Player is loading...
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 · 11:24 a.m. · 28m 24s
Embed
Transcriptions
Note: this content has been automatically generated.
00:00:02
you feel the synergies them in the common interest in in the past science uh
00:00:07
and it's really fascinating for for us to see this and i would like to show you some example on what we're
00:00:13
trying to achieve by using data science in oncology uh and and show you a you know a little bit of the uh
00:00:20
do work that uh we we hope could help us guide
00:00:23
treatment decision uh even the not so distant future and i thought
00:00:28
hey with with it so but there were to call curve
00:00:31
of survival in patients that for example have made aesthetic uh melanoma
00:00:36
uh and we now have um behaviour of this type where we have a small fraction of the
00:00:40
patient you see here about twenty person that or long survivors so these bar can be higher or lower
00:00:47
but the very important aspect is that there is a bar so there is no plot or we can have this long term benefit
00:00:52
with the minute therapy and of course the treatment that you deliver
00:00:56
and only working twenty percent is far from ideal there is toxicity
00:01:00
the other side effect and so on and so forth and you
00:01:03
lose time if you don't have the correct assignment the correct nation
00:01:06
to to drug and so the question is can we find via
00:01:10
markers digital buyer markers that help us pinpoint those patients can we
00:01:16
identified using data science the orange patient in this population and can we restrict the
00:01:21
treatment duties and of course that's the ideal scenario because then we are able to provide
00:01:26
a drug that will have maximal benefit in this uh population grant that we can uh
00:01:32
identified those patients of course the question is what do we do for the rest well
00:01:35
we can find buyer markers for the other uh and uh members
00:01:40
of of the of the population and so on and so forth until
00:01:43
we try to reach a you know as far as we can maybe eighty eighty ninety percent of the population so that's exactly the goal
00:01:49
and to achieve these uh uh the important uh aspect is to to find those those maya marcus was predicted by
00:01:56
the markers that will tell you which patient will respond to to what and so you know to really move this forward
00:02:02
uh we have started to to do very deep finally takes of the chamber and the trimmer micro environment i mean
00:02:07
this is a huge effort uh internationally trying to uh uh
00:02:11
assets what are the key players in ten or said or
00:02:14
a a composition a ex uh the level of activation of the set i mean we have had fantastic talks on this
00:02:20
so just just just a minute ago and of course as we're talking here about huge data sets a data science is becoming
00:02:26
not an option but an absolute need to uh to be able to make sense out of the data
00:02:32
so there is the second revolution that i would like to
00:02:35
just a mention here is the way we generate a uh um
00:02:40
knowledge and and evidence for treatment so usually in uh when we have a
00:02:45
heterogeneous patient population like the one i had depicted here was the various colours
00:02:49
what we do is we try to minimise the
00:02:52
variability of the population by putting intrusions an exclusion criteria
00:02:58
and in that context uh we can run prospective randomised trials that are extremely efficient
00:03:04
at determining whether an experimental treatment is better yes on the wood and the the reference
00:03:09
and of course there we can bring back the information of the conclusion in that case that the experimental treatment is better to
00:03:16
use well selected population so where's the cats well the catch is
00:03:20
that we usually don't do that because once we have those conclusion
00:03:24
we apply the results to the general population and that's a big issue in colour g. because if you do
00:03:30
the math the number of patients are treated outside of
00:03:33
the street inclusion exclusion criteria of the trial is huge
00:03:37
and that means we we actually are losing a lot of information and
00:03:41
just like wracked italian example uh and my my feeling on crutches melanoma
00:03:46
and we had no it w. minute they're deeply marvin the volume up see chili
00:03:50
for in in you when you mention it in all the trials the bring that
00:03:54
that's that the patient or excluded it because like three or four years to start
00:03:59
to see that oh yes there is activity in the brain to find a have
00:04:03
prospective trials that show exactly exactly the same so there is the same response rate
00:04:08
in the brain then in the rest of the body something we did not imagine
