's a good question so um no one should can email us because
00:00:25
we were not able to get that status so it's you know it's the limitation of uh
00:00:31
uh you know the the c. r. f. that uh or have been collected by by the
00:00:36
industry and so if it's not in the star if it's a huge effort to get it up
00:00:40
and we were not able to benefit on with the aim is to do that but i think to your point that's the beauty of free will data
00:00:48
because you know if not you have a data warehouse where all the information is present then you'll
00:00:53
be able to to try and capture that uh so that's why i think it's very important that we
00:00:58
we all all these this possibility to do
00:01:01
data expiration within the hospitals because those a hypothesis
00:01:05
will come for the time and so if we are able to have a dynamic mechanism by which
00:01:11
we we can you know i had a a new a new data tied to the to
00:01:15
the data set i think it's it's really uh and uh it's something that will be transform active
00:01:21
uh huh
00:01:26
so okay yeah thank you thank you very much for great could use um
00:01:34
it's great to redirect read after that
00:01:40
all i want there if you need on you for where you improve patient
00:01:48
that you can share typically without god have cars and
00:01:51
also i don't know whether this challenge lee or barriers
00:01:57
for ah for ah papers showing you ensure they huh what is the most big ears
00:02:06
most gifted betters okay thanks so much it's a very it's very important point so i think
00:02:11
you know all the the host the the second question first all the hospital that we're dealing with our
00:02:15
our university hospital with large data warehousing now we
00:02:18
had a very sullied prescription since then on on ha
00:02:22
to explore the data so all of these has been norm then and we can really tell
00:02:26
that the the data warehouse people what to do that to do this i think to not embark
00:02:31
a new hospital visited our whole system would be really relatively easy
00:02:35
now the problem is if you have structure where there isn't a data warehouse and
00:02:39
and and we're working with peripheral hospitals to find ways to export no sort of uh
00:02:46
generate export form for a patient case that can be processed through in the l. p. n. yet
00:02:51
entered into the system so they they might be ways like that's to to to have and of course
00:02:57
the the the more uh you know the the t. v. you usual ways to be cool but
00:03:02
with the c. r. f. and so you know the data manager that enters the that but that's
00:03:06
no extremely expensive and time consuming but nevertheless i mean
00:03:09
that's hot a psychic is operating they're moving towards the the
00:03:12
data warehouse system as well so hopefully s. a. k. k. will be will be capable of of providing it's not
00:03:18
bad for the constant it's clear that um you know we we yeah we need to have patients that have field
00:03:25
several forms i mean first of all i general consent because
00:03:28
we wanna be able then to reuse the data upon approval
00:03:31
but it is committee for for the new study and also specific consent and uh because for example for for
00:03:37
these these program we actually need to do is is second biopsy and so for this one we need clearly
00:03:44
a a consent form which is specific for for the section and of course then with the speech and uh
00:03:51
all the the security the data transfer in use agreements
00:03:55
between the various hospitals have been working all details so we
00:03:58
had to put the lawyers of all the hospitals uh and you know what the it takes but it's done
00:04:03
it because just three years but now it's done and we are we are able to uh to to move forward
00:04:11
i think so it's i don't your purse describing how to work to to match here um extracts are
00:04:17
features from patients and to do the jury trials to personalise future treatments and that that that's really exciting
00:04:22
and and my problem with workman's comp observation data is that you can get a correlations radar it on
00:04:27
not cause and and that can be a problem when trying to to burst master p. based on that
00:04:32
but is this as a surrogate tomatoes approaches to to try
00:04:35
to discriminate what's correlating what's quality so would it be um could
00:04:40
begin vision to to add this in to ensure pipeline so
00:04:43
that over time is detection rates and you get more more features
00:04:46
uh you know your your your system would would get more more to close to two closer to
00:04:51
uh features that absolutely this is something we're discussing very frequently i mean we we we wanna be
00:04:58
uh you know talking with score show uh association and and uh just uh you know uh
00:05:04
a random association so we really need uh to be able to implement that then uh
00:05:08
you know that's something which really easy in in our little road map for the future because
00:05:13
of course you you wanna be able to filter out and and we're following in their uh a
00:05:17
lot of exciting uh you know approaches to be able to do that no one approach if you
00:05:21
do if you do the biology too perturbed the system but of course he was kinda leader you
00:05:26
cannot do that's why thing not those new technology will be extremely a useful uh for this absolutely
00:05:34
yeah exactly yeah yeah that's my question was related to this question because
00:05:38
alright you you you have your data warehouse now collected data from the
00:05:41
current treatment but you will have more more more interviews treatment industry more literate
00:05:47
be more new um technology coming so you have to completely for your model
00:05:53
for training new data so i don't know if you're already facing this challenge
00:05:57
and the second question that now you're working one exception so but the ones that are not responding to the
00:06:03
to basic to repeat it to come to your automatics you think future would be applicable to everyone
00:06:10
yeah so it's it's um you know it's a good question for the the my kitchen already takes late stage patients and uh for a
00:06:17
chuckle reason because we don't wanna put this in front of well validated
00:06:20
standard of cap that's been diskette if you find no the molecular signature
00:06:26
where the patient or do we extremely well on on the
00:06:29
site specific treatment intervention you wanna move that sooner and so
00:06:34
the the way to do that is actually to use the exact same network to the stream of automate the to spot
00:06:40
patients winner of course then you will have to deal the prospective trial always proper a consent and all that
00:06:46
but then i think you know maybe doing a very small face to uh you might be able to really
