►
From YouTube: DevoWorm #24: GSoC project updates, issues update, networks as differentiation waves and epithelia
Description
GSoC Coding period (weeks 3 and 4): Updates on Digital Microspheres and D-GNNs. Activity Review (Github Issues, project tracker). Discussion about organoid modeling and differentiation trees, waves, and hypernetworks. Small-world networks as epithelial cell populations. Attendees: Susan Crawford-Young, Harikrishna Pillai, Morgan Hough, Karan Lohaan, Longhui Jiang, Jiahang Li, Wataru Kamakari, and Bradly Alicea
B
B
B
C
C
C
C
So
what
are
you
planning
on
doing
with
the
like?
I
know
that
the
you
know
labels
are
important.
You
have
the
cell,
I
guess
you
have
some
data
in
you
know
different
data
sets.
You
have
different
types
of
data
and
you're
going
to
put
them
together,
or
are
you
going
to
use
all
the
data
sets
in
different
ways
or
what
are
your?
What
are
your
plans
for
that.
B
C
Okay,
yeah
yeah,
I
think
so
yeah.
I
think
you
should.
You
know,
treat
the
every
data
set
as
a
potential
source
of
information,
so
some
of
the
data
sets
are
going
to
when
you
start
to
go
and
try
to
segment
them.
Some
are
going
to
have
certain
types
of
information
highlighted.
Others
are
going
to
have
other
types
of
information
highlighted
so
like
when
we
did
our
meeting
on
thursday.
C
We
talked
about
you,
know,
thinking
about
the
source
data
and
how
it
can
affect
like
what
you
get
at
the
end,
so
you're
processing
the
data,
you
segment
it
and
then
you
get
this
this
network,
but
it's
contingent
upon.
You
know
what
you're
able
to
extract
out
of
the
data
sets
and
so
like
something
like
a
fluorescent
data
set
is
going
to
be
able
going
to
allow
you
to
extract
certain
features
more
easily
than
other
data
sets.
C
Now
in
the
past,
we've
had
this
same
issue
where
we've
used
different
algorithms
to
segment
different
types
of
data.
So
you
know,
if
you
have
a
really
nice
high
resolution
image,
you
can
use
certain.
You
know
you
can
use
an
algorithm
and
it
can
pretty
much.
It
can
pretty
easily
pick
out
a
lot
of
the
features
that
you
want
to
use
to
segment
individual
cells.
So
you
can
segment
the
nucleus,
you
can
segment
the
membrane
and
you
can
segment
some
other
things
in
fluorescent
images.
C
C
You
know
they
have
some
quality
control,
not
really
sure
what
the
quality
control
is
on
each
data
set,
but
I
think
you
can
safely
assume
that
they're
workable,
you
know
you
can
you're
able
to
work
with
them
and
then
you're
able
to
extract
these.
You
know
you'll
have
to
find
you
know.
Usually
we
use
the
nucleus
as
a
marker.
We
use
some
centroid.
C
We
call
it
a
centroid,
because
you
know
we
don't
want
to
call
it
the
nucleus,
because
it's
not
really
the
nucleus
per
se,
but
you
would
use
something
like
you
know
that
fluorescence
signal
that
would
be
the
centroid
of
a
cell
and
then
your
embeddings
would
be
some
criterion
on
those
dots
which
would
be
the
nodes,
so
it
could
be
distance
like
we
had
mentioned
in
the
meeting.
It
could
be.
C
You
know
some
other
criterion,
but
but
I
think
the
important
thing
here
is
to
extract
the
data
to
to
segment
the
images
you
may
have
problems
with.
You
know
error
like
false
positives
or
things
like
that,
but
that's
something
that
you
can
control.
You
know
it's
something
that
we
can
control.
We
know
how
to
do
that,
so
just
to
keep
those
things
in
mind
that
this
this
step
is
actually
pretty
key
to
like
getting
these
correct
network
embeddings
so
yeah.
This
is
nice.
C
C
C
A
A
I
mean
I've
described
some
details
of
our
him
and
he
has
some
experience
on
it
and
I
think
he
is
very
interested
in
joining
such
projects
in
our
research
idea,
and
so
I
think,
because
his
experience
in
such
as
computer
vision
and
such
as
some
other
deep
learning
tasks.
