Player is loading...
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
Wednesday, Jan. 25, 2023 · 12:32 p.m. · 07m 51s
Embed
Transcriptions
Note: this content has been automatically generated.
00:00:02
i think you re talk report for a project so first question on what we're uh
00:00:09
do you have any sites in terms of the zinc tangled generative factors that you get
00:00:15
uh our of your variable product that's question number one in new york broader perspective one
00:00:21
the rule of uh integrated prior knowledge for example pets with which
00:00:25
the what you call relations within so what so here's publicly for fear
00:00:31
actually we i can be designed in a very terrible in a in a a
00:00:39
in a way so that we can that even better fix between just everything so here we are
00:00:44
removing bitch effect well based on the origin but the what we're also doing for example these to be
00:00:50
using this kind of approach to in if you can look at the the ulcers and attach effect of
00:00:56
print it that's uh oh patients that are simple that we're not at that and samples that were treated
00:01:02
so you can see that it meant as a badge effect so in a way we're training them or not
00:01:06
uh that can know where and a statement of fact
00:01:11
and then for the patients or for the sample that that
00:01:14
we're not treat that we can predict how the treatment whoop looks like so i'm saying that it is very adaptable
00:01:21
and you can put here let's say one adversarial nectar that would be
00:01:26
for some reason discriminated another one you can train on on on the friend thing so it can be let's say
00:01:32
but funny or or or treatment effect or anything yeah it's
00:01:36
quite flexible that we can do with these variation often conversed
00:01:43
uh
00:01:49
in the late but it's such a this i example in the
00:01:55
lakers space with the old discriminate between was based on the tuner type
00:02:00
yeah so i but we're actually here using the the call the data because that's uh
00:02:08
that's what we i think that is that in later and uh in this is a very nice properties of the article that is actually that
00:02:14
we can actually come back later what happened with the project and so either the projection is tylenol cell line
00:02:20
to patient it can be and it'd be a patient to treat that patient so it can be very adaptable
00:02:36
so i'm curious to understand how him when we did not the drug we want want to be sure
00:02:41
that is focused on getting the only state right of two more development that when you look at sin nine
00:02:47
i imagine things which are crazy in a highly
00:02:51
pretty primitive man now where everybody pentagon press ways off
00:02:55
happy homes how do you identify do yeah yeah and it was like a
00:03:01
like the trajectory of the disease as you can yeah you know that the so
00:03:05
for example when we do the snow out screens we want identifying genes that there
00:03:12
i eleven for only particular cell types so if a cell line was sensitive to that g. then
00:03:18
it's not a good time yeah but the the cell line is relevant only for i don't know
00:03:22
some specific steptoe a subtype of the okay we our live on one day and it's a
00:03:27
target that we're interested in now depend on what kind of treatment we're developing if it is
00:03:33
let's say a delight and their base so this is uh in about this before the other recent development
00:03:38
where we are in for example believing affiliation to to
00:03:42
my stocks and uh so we they'll either allegation based
00:03:46
on cell service propane still we can aligned on to that binds to the surface uh all that to marcel
00:03:52
and then for example like a there is basically the typical that we're not attacking the healthy cells and then
00:03:58
we're looking for let's say propane that expressed on the
00:04:02
surface of cancer cells but they're not experienced enormous else
00:04:06
so that's very important criteria and for example even speak about eight alignment
00:04:11
there but it's not a very help a shell to sell it is
00:04:14
saying that they it what what education for example skin is not case sensitive
00:04:19
as the example the small in this time so if something is expressed in the small in this ties
00:04:24
in this sells itself then it's normal part yeah so i mean they are different the dodgers depending on
00:04:32
yeah oh what kind of therapy we're bursting for that specific that i
00:04:38
i was partially replied to your my concern was or about what you talk about drug to scroll ritual
00:04:46
looks like you use all your project is you perk certified so i started to watch more corpus but you should put that
00:04:54
um you is way to speed up the clinical trials to speed up your
00:04:58
your your your preclinical the to to understand if it's trucks work or not but
00:05:03
uh at user but you you you should look for for the purposes for example you're good good
00:05:09
result for production but you too much you know that there's a big endeavour first which would explain
00:05:16
and it was the outcome of that much running of to keep learning of of of the picture so
00:05:22
a white persian but yeah wrong so uh i just want to go to to to to to uh to
00:05:26
ask you a okay you have got revoke your your core of the production of the biggest users so you improve
00:05:32
to to to trust works more last resort to very good use but you
00:05:37
can you explain also interaction between the molecules right and and the reaction
00:05:42
to trucks back should huge so that you're going to those tires yeah
00:05:46
well for every experiment that we're on let's say we mind data and
00:05:51
then we did arrive with the least bit targets week after the target validation
00:05:55
so all all a slot we do well is the band we
00:05:58
cast on the stuff that gets me score them based on different properties
00:06:02
and then whether it is like promoting that passed this is or doing something else or we just want to categorise the
00:06:09
basis and then with a okay these are the jeez that are expressed in these these but not think al this i think
00:06:14
and then these are our perspective time it but then in every time is that we
00:06:19
did i've been we do experiments to validate in our example machine is the question of
00:06:25
columbia violet i because we we're not going to one and one hundred cell lines we would want to play maybe
00:06:31
at the beginning and then it's very important to pick the right cell lines
00:06:35
to why they these targets of the one that really resemble the patients the most
00:06:39
so that's wanting another think with this small but now we're using it
00:06:43
when we do the and stage clinical trials that typically and rolled not so many patients
00:06:48
maybe twenty thirty and that's never enough to do some sophisticated by market discovery and so on
00:06:54
and that's why we're combining the overall date is data and also preclinical data which we typically have a lot
00:07:00
and then we that's how we want to boost our by market discovery
00:07:04
and that's why but then we see the significant better factor that the by mark is that we seem to the clinical
00:07:10
models may not be relevant in patients so this made them as fast as it was just the it'll station in we yeah
00:07:17
we do way more experiment uh yeah but then you really
00:07:20
needed some some way in order to do things better not
00:07:24
for example in the pascal into that let's say we
00:07:27
derive list of file the clinical markets list of clinical
00:07:31
to be doing this section and that was it but now it's a weekend with way better every day i
00:07:43
yeah
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.
422 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
A story of unification: from Apache Spark to MLflow
Reynold Xin, Databricks
June 12, 2019 · 9:15 a.m.
1271 views