00:04:12
at all at the beginning but those three four five years gap and uh you
00:04:17
know he's really something we would like to avoid in the future so can we actually
00:04:21
uh uh no speedup uh the momentum and also have access to a
00:04:25
the largest the asset that by tapping directly into the real world data
00:04:30
so that means we're no longer relying on on um the prospective trials
00:04:34
that are not or what selected will control i mean high quality data
00:04:37
but can we use it real well data which might be of lower quality because they they would not done
00:04:43
per protocol however it's a much larger number so maybe we can use data science to filter out the station
00:04:50
make cohorts that really makes sense for the question that that we are asking
00:04:53
so that's exactly where we would like to go uh in the very near future
00:04:58
so well the data types uh that we can uh it's force you to assemble for a precision
00:05:04
oncology program uh we have a chemical data uh actually example how we can we can manage these
00:05:10
which reported dot com are becoming a hot topic in in that uh feel that
00:05:15
and of course uh all the gentle makes uh just the transcript to make so that
00:05:19
we have seen in on the images that are becoming a integrated part of the data set
00:05:24
uh that we are ignore happily as starting to mine too fantastic a collaboration on a
00:05:29
lot of the the colleagues are here in in the room that would like to to really
00:05:33
acknowledge these fantastic a collective work so not once we have this whole data set it's
00:05:39
very difficult to bring that to decision because we need a large reference a data sets to
00:05:45
be able to compare what is normal what is expected what is differing from from the
00:05:49
norm and i'm getting is then to bring that to eight molecular to more which is the
00:05:55
uh action arm of precision oncology program where we can make a decision and educated decision on
00:06:01
how to treat the patient knowing all of the information that we have on the left so we're
00:06:06
working towards that not all the components are getting place but that's
00:06:10
something that we aim to achieve in in the in the coming years
00:06:14
i think to really make this a virtuous cycle that we really need to close the loop and and
00:06:19
to do that we have to bring the information of the patient that we treated uh with this knowledge
00:06:25
into the the assets so if we can do that that we can
00:06:29
really go to reinforce learning where we can when we are able to actually
00:06:33
add information as we progress through the number of patients that
00:06:36
we created this mechanism not show you as an example of that
00:06:41
so just um uh as a quick a teaser we all know the the
00:06:46
big groups of a machine learning unsupervised improvise or reinforcement learning uh you have all
00:06:52
seen this but many time i'm sure yeah but there is the analogy in in oncology
00:06:57
uh which which uh you know it's not a one to one mapping
00:07:00
it but you can actually go to new technologies for the unsupervised prognosis for
00:07:06
the supervisor you would then use the outcome as a way to to
00:07:09
supervisor training and for treatment no reinforcement learning as i just mentioned could be
00:07:15
a really the way to go because then you can learn from from each patient so there isn't i mean this is really a way to schematic
00:07:21
but nevertheless i think they are really not the technologies in terms of
00:07:24
machine learning to really be able to help in different aspects of personal coach
00:07:30
so as we have mentioned uh data is really important the the size of the data and and the reference that
00:07:36
and we have embarked uh five years ago into a national problem uh led by this
00:07:41
use per slice has network and we we
00:07:44
reply to core uh for for domain experts getting
00:07:48
together and we propose this she's a purse my psychology uh and this program basically was
00:07:55
the idea to put the data together in the entire country uh we we wanted to really
00:08:01
be able to uh bring all the university hospital but we actually
00:08:05
were even more ambitious we wanted to also have been known university hospital
00:08:09
in oncology turns out that we had d. s. a. k. k. which is a very nice partner yeah because they are linked to all the
00:08:16
non university hospitals and so we build a day passes them uh uh
00:08:21
that is completely coherent between the university hospital and the non university hospitals
00:08:26
for the university hospital we have a way to speak the same language uh at the data warehouse level
00:08:33
and for the non university hospitals we have uh it's your f.