So
I
think
probably
he
can
he's
very
helpful
for
us
and
he
could
you
could
help
us
to
build
a
better
juicer
project
and
yeah.
Could
you
please
briefly
introduce
yourself.
D
C
Yeah,
well,
that's
good
yeah!
I
know
jiang
told
me
that
you
were
going
to
join
us
he's
going
to
join
in
with
him
and
motaro
and
enjoying
the
meetings
and
that
sort
of
thing
so
welcome
to
the
meeting
yeah
yeah.
Thank
you.
Thank
you.
Yeah
and
yeah
we're
giving
updates
on
different
gsoc
projects,
so
otaro
started
with
some
of
his
work
on
you
know.
C
Extracting
data
from
data
sets
going
through
the
data
sets
understanding
what
they
are
and
then
you
know
there's
some
quality
control
and
then
you
know
there's
some
extraction
of
data,
some
segmentation,
and
so
all
of
that
should
be.
You
know
the
sort
of
the
key
to
this
project
because
you
have
to
have
good
starting
data
to
get
to
the
embeddings
part
and
to
some
of
the
other
things
you
want
to
do
so
yeah
I
was.
I
was
really
impressed
with
the
meeting
on
thursday.
C
A
A
E
C
Yeah,
that's
good
yeah,
and
so
what
taro
mentioned
about
the
cell
tracking
or
the
yeah,
the
cell
tracking
data?
And
I
don't
I
don't
know
yeah
what
this
I
mean.
One
of
the
data
sets
has
it,
but
it
isn't
like
directly
relevant.
I
mean
some
of
this
too.
When
I
say
quality
control,
I
mean
what
I
mean
is
like
matching
up.
C
Some
of
the
different
data
sets
so
like,
if
there's
cell
tracking
data
that
you
have
you
know
making
sure
that
it
makes
sense
with
respect
to
the
sort
of
embeddings
that
you
get
or
the
the
segmentations
that
you
get.
So
that's
that
some
of
that
is
just
kind
of
like
you're.
Just
gonna
have
to
check
it,
like
you
know,
plot
it
out
or
yeah.
C
Looks
like,
but
you
know
that
that's
something
you
know,
and
I
don't
know
you
know
every
time
you
take
data
from
a
different
source,
there's
always
a
potential
for
like
error
or,
like
you
know,
a
mismatch
of
things,
because
not
every
group
uses
the
same
methodology
to
get
to
the
output
data,
so
sometimes
cell
tracking
in
c
elegans.
It's
it's
rather
easy
because
it's
relatively
easy,
because
you
know
there
isn't
a
lot
of
variation.
We
know
what
the
cells
are
and
what
their
tracks
you
know
kind
of
what
their
final
positions
probably
are.
A
Yes,
yes,
yes
yeah!
I
understand
that.
Yes,
I
think.
Maybe
we
will
start
with
trying
c
elegans
and
just
make
make
a
demo
so
that
we
will
make
sure
that
all
things
just
work
like
what
we
imagine
and
then
we
will
move
to
other
large
data
sets
such
as
tracking
challenge
or
such
as
some
other
data
set.
They
will
use
in
the
devlin
yes,
but
we
will
start
with
the
c
elegans,
because
I
think
it
is
small
and
simple.
Yes,
yeah.
A
Okay,
yes,
yes
and
yeah
yeah,
just
like
what
I
said.
We
I've
mentioned
the
paper
yeah
last
week
and
I
think
they
have
some
source
codes
which
will
be
helpful
for
us
and
I
will
try
those
source
codes
because
they
also
deal
with
those
large
data
sets,
such
as
those
sales
tracking
challenge
data
sets,
and
I
will
check
how
they
implement
their
codes
to
deal
with
those
large
datasets
to
to
segment
them
or
to
extract
positions
of
these
sales.
Yes,
I
will
check
their
source
codes,
yeah
yeah,.
C
F
Yeah
so
my
week,
like
I
outlined,
I
think
in
the
I
think,
previous
update.
You
know
that
I
was
improving
the
contour
extraction
for
the
initial
stage
of
you
know:
building
the
3d
model
from
the
2d
image
so
for
this
particular
one
you
know
this
is
like,
like
so
far
how
we're
going
about
you
know
if
you,
if
you
can
see
you
know
these
outlines
how
they're
mapping
about
this
axis
of
rotation.
