Player is loading...
5 minutes questions: NIPMAP: Niche Phenotype Mapping of Multiplex Histology Data by Community Ecology.
Dr. Jean Hausser, Assistant Professor, Karolinska Institute, Sweden
Wednesday, Jan. 25, 2023 · 10:43 a.m. · 09m 20s
Embed
Transcriptions
Note: this content has been automatically generated.
00:00:01
mm kay
00:00:10
i think it's worth very nice full some um just the question
00:00:15
oh maybe i missed it how are how sure are you that you are looking at cancer cells and what other
00:00:22
what do you use to specify that they are cancer cells because you
00:00:26
know the slide you can have definite mine fight to weaken how rituals landfill
00:00:32
how do you deal with that right yeah so you asking um how can we
00:00:37
be sure that that that the data is is this a typing data is accurate our
00:00:42
if we see something that's called a kansas i is it when a kansas and how
00:00:45
does it affect the infinity big patterns that we then identify here is that the question
00:00:50
yeah so yeah that's that's a great question so so neat map which this approach is designed
00:00:55
as the and the top layer of the of the multi step process that it takes another two
00:01:01
to do this kind of of work right so there is a image instead and then there's the sex says segmentations that
00:01:07
and then there's a set a a typing step install this has to be accurate because if you have ever
00:01:12
at this stage this is this can generate force negative and false positive when we look at it those uh
00:01:18
affinity big inference so neat and doesn't try to correct this is this is the responsibility of all the all the layers of the
00:01:25
of the um uh uh not to pick sister just stack 'em but this being said there's been good a
00:01:32
significant advances in over the years in improving these writing
00:01:36
in correcting errors in the latter spillover of of signals
00:01:40
and and this is still ongoing this paper that could improve on this uh continuously
00:01:45
in addition there's a level of robustness in any pap analysis which is that it need not be set is a good approach
00:01:50
so it will only reports speech if you know types eve
00:01:54
those or systematic so you would need to have a systematic error
00:01:57
for those to forbid the analysis here if it just happens if you says it won't be picked up and did answer question yeah
00:02:09
yeah thanks for the presentation was super nice or question in um you know you just
00:02:14
say you relied on the cell types in and then you have got on top of it
00:02:19
you know we should be other markers actually useful calling cell types right so you kind of reusing the same markers
00:02:26
i've used it finds out actually it's kind of advice we doubled the menus the same data twice
00:02:32
do you actually use all the markers still every irrespective of how the the
00:02:36
cell types work all or you exclude some of them because the reusing the sort
00:02:42
yeah i agree question nice clinicians to clarify what you're saying is that there's markers and we can
00:02:47
if we use them twice then it it's not the proper way to data analysis right we should
00:02:51
'cause when we can rediscover thing that that we're this over the previous stages of the analysis of course um yeah
00:02:57
so it so the answer is um we actually speed the markers into two sets right so yeah this forty six markers
00:03:03
and some of them are used to determine the cell types is like
00:03:06
lineage markers and then other markers are used to determine function infinite types
00:03:12
and so at the first stage only look furnishes we use use the first set and and the second stage where you put affinity big
00:03:18
organisation then we use use only the second sets so this competed does
00:03:21
john sets of of markers that we use for said typing infinite typing
00:03:30
i think you flew them i still uh how do you size the spatial
00:03:34
resolution them up a little pixel is is you do some segmentation to get to
00:03:41
one label for every soul or something uh what was the p. c. straight streets or
00:03:49
e. friends um so i'm not sure that the the question it accurately or you are we
00:03:54
asking how do we decide what is the size of the sites that we study when you
00:03:59
consider composition before doing principle commits analysis and not long after for a once the the p.