00:08:36
b. is the web based forms uh that can be field even
00:08:40
uh so that we actually encoded data the same way at the end
00:08:43
of the day being able to recombine everything into a the same who'd
00:08:47
and you you have seen that i have highlighted here the lemon cut because i think we can really do great things
00:08:53
together from from the valet all the way to geneva i think there is a a nice critical mass if we can gather
00:08:59
all these data together and they're really strong efforts to to do that so when we started that project this was really
00:09:06
a a a heritage task because we did first the
00:09:09
mapping of the infrastructure at the various hospitals re release
00:09:13
that basically it's all different and there is just no way to reconcile the data at the at the lower level
00:09:19
however we wanted to be able to use them uh for joint project so the question is how do you feel this gap on
00:09:26
and the strategy that we had used was to a project the data of each hospital into
00:09:33
they come on a vocabulary did asset and that's exactly what we have done so we have defined
00:09:39
and in forty plus uh variables some of these are
00:09:42
extremely complex of variables recurring and so on so it's
00:09:45
actually quite a large status that we used to call it a minimal data set but it's no longer minimal that
00:09:51
it s. that can be more or less automatically extracted
00:09:55
from the data warehouse of of some hospitals but some
00:09:58
are more advanced some require more work but we'll we'll get there we have not three more years of funding it
00:10:04
and we believe that uh it within those three years we should be able to get close to automatic station
00:10:10
e. most hospitals um we also have ways to cry
00:10:14
every uh more advanced data uh directly from essentially but
00:10:17
that's or less less important so that i can be a structured field uh uh i see you three cold
00:10:24
response to that treatment is on or how do you call the treatment the same way everywhere in the country
00:10:29
that can be also text data and here is an exercise that we have done on
00:10:34
the radiology reports that we had gathered from the various
00:10:37
hospitals and and we have trained uh and a and an
00:10:40
l. p. natural language processing to to uh i signed
00:10:44
the patient to two categories all three to respond or non
00:10:48
respond or or so on and so forth progress or whatever so we we have tried many things and to cut
00:10:53
a long story short uh this very rudimentary n. l. p. it's it's less rudimentary now we we we have improved
00:10:59
but they're just out of the box we we get the eighty ninety percent accuracy and it compares you know
00:11:05
it's basically the same as what the the viability we get when we take several commissions reading the report so
00:11:11
basically means that this is extremely useful because then we can assign all good patient is progressing uh at
00:11:18
this time because we we have the the report that says that so why are we interested in today's well
00:11:24
uh first uh applications so that's just actually the clinical data uh what you see here on
00:11:29
the left is the prospective trial run by b. m. as near mess on on on the walk
00:11:35
it's a w. not directly on the top a single agent
00:11:38
p. d. one blue curve and then single agent at u. p.
00:11:41
maps actually from the bottom and you have over us rubber on the top and progression free survival in the bottom so that's
00:11:48
basically what we used to guide our treatment issues decision in in the the minimum
00:11:53
much more bobby we always referring to these two to make decision so the question is
00:11:57
now if we take the data of the country or a hospital what
00:12:02
do we see and you can see that uh although we have small numbers
00:12:06
we do see fairly senior id hate your boss in
00:12:09
terms of overall survival as well as in terms of
00:12:13
a a progression free survival insight that's where where we needed d. n. l. p.
00:12:17
to call it the patient a a progress are based on the radiology report that allowed
00:12:22
us to to create the curve on on on on the lower court so this is
00:12:25
very interesting because if we now realise that or davis it internally or a reproducing known
00:12:32
a fax from from the the the literature possible i mean this this uh you know is is
00:12:37
of course pretty reassuring but it's also way to ask him any additional
00:12:41
questions for example you could create trials that will never exist uh so you
00:12:46
could imagine really taking it uh you know comparison that are on the
00:12:50
tickle to do or that are more suited for a specific uh uh situation
00:12:55
and you can imagine that with these chicken stock uh you know putting the a lot a.