F
So
this
is
again,
you
know
just
a
linear
programming
algorithm
that
I
designed
you
know
based
on
this.
So
what
it's
doing
is
it's
taking
2d
images
taking
their
outlines
and
their
centroids
as
well
and
based
on
the
centroids
of
all
these
stacks
of
2d
images.
You
know
it's
trying
to
build
the
frame,
so
this
was
the
you
know,
linear
programming,
way
of
getting
to
the
answer.
F
The
the
things
that
you
know,
hari
krishna
also
told
you
about
that.
We
were
trying
cold
map
and
you
know
neural
radiance
fields
for
improving
this
particular
step.
So
the
thing
is:
there's
still,
you
know
a
bit
bit
of
a
learning
curve
to
get
to
do
that.
So
that's
why
even
hari
krishna
was
having
you
know,
issues
in
rendering
a
model
using
like
improving
the
current
techniques
that
we
are
using
just
going
a
step
further
than
the
earlier
programming
way.
C
F
F
F
So
this
is
like
the
early
rendition
way
of
you
know
getting
better
at
approximating
the
final
frame
of
the
3d
model,
so
this
was
slightly
improved
because
earlier
you
know
I
was
I
had
to
resize
these
outlines
along
the
common
axis,
because
some
of
the
images
you
know
they
have
the
axolotl
subject.
You
know
slightly
farther
away
or
slightly
close
to
it
like
you
can
see,
you
know.
Still,
there
are
these
two.
F
C
F
C
C
G
F
G
F
Yeah
so
I'll
be
improving
the
again
I'll
have
to
come
back
to
the
step.
Unless
I
you
know,
I
can
see
how
the
final
mesh
is
forming
from
the
base
base.
That
is
this
so
I'll
be,
like
I
told
you
about
the
function
you
know
of
getting
more
lines
based
on
the
existing
ones,
so
I'll
see
how
that
works
out.
Otherwise,
we'll
keep
on.
F
F
C
I
C
I
I
So
that's
because
it
doesn't
like
understand
that
the
images
are
continuous.
It
feels
like
the
images,
are
not
continuous,
so
I
tried
some
image
augmentation
with
the
data
set
like
rotating
the
images
or
translating
the
images
in
different
directions,
so
that
I
can,
I
could
increase
the
numbers,
but
still
the
result
was
the
same
so
either.
I
will
try
doing
something
with
the
data
switch
again
or
I
will
move
on
from
this
algorithm
and
try
something
else.
C
I
C
C
H
H
H
C
Yeah
all
right
well,
thank
you
very
krishna
for
the
update.
That
was
very
good
and
I'm
glad
to
see
you
know
some
of
the
troubleshooting
going
on
and
some
of
the
experimentation
that
people
are
doing
projects
it's
very
important
to
go
through
the
project
and
to
kind
of
hit
those
marks.
You
know
where,
if
you
have
trouble,
you
know
if
things
go
well,
it's
great,
but
a
lot
of
times
people
are
going
to
have
obstacles
that
they
need
to
overcome.
C
Okay,
I
wanted
to
say
also,
I
know
he
left,
but
wataro
made
a
pull
request
on
divo
learn
and
he
made
a
small
improvement
to
it,
so
it
was
accepted.
Thank
you,
wataru
for
that.
That
was
some.
A
small
fix
to
the
diva
learned
platform
and
ajiya
hong
and
otaro
were
working
with
that
and
they
found
some
problems
with
it
that
needed
to
be
addressed,
and
that's
really
the
the
essence
of
open
source.
C
C
All
right
all
right,
you
can
see
my
screen
good,
so
we
do
have
this
list
of
major
tasks.
I
always
bring
this
up
every
once
in
a
while
to
talk
about
some
of
the
projects
that
we
have
ongoing
in
the
group.
It's
pretty
diverse,
as
you
can
see,
there
are
a
lot
of
things
here,
a
lot
of
issues
that
are
sort
of
unaddressed.
We
have
some
things
that
are
done
finished.
We
have
some
things
that
are
in
progress.