00:04:05
c. e. c. plates are like to apply it's on the wall you much yeah how it's
00:04:14
um yeah and the trend understand the
00:04:21
so so it out
00:04:26
okay so you're asking on what resolution do we perform a tissue segmentation intern issues you are right
00:04:32
well this this arbitrary reckon just make it as uh
00:04:36
as fine grain uh as you want like it's um it's
00:04:45
alright okay so actually yeah okay okay so so then the question of i believe is is
00:04:50
the first the first day of the first we have i i try to formulate attributes that
00:04:54
how do you decide what is the what is the size of the sites that that you wanna study because the the term is the
00:05:00
the landscape of the tissue with our audience then save the tissue right okay so what we've
00:05:05
done is to um explored with with this question systematically die and coming to it there we go
00:05:14
that's the slide so what we've done here is to change the the size of the of the sampling or a area
00:05:21
in from you know from zero uh my computer up to two hundred my commuters and then we've looked at
00:05:27
um how how does how this structure emerges sarah composition of the sites
00:05:32
so the intuition there is that if your site is too small gonna get one set
00:05:35
of the times is gonna be just no structure right no correlation structure interprets becomes analysis
00:05:40
so it and then as you start to increase the size of the sites
00:05:43
you gonna start to see correlation between between said types now because this is
00:05:47
where you start to pitch to show architecture so this you see here right as you increase the side of the of the size
00:05:53
the the there's more you did less and less principal components in other to explain
00:05:57
a tissue architecture and so we decided on the size of twenty five my computer because it seems that if use
00:06:04
more increase the size even more there's no more um new architecture that that that it quite this is into the
00:06:10
principal compromises and so this would suggest that an ad for
00:06:14
a landscape forty ish architecture could be twenty five my computer
00:06:17
which has to europeans like two to four says which maybe perhaps not biologically may make
00:06:22
sense right this maybe this or it could be that like it's been shown that um
00:06:27
uh oxygen diffuse isn't on my computer with from but this is
00:06:30
uh the future signals often affects says up to only micro meters away
00:06:35
and proximate signalling we definitely need research so that would be to my computer so it's kind of in
00:06:39
the ball park of of what we expect from from t. should better g. so the dancer question yeah
00:06:47
oh i think so much with remus took so how that you know
00:06:53
the full question so when you fix than the size of june you should
00:06:57
would you compute exactly you should you complete coverage of
00:07:01
those markers lose we concede and some texture uh_huh rooms so
00:07:07
yeah so suggesting what is the um what what is the what is the
00:07:12
table on which would perform the simple compromises that you understand that there is a
00:07:15
the rose or or or or sites that you study what other columns what are
00:07:19
the features to the feature is is just counting uh the different types of says
00:07:24
right is that how many cancers says how many t. says how many pieces and in this we
00:07:28
we do this because this is what's been shown to work uh already for they could to commute ecology
00:07:33
in which you only look at its species and then you look at at morphology in a separate analysis yeah that answer question
00:07:47
hi thank you have a very nice very nice though so i think
00:07:52
that's a nice is somehow use basically the extra percent clear distinct operations
00:07:57
i was wondering about the thing that i don't h. c. that i make sure that this
00:08:01
is what these populations are half an external well you look into it it was one thing
00:08:07
here in the only question yeah see so to do you saying the it's
00:08:12
clear that this either close to the end points or or else's opinion say
00:08:16
exactly yeah unique a homogeneous finishes are right i it often h. i'm nice
00:08:22
and extent of cancer say it's i mean it's it's i was wondering what is
00:08:26
it may be a hundred percent it actually writing in sales i would have to look at it
00:08:32
this is a total yeah that's interesting so the um like we haven't looked at
00:08:36
this in detail because once we knew what the niches where we felt were satisfied
00:08:40
but the way we just this question was to generate this to show architecture table which is
00:08:45
a summary of the construct affinity architecture and you can see for example that the inflammatory cans image
00:08:50
has specific cell types right so you get a energy that's one positive uh turner says
00:08:55
you get a macrophages that have um markers of of of of protein synthesis
00:09:01
um any gets a activated you regulatory said so you denis is
00:09:06
would would would produce this is this is that the answer the question
00:09:10
i it's an open question so yeah that's not clear consensus i thank you
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
SIGCHI Social Impact Award Talk
Leysia Palen, University of Colorado Boulder, Boulder, United States
April 20, 2015 · 2:34 p.m.
1711 views