00:12:59
i. system on top of it and start learning uh uh where there is opportunity for for
00:13:04
a improvement for the patient that's so that's the gringo that's where we go in it so
00:13:08
it's a lot of work to get there but i think it's it's a really great opportunity so
00:13:13
in particular what have we learned from that exercise so we we went
00:13:16
back to his data and since the data is reproducing and know when
00:13:20
um again outcome for the patient we looked at additional metrics
00:13:26
that could be infringing this in one of the metric was the
00:13:29
uh uh um medication that the patient is taking at the baseline when we start the
00:13:34
uh i'm a therapy and you can see that uh here
00:13:38
we have the stratification for guys that's proton pump innovators it's
00:13:42
uh what you take when you had a i see a stock market in aid kit and actually it turns out that if
00:13:49
you do take your lines you are doing much worse than if you don't uh so do of course it's always the case
00:13:55
when when he does this results from that assigns we scratch our hands and then after five seconds to say oh of course
00:14:01
the the p. p. eyes are changing the th of you've got a and we know that the got floor
00:14:06
of these strongly associated with the chance of responding or not yes or no to the to the inner therapy
00:14:12
so uh this was a a hint i mean this is not validated yet up
00:14:16
but it's it's interesting hints because you know it's coming from the data it's hypothesis generating and that's something that you
00:14:22
know could change packed is if we can show that was probably not a good idea to start a new carpet
00:14:26
on a point of pop any beecher so interesting i was telling about the this dichotomy between real well they don
00:14:33
and and prospective trial so with this really well they that we went back to be a mess with the the
00:14:37
the trials and we asked them to mind their own data set to see whether uh we can find these uh
00:14:44
a pattern yes and also not in the prospective trials from the m. s. you see here check mail sixty nine
00:14:50
you can see that they realise here it's recognition so the same direction so if
00:14:54
you use the t. i. uh you actually have a progression free survival that appears to
00:14:59
be lower than if you're not so the story is not settle the it's not
00:15:03
the final word because we don't see that in other trials like checkmate of sixty seven
00:15:08
our impression that uh is that um i there is the decomposition of the patient uh
00:15:13
plays a very big role uh so we we really need not to to dissect that and
00:15:18
so i mean with the the evidence that we have all it's it's it's to me
00:15:21
to recommend not to take p. p. eyes when you doing notoriety but even as i think
00:15:25
this this is a nice illustration on how we can use real data to ten challenge
00:15:30
perspective and then to perhaps go to a change uh of get just at the data level
00:15:37
so now they're of course uh a great opportunities using a additional data sets and one of them is
00:15:43
is uh uh images uh and i would like you did you just to show you an example of a
00:15:48
work we have started to do in the field of digital
00:15:51
pathology and just about allergies are interesting feel because it's uh
00:15:56
and image analysis apply to the court of of the trimmer i mean
00:15:59
we have seen great example angry talks uh just just a minute ago
00:16:03
we have a you know nice a collaboration with the pathologies low on the light
00:16:07
in a lot on uh and uh plus a whole job hunting geneva ingenue that entire
00:16:11
the uh department is digital eyes that so it means we
00:16:15
we uh we have access to july slide i think this scanning
00:16:18
like seventeen thousand a sly and uh every week so it's
00:16:21
it's a huge data set that is being generated the question is
00:16:25
what how can we explain that so what you have here on the left is um a a any chain it's the standard
00:16:32
uh colouring and uh which uh was then used to to get
00:16:36
infinity picture characterisation of the cells and then we can use it
00:16:40
uh in a network uh abstraction and and we had docking at that time who uh who is here uh with whom we have a
00:16:47
fantastic collaboration today that was barking pascal pasta and and michelle quite
00:16:51
there who is also in the which been uh you know very uh
00:16:55
a proactive in this problem so that's ongoing but it's it's a very
00:16:58
interesting collaboration uh uh between the clinicians and and then e. p. f. l.