C
We
have
some
action
items
we
have
holds
and
to
do's.
The
to
do's
are
mainly
things
that
we've
come
up
with
as
ideas
and
we're
not
really
there
yet.
So
some
of
these
things
are
maybe
things
that
have
fallen
through
the
cracks
and
holding
to
do,
but
there
are
things
we
can
get
to
with
you
know
if
someone's
interested
in
it,
they
can
go
ahead
and
do
that.
So
this
is
in
the
diva
worm,
repository
under
group
meetings
and
it's
the
project
there's
only.
C
I
think
only
one
project
board
here,
so
this
is
major
tasks
for
20
20
22..
So
this
has
been
something:
we've
been
working
on
since
2020
kind
of
going
through
some
of
these
things.
So
we
have
things
like
papers
and
we
have
lectures
that
people
might
want
to
do.
We
have
these
different
projects
that
you
know
we
will
make
a
presentation
somewhere
and
then
you
know
we'll
have
follow-up
on
that
and
that
this
is
the
place
we
put
this.
C
C
C
C
We've
also
have
an
interest
in
in
working
with
virtual
worlds
and
simulations
if
you've
not
seen
the
open
worm
viewer,
you
you're
invited
to
go,
look
at
the
browser,
it's
browser.openworm.org
and
that's
an
example
of
one
of
these
virtual
organisms
or
how
you
can
build
a
virtual
organism.
Basically,
you
have
the
worm
you
can
zoom
in
on
different
cells.
C
So
this
is
something
that
I've
had
as
issue
number
113.
It's
if
you're
interested
contact
me
on
that
there's
also.
There
are
also
things
like
the
mathematics
of
diva
worm
document.
This
is
actually
something
that
is
now
made
into
a
poster.
So
I
may
move
this
over
later
to
in
progress
or
finished,
because
we've
kind
of
moved
in
you
know
this
is
kind
of
like
a
coherent
poster.
Now,
so
that's
good.
C
We
have
a
video
highlight
of
this
new
paper
that
isn't
quite
out
yet,
but
when
it
comes
out
as
the
book
chapter,
this
bacillary
a
neural
cognition
paper
that'll
be
out
as
a
video
highlight
and
we
have
a
youtube
channel
where
we
put
different
things
out
so
that'll
be
one
of
the
things
that
come
out.
Maybe
at
the
end
of
the
summer
we
can
make
video
highlights
of
the
different
gsoc
projects
as
well.
That's
something
we
can
put
on
the
list.
C
So
we
have
all
these
ideas
and
they
kind
of
like
you
know,
they're
kind
of
loose
ends
that
get
thrown
out
at
the
meetings
and
you
can
see
there
are
a
lot
of
them
here.
So
if
you're
interested
go
through
this
list,
if
you
see
something
you're
interested
in
contact
me,
you
know
you
can
put
in
it.
You
can
put
in
addition
to
the
issue.
So
you
have
the
issue
and
you
just
make
a
comment,
and
I
should
get
an
alert
about
that.
E
Hi
there,
yes,
so
it
is,
it
is
school,
run
time
here
in
san
francisco
yeah,
so
I'm
I'm
actually
just
getting
some
things
together
to
get
my
daughter
to
school,
but
organoids,
you
know
yeah
the
certainly
some
of
the
self-organization
of
just
stem
cell
work
is
it's.
It's
been
what
we've
been
looking
at
recently,
but
not
not
much
new,
not
many
updates.
Sorry,
oh.
C
That's
okay!
Well,
I
just
wanted
people
to
be
aware
that
you
were
interested
in
the
ocean
yeah,
so
yeah
yeah,
please!
Yes,
yes,
so
we've
we've
done
some
stuff
we've
well.
We've
talked
about
organoids
in
this
group.
There
are
neural
organoids,
which
are
where
you
have
neural
cells
that
you
can
grow
up
into
these
little
organs
or
organoids.
C
So
you
have,
you
know
like
a
bunch
of
neural
stem
cells
and
you
differentiate
them
into
different
types
of
neural
cells,
and
you
get
these
little
structures
that
look
like
no.
They
may
not
look
like
brains,
they
could
be
brain
structures
and
then
you
can
test
hypotheses
about
function.