00:06:52
uh it gets uh your your your answer uh whether yes or no it it it works
00:06:56
the for the first question and okay i think um you know it's an evolving model i mean we we had new treatment coming
00:07:02
in all the time and so the inspection system is below the
00:07:06
uh with this in mind so we we know any any new
00:07:09
a a treatment that comes in will be automatic integrated and taking
00:07:13
care of where of course we cannot see anything is uh use on
00:07:17
experimental treatment uh that have uh not chip didn't dawning practised
00:07:21
in the hospitals uh so but that's something we we can
00:07:24
actually provide a basis for because for example we can work
00:07:27
on a synthetic control arm so if you have a new treatment
00:07:31
exactly so we have worked on that for for t. l. therapies for example senator up it's
00:07:36
uh we wanted to know 'cause we have use that for for uh you know achieve in
00:07:40
in second line off the failure of the p. d. ward and it it was sixty seven
00:07:44
the actual you and you know we pay people were doing reasonably well but but doesn't that
00:07:49
mean and so we we build a synthetic controlled from the very same did i showed you
00:07:54
in order to really shocking what what's the fate of the patient after failure of of the people
00:08:00
and we saw that this was the small ones and we were actually
00:08:03
able to really didn't know i had the hypotheses that yes still therapy does
00:08:08
happen this setting because we had the reference uh for me people but you
00:08:12
know it's now five years down the road that everyone will get killed therapy
00:08:17
and so we'll have to really know they'd asset in the no we will be for the for the for the rest so i think
00:08:21
canada i mean those system they need to be able to to integrate the new data type on the on the the snap
00:08:27
of the finger so that we can redeem uh you know and maybe react even have the that is meaningful for the future
00:08:36
oh
00:08:39
uh_huh
00:08:46
uh_huh uh_huh so yeah you mentioned that you were are you want to put each patient in the loop to do
00:08:53
more reinforcements type learning and just sort of suggests you know
00:08:58
for a fact big of a computer during out which s.
00:09:02
actually like game and fall until i until the end user and something or if you know
00:09:08
it gets lost or not is you know you want to do something zoom revere sense that which stock
00:09:16
things really selector patients sort you mention that you we selected some pages standard
00:09:25
so i'd say it's a very good question of course we we don't wanna pay game with with patients and uh and you know we
00:09:32
you know any opportunity that we see if it uh for the patient lecture nate and and
00:09:37
we we have to we treat the patient uh you know that's our first and only mission
00:09:41
now of course if um you know you record this information and it's gonna be below or
00:09:47
by a lot of things because for example you know you assigned a patient to didn't treatment
00:09:51
for somebody doesn't responded that he get another treatment coming from somewhere else that works and that's gonna inference survival
00:09:57
so you know you can do many things you can first look at progression free survival and
00:10:01
that's a viable because that's gonna be in metric of back line that you have been should
00:10:05
do so that's a way to really know a look at the the more a specific uh
00:10:11
the performance criterion for for for your line and but i'm not clearly i think we we
00:10:17
yeah you'll have to use also hear the statistics uh and and to be
00:10:21
able to say that in general when those patients were treated with this line their
00:10:25
fate word was good or not as good and i think that's that's what we
00:10:29
want him or not at this level but the budget the questions that they can
00:10:34
yeah
00:10:36
i i last question is men's to make the transition with next stocks now listening to your talk
00:10:43
was very much we want us to that it's meant fog clinic to improve that can we cannot come
00:10:49
and then we have lots of companies so i wanted to understand how this platform and the date and you're connecting
00:10:55
can indeed as weren't used tried it companies to promote the development and you
00:11:00
try and whether it's meant to be are now perhaps how can we overcome them
00:11:06
very good very good question i think you know it's a very interesting uh a program to work with companies
00:11:13
and before that just a very small note it's also
00:11:16
very interesting program for the academics because no with this type
00:11:20
of precision oncology i think this puts back the academics into
00:11:25
the driving seat which is very nice because you know we're
00:11:28
using drugs that are already proved that uh and so we were basically
00:11:32
just chanting the patient to the right treatment arm combination normal sequence on
00:11:37
but i think it's important because that's the code academic process not what is nice is that these data sent
00:11:42
in in this process there are ideal to collaborate with
00:11:45
the industry going for example the industry could not had no
00:11:49
makes to to to do to the game ends studies or maybe add the court in their war came up with
00:11:55
us to to provide is more face to prospective trial
00:12:00
for example if we identify in the data set that eh
00:12:03
you know you need p. d. one in like three
00:12:05
in that type of signature of the email said infiltrate whatever
00:12:10
then you know a company could sponsor h. y. always be one or p. d. one like three
00:12:15
uh in this patient population and we can use this same
00:12:19
network suisse wide to run those prospective trials because all he he's
00:12:23
in place the the to do that so i think the this
00:12:26
problem is really meant to be a you know interacting with the
00:12:30
uh uh the formal sector and s. p. h. and has this uh within
00:12:34
the tubes in it's the it's made so that we can actually do these
Conference Program
The Idiap Research Institute in Martigny is launching a new public series of symposiums
Lonneke, Idiap Research Institute
Jan. 25, 2023 · 9:39 a.m.
420 views
Interpretable artificial intelligence for cancer personalized medicine
Dr. Maria Rodriguez Martinez, Group Leader, IBM
Jan. 25, 2023 · 9:48 a.m.
5 minutes questions: Interpretable artificial intelligence for cancer personalized medicine
Dr. Maria Rodriguez Martinez, Group Leader, IBM
Jan. 25, 2023 · 10:10 a.m.
NIPMAP: Niche Phenotype Mapping of Multiplex Histology Data by Community Ecology.
Dr. Jean Hausser, Assistant Professor, Karolinska Institute, Sweden
Jan. 25, 2023 · 10:21 a.m.
133 views