00:17:02
on on this from that and and really a great problem that we're pushing
00:17:06
together electrician example really uh on on how this uh has been used for
00:17:12
a digital pathology with the the work of a into general chit uh who is really
00:17:16
in a world expertise in analysing a chaney uh slides and he came up with a
00:17:22
it uh a deeper learning a way to correct
00:17:26
a rise the lymphocytes into a uh in a t.
00:17:30
shirt and you can see perhaps uh that the belief that are highlighted here with the the the little
00:17:35
blue uh a shallow and that's that's where the
00:17:38
the uh the planning algorithm has identified leave aside
00:17:42
and then we can apply matrix of uh the leave aside distribution
00:17:46
uh are they really in future team uh they are you close to
00:17:50
together do we have clustered big cluster small cluster so so we're
00:17:54
analysing tons of parameters but i just to show you an example because
00:17:58
i think it's it's quite a telling a it's a it's a
00:18:02
cohort of melanoma that we are analysing here a hundred and twenty melanoma
00:18:07
and on the left you see the density of three percent which
00:18:11
doesn't seem to be struck to find the patient with respect or survival
00:18:15
not if we start to look at the structure of that infiltrate
00:18:19
and and the measure which is the future hundred and seventy eight
00:18:22
uh that feature is the measure of on heightened feature achieve
00:18:26
uh these leave aside arc that seems to be correlated that
00:18:30
without come so that shows that probably these huge the assets of digital but energy that we have in our hard drive
00:18:37
could really bring a lot of information on how to guide directly for for the patients so that's exactly what we're
00:18:42
trying to push again collectively with with all the the menu card in the on because now they are really nice initiatives
00:18:49
to further rate additional pathology within the whole country uh and that's led by the talkers laugh from from zurich it
00:18:55
and and that could really beaches formatted because if we have now ways to share data
00:19:01
at the national level uh i think this would be really uh uh anyway fantastic uh aborted
00:19:08
now i like to to switch to a radio makes uh in here uh really show great
00:19:13
work from the danielle have blair who is also a in the room together with with initial concept
00:19:18
uh then has really a pioneer this uh together with
00:19:21
the office of the person who is here in integrate collaboration
00:19:25
was also john prior uh the idea of of trying
00:19:28
to um use more of the information um of the uh
00:19:33
fine structure of the uh pet c. t. or um
00:19:37
um c. t. scans in order to be able to uh
00:19:40
a guy precision ecology so the id first is to select the regions of the tomb or uh do doing these
00:19:47
considering the the segmentation and then once we have these small volumes of trimmer how can we use
00:19:53
uh the last science to extract information and that's the whole field of afraid i'm mixing killer
00:19:58
uh i had a percentage is is really leader the world one in in this heat so
00:20:02
we we um you know how these these great collaboration together with with with that gene now
00:20:07
you can extract a lot of features and then try to associate those features a two and
00:20:13
he yelled come and that's exactly the work that the uh then yell and colleagues have have taken
00:20:18
up with meeting shall and and i'd hear that trying to look at some of those properties
00:20:23
this is ongoing work uh not published a so you d. need validation which is on going to
00:20:28
with zurich where they had similar data set up and how would
00:20:32
we we are you finding one of the feature that appears to be
00:20:36
uh oh again associated uh with benefit you see here that if
00:20:40
you have a feature low or high uh your faith again on
00:20:45
p. d. one switch the predictive by the marker appears to be
00:20:48
very different so you can see this is very short example that's using
00:20:52
digital pathology core review makes we could build a a predicted by
00:20:58
the marker can be the it's four p. d. one if you
00:21:00
could see the melanoma so again i mean and let us know saying that any of this is really for the king but it just
00:21:06
it shows that probably the release information there that can be mine in that will in the future most likely
00:21:12
uh uh integrate be integrated into more what's in order
00:21:15
to really guide a precision treatment assignment for the patients
00:21:20
so how do you do that and that you have all those great buyer markers and
00:21:24
and and you wanna treat patients in for the last but i would like to show you
00:21:28
example on how we were doing that yeah so we have created a molecular to more bored than a
00:21:33
donkey between geneva and was on the uh and that's
00:21:37
two more bob basically uh comes from the commission so
00:21:40