You
can
look
at
like
disease
states,
you
can
do
that
other
sorts
of
things
and,
of
course,
they've
done
organoids
for
every
different
organ
in
the
body.
C
C
So
if
we
have
a
plate
of
liver
cells,
we
can
sample
those
and
say
with
the
liver,
you
know
maybe
like
metabolism
in
the
liver,
or
we
can
talk
about
like
gene
expression
in
the
liver
or
something
like
that
in
organoids.
You
have
this
added
dimension
of
of
phenotype,
so
you
have
this
first
of
all,
usually
there's
a
third
dimension
to
the
organoid,
but
also
you
have
this
function
of
the
organ
itself.
So
you
start
to
get
things
like.
You
know:
different
types
of
associations
between
cells
and
things
like
that.
C
There
is
to
look
at
some
of
the
things
that
are
going
on
in
that
process
and
so
organoids
bridge
that
gap
and
so
morgan's
been
interested
in
trying
to
get
some
things
off
the
ground
with
that
and
I've.
I
I
think
you
know
longer
term.
I
see
this
as
being
kind
of
a
promising
area,
there's
also
the
idea
of
computationally
modeling
organoids.
C
So
you
know
this
is
something
that
we've
done
with
computational
modeling
of
embryos.
We've
talked
about
physics,
modeling
we've
talked
about
some
of
these
other
things
and
organoids.
You
know
we
can
also
model
organoids
in
terms
of
sort
of
the
early
origins
of
cells,
self-organizing
and
different
types.
H
C
H
E
E
G
H
The
3d
tissue
model
it
keeps
telling
me
it's
set,
is
empty
and
that
it's
unsolvable,
which
is
typical
of
a
tensegrity
object.
But
then
I
don't
know
how
to
find
the
data
that
it's
throwing.
So
I
don't
know
where
it
is
in
the
set
of
data
that
there's
an
empty
line
or
whatever
that's
it's
going
on
about.
So
I'm
having
some
trouble.
C
C
C
Well,
yeah,
it's
always
a
struggle
with
modeling,
though
it
seems
so
easy
and
then
it's
not
yeah
so
jia
hong
had
to
leave.
So
thank
you
for
attending
jiahung
and
it
looks
like
guatara
also
left
so
yeah.
Why
don't
we
go
back?
I
have
some
papers
to
present
and
I
have
a
couple
of
things
actually
to
talk
about.
C
So
let
me
share
my
screen
again.
Thank
you,
morgan
and
susan
for
your
updates
and
ari
krishna.
Thank
you
for
attending.
So
this
is
a
couple
things
here
to
talk
about
today.
First
of
all,
dick
as
dick
gordon
sent
me
some
figures
from.
I
think
this
is
from
one
of
his
books
on
differentiation,
waves
and
differentiation
trees,
and
so
we
were
talking.
I
I
don't
have
the
paper
right
now,
but
it's
it's
up
on
researchgate.
C
It's
a
paper
I'm
going
to
be
presenting
on
later
this
month,
at
the
network
science
conference
on
embryo
networks
a
little
bit
different,
they're
hyper
graphs.
That
represents
multiple
features
of
development,
in
what
I
consider
to
be
a
c
elegans
like
embryo,
and
so
I
kind
of
walk
through
that
in
the
paper.
So
it's
a
little
bit
different
than
the
gsoc
project,
but
it's
related
now
dick
sent
me
these
pictures
and
I
think
they're
from
his
first
book,
the
1999
book,
and
so
there
are
a
couple
figures.
C
I
think
that
triggered
him
during
that
talking
about
that.
So
the
first
one
is
this
four
dimensional
geometry
of
the
development
of
a
regulating
organism.
So
we've
talked
about
regulating
organisms
that
are
where
the
cells
send
signals
to
one
another
and
it
determines
the
fate
of
the
cell.
So
in
c
elegans
we
have
what
we
call
mosaic
development,
which
is
where
the
cells
have
a
deterministic
fate
from
the
mother
cell
to
the
daughter
cell.
That
information
is
passed
on
and
then
by
and
large
those
cells
will
retain
that
all
fate,
ultimately
in
adulthood.
C
So
the
the
fade
is
determined
from
the
birth
of
the
cell
in
irregulative
or
development.
However,
different
cells,
depending
on
their
gene
expression
state
depending
on
their
neighbors,
if
you
take
stem
cells
for
example,
and
put
them
into
a
heart
or
a
liver,
they
take
on
the
heart,
some
heart
or
liver
fate,
and
so
this
is
a
regulative
embryo.
So
this
is
where
we
have
this
map.
This
axis
here
is
developmental
time,
so
these
slices
going
up
is
increasing
in
developmental
time.
So
this
is
the
the
one
cell
egg.
C
C
C
So
that's
one
and
my
the
hypergraphs
actually
that
I've
been
working
with
actually
look
something
like
this
there's
a
very.
There
are
a
lot
of
similarities
here
and
then
this
is
another
example
where
you
have
developmental
time.
You
have
this
again,
the
egg.
This
is
mosaic
development,
so
this
is
like
c
elegans
development,
where
each
where
the
mother
cell
determines
the
fate
of
the
daughter
cell,
just
by
virtue
of
passing
on
its
genome.
C
So
this
is
the
3d.
This
is
the
egg
is
the
single
cell,
and
then
you
get
two
different
cells.
They
have
different
sizes
here.
This
is
an
asymmetric
division,
so
we've
done.
We
did
a
paper
on
this.
Where
we
looked
at
in
the
c
elegans
lineage
tree,
we
looked
at
the
different
sizes
of
cells
and
it's
it's
it's
an
imperfect
thing
to
approximate,
because
cell
sizes
are
not.
C
You
know
you
can
measure
that
you
can
approximate
the
size
by
segmenting
images
and
calculating
the
volume
of
that
image,
but
it's
not
it's
not
a
perfect
way
of
doing
it.
In
any
case,
you
can
differentiate
between
symmetric
divisions
where
the
daughters,
the
two
daughter
cells,
are
basically
the
same
size
and
then
asymmetric
divisions
where
the
daughter
cells
are
of
different
sizes.
And
so
that's
what
you
see
here
in
this
slice,
where
the
daughter
cells
are
two
different
sizes
and
then
in
this
slide
at
the
top,
you
have
this.
C
C
C
So
I
like
this
visualization
because
I'm
really
a
fan
of
a
lot
of
this
bibliometric
stuff,
but
it
does
show
some
of
these
topics
some
of
them.
We
don't
talk
about
in
the
group
much,
but
some
we
do
so.
This
is
specific
to
network
theory,
and
it
just
shows
like
the
frequency
of
these
different
topics
over
time
from
1998
to
basically
the
present,
so
there
have
been
hyper
graphs
and
simplical
complexes.
We
talked
about
that.
I
talked
about
that
with
g
hong
on
thursday.
C
C
C
Strogets
and
watts-
and
so
you
can
see
that
there
was
a
trend-
you
can
see
a
trend
towards
that
maybe
20
years
ago,
and
it's
decreased
since
then,
and
so
there
are
all
these
different.
You
know:
networks
embedded
in
metric
spaces.
This
is
another
thing
that
we've
kind
of
talked
about
in
the
group.
This
is
kind
of
grown,
maybe
about
15
years
ago
and
stayed
constant.
C
Then
finally,
this
is
a
picture
of
a
comparative,
a
set
of
comparative
embryos,
just
to
show
you
the
size
of
different
embryos.
So
this
is
the
axolotl
embryo.
This
is
the
embryo
that
susan
and
karan
and
hari
krishna
are
working
on
it's
very
large
embryo,
as
you
can
see,
and
if
you
compare
that
with
xenopus
and
lamprey,
you
see
that
it's
much
bigger
than
xenopus
and
lamprey
xenopus
being
smaller
but
still
sort
of
in
the
same
family
as
axolotl.
C
But
then
lamprey
is
this
what
they
call
phylogenetic
outgroup,
which
is
in
it's
in
the
area
of
like
fishes
in
that
area
of
the
tree
of
life?
So
this
is
a
much
smaller
embryo.
I
don't
know.
If
there's
a
correlation,
I
don't
think,
there's
a
correlation
between
embryo
size
and
phylogeny.
It's
just
that.