that clinicians has to identify a and a patient
00:21:44
that is suitable when usually those special operation that had
00:21:47
failed to the standard of care we don't want to be you know the doing this this and advance the
00:21:53
uh technologies corporation where there is the standard of care treatment but once that's
00:21:57
uh as fail of course there is opportunity to use all this information and basically we we
00:22:02
get the material and then assemble it a fantastic keen on both can sting even less on
00:22:08
where uh we can have the last ninety is the but oh jeez molecular pathology is biology since one and so forth
00:22:15
think about the problem started to use more and more data science in
00:22:19
the mix and then bring back any information to the to the treating
00:22:23
doctor is very interesting because i'm part of the stream of all but sometimes i'm also the treating doctor and then i received the results and
00:22:30
is very interesting to see how these going back to your decisions it's
00:22:33
it's quite a a in in training and and no change of of mine
00:22:38
so we started that into doesn't seventeen missus mean if it
00:22:41
incredible success um we we were thinking to get probably of four
00:22:46
of the if not the force of the volumes that that we we got that so that's network as being open of course
00:22:52
to uh that both used to hospital but also to the rest of the of the um come to one and all the
00:22:58
french speaking park and you can see that we have no more than two thousand five hundred patient we're reaching out to three thousand
00:23:03
and it's a big number because uh we get about four hundred cases per your
00:23:08
and by comparison to him badly as he is about two hundred fifty cases so
00:23:12
i think it's a it's a very decent volume and it does show that if
00:23:15
yes we bring everyone together in in this region uh we can have the numbers that that
00:23:21
uh no means something in terms of that data sense because he would we're treating every battle ready
00:23:27
a little bias from milan but uh other than that i think it's it's her present and
00:23:31
and the to just show you that this is not only you know uh you know powerpoint a
00:23:37
science but there is a patient being treated that i'm an example here of a young patient despite business
00:23:43
ambition had a a in an advanced or coma which uh it was not responding to chemotherapy so he
00:23:49
was under a intense chemotherapy progressing very fast probably
00:23:53
had no fume ones uh in front of him only
00:23:56
uh and then the top button as who actually is not the university hospital he he's
00:24:01
a private practitioner in in those on uh it was sent us the the tissue and
00:24:06
uh of course with all the consent on the on the the right way uh we
00:24:10
did the analysis and we know that this are common does not respond to the minute there
00:24:15
uh but when we did the analysis we we found an amplification of a c. d. to seventy four
00:24:22
and see the two seventy four is actually p. d. l. one which is a gene that uh expressed
00:24:28
only to herself will basically block t. cells from doing
00:24:32
their killing activity as we have discussed in the previous talks
00:24:35
is very interesting because um that's a trimmer that usually is not seen by the immune system
00:24:41
here needed this escape mechanism to grow and so this is interesting because it it
00:24:47
the goes back to the idea of the reference if we didn't have the reference that
00:24:51
says it's not usual uh to have p. d. l. one implication is this to
00:24:57
her we probably we would have nots but to this as a as a therapeutic opportunity
00:25:01
but since we had this we were able to say well maybe that trimmer which in general is not you know jenny
00:25:07
this specific one is and that's why we we know decided
00:25:11
to treat a patient with it p. d. one blocking antibody
00:25:15
and you can see the vision when in in your complete remission uh i think that lasted the at
00:25:19
least in complete remission i will have to to talk with yeah bonus to know the the latest news
00:25:25
but this patient was progressing very fast to has been has had a
00:25:28
complete change of trajectory at things to this intervention so it really shows that
00:25:32
using a you know large approaches uh all makes and
00:25:36
and very soon data science probably will change the fate
00:25:39
of of many patients so how can we push these further sought to finish about like to just show you the
00:25:46
a person action quality second round so now we we have a really applied for speech and and and uh
00:25:52
we we've found that for a a new project which will continue the data acquisition of course in in consolidate this
00:25:58
but at the same time the move to prospective treatment of patients using multi all makes
00:26:04
prior to the molecular to more so what you see here that's the first panel we
00:26:08
do exactly what i showed you with this temptation which is to n. g. s. m.
00:26:11
uh and then we we make a a treatment the location for the
00:26:14
patient but in parallel those that's gonna be three hundred patient just as
00:26:18
a control not random eyes it's just a a different court and then
00:26:22
we'll get fresh tissue material from this patient and first we'll have the rapid
00:26:28
uh evaluation we started it'll make so that's uh all mixed with no on else uh where
00:26:33
we know more less the action ability and and we know exactly what to to look for
00:26:38
and that will ball back in clinical time uh to do molecular to mobile so
00:26:43
that's a huge challenge we're we're implementing this in the whole country at the same time
00:26:48
to be able to make sure that we get these um all to mix analytic
00:26:52
something done in real time so that we can make the decision to three four weeks
00:26:56
later at the molecular tomorrow so this is taking shape as as we speak we hope
00:27:00
to have the first patient even in about one two months and and then we had
00:27:05
a more extended will make some and where we will go for full blown thing
00:27:10
it's already sick guessing is a t. c. r. herbal korean is and special transcript that
00:27:14
makes uh that that uh will be really very important for the future and then we'll
00:27:18
assemble all these data for our research project which is really geared at trying to understand
00:27:24
why certain to more do respond to minute darby and others not uh in intertwining tour a
00:27:29
trimmer type compare isn't it or something with this other uh finish up and really try to
00:27:36
re emphasise that uh i think we we have been able through a combination unit
00:27:41
are happy to really bring the bar and i think there is really no longer paternity
00:27:45
of course to introduce new treatment which of course we we all do research on but
00:27:49
also to better select patient for therapy i think if we are able to do these
00:27:54
there is a chance that we can significantly um um raise the bar because we lose a lot
00:27:59
of patients when they are not assigned to the correct line because they will get more rapid course
00:28:03
of disease apprentices is and then we were not able to to salvage them anymore so i think
00:28:07
if we are able to to do this and with the use of of a i ended a science
00:28:12
i think we'll we'll move the field for junkie together and there are many people at the hospitals
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.
421 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
5 minutes questions: NIPMAP: Niche Phenotype Mapping of Multiplex Histology Data by Community Ecology.
Dr. Jean Hausser, Assistant Professor, Karolinska Institute, Sweden
Jan. 25, 2023 · 10:43 a.m.
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
Jan. 25, 2023 · 11:24 a.m.
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
Jan. 25, 2023 · 11:54 a.m.
Applications of AI in oncology drug discovery
Dr. Slavica Dimitrieva, Associate Director & Senior Principal Scientist, Oncology Data Science, Novartis Institutes for BioMedical Research
Jan. 25, 2023 · 12:07 p.m.
5 minutes questions: Applications of AI in oncology drug discovery
Dr. Slavica Dimitrieva, Associate Director & Senior Principal Scientist, Oncology Data Science, Novartis Institutes for BioMedical Research
Jan. 25, 2023 · 12:32 p.m.
Recommended talks
Digital detection of drug adverse events in rheumatology - a whole population analysis
Paul Hasler
Feb. 1, 2019 · 2:57 p.m.
1145 views