Sometimes
you
get
really
big
eggs
in
some
species
and
smaller
eggs
and
other
species,
and
but
this
is
just
a
size
comparison
here
of
these
different
types
of
embryos.
C
So
then
I'm
going
to
get
into
some
papers
here.
Actually
I
want
to
talk
about
this
paper.
This
is
a
really
nice
one
and
I'm
going
to
zoom
in
on
a
little
bit.
If
I
can
here
it's
down
here,
okay,
so
this
is
a
paper
and
going
back
to
this
idea
of
networks
and
we've
talked
about
networks
in
embryos,
and
we
talk
we're
talking
about
making
these
graph
embeddings
and
we're
talking
about
different
hypergraphs
and
things
like
that.
C
This
paper
is
about
endothelial
cell
signaling,
so
this
is
within
a
layer
of
cells
called
the
endothelium,
and
this
is
the
cell
signaling
study
of
these
cells,
and
they
say
the
title
says:
small
world
connectivity
dictates
collective
endothelial
cell
signaling,
so
in
the
endothelium
you
have
these
cells
and
they
have
signaling
between
them.
As
I
said
in
regulative
development,
you
see
often
see
the
cell
signaling
serve
to
organize
things,
so
it
organizes
collective
behavior,
so
they're
proposing
this
idea
of
small
world
connectivity
being
the
driver.
C
So
this
is
the
significant
statement
here.
The
endothelium
is
the
single
layer
of
cells
lining
all
blood
vessels
and
acts
as
a
central
control
hub
to
regulate
multiple
cardiovascular
functions
in
response
to
hundreds
of
physiological
stimuli.
So
this
is
an
important
part
of
the
body.
Important
function,
physiological
function.
C
The
detection
of
various
physiological
stimuli
is
distributed
in
spatially
separated
sites
across
the
endothelium.
So
this
is
where
you
have.
The
detection
of
stimuli
by
cells
cells.
Have
this
sort
of
perceptual
aspect
to
them
that
we
don't
talk
about
as
much
as
we
probably
should,
but
they
can
sense
things
in
physiological
states
of
the
organism.
They
can
sense
environmental
states,
so
they
respond
in
kind
and
so
the
the
distribu.
The
distribution
of
this
is
spatially
separated
in
different
places.
So
you
don't
have
it
not.
Every
cell
is
equally
exposed
to
these.
C
C
C
Well,
thank
you
for
attending,
so
this
is
so
this
is
you
know
these.
These
cells
are
connected
in
different
ways,
but
they're
also
connected
at
different
strengths.
So
some
of
the
cells
serve
to
distribute
information,
some
of
them
just
kind
of
receive
them
on
the
periphery
of
these
networks,
and
this
is
the
kind
of
these
types
of
organization.
This
is
what
they're
proposing
the
organization
confers
a
high
spatial
propagation
speed
and
a
high
degree
of
synchronizability
across
the
endothelium.
C
C
C
So
they
use
graph
theory
to
sort
of
get
at
these
issues
and
they
have
a
nice.
I
can't
seem
to
load
the
rest
of
the
file,
but
this
is
basically
what
they
have
to
say
here.
I
think
it's
an
interesting
study
and
that
they're
able
to
you
know
they
extract
these
networks
from
physiological
data
and
they're,
able
to
characterize
these
networks
and
say
something
about
the
how
the
sort
of
the
function
all
right.
So
then,
if
we
go
down
the
rest
of
this
paper,
we'll
see
that
they
have
a
figure
here.
C
So
this
is
a
random
network
and
the
usual
uses
to
calibrate
some
of
the
things
that
they're
when
they
compare
skill,
free
and
small
world
networks,
they
have
to
compare
it
to
something
because
the
the
structure
deviates
random
in
some
way
and
so
the
scale
frame.
This
is
an
example
of
a
scale
free
network,
and
these
really
characterize
in
terms
of
the
degree
distribution
or
how
many
connections
there
are
into
each
node,
and
so
they
did
this
as
an
average
or
they
do
this
over
a
distribution.
C
And
so
this
is
the
example
here,
which
is
usually
scale-free
networks,
as
well
as
small
world
networks
exhibit
long-tail
behavior,
which
means
that
there's
a
long
tail
of
district
of
the
distribution,
so
some
nodes
have
more
connections
than
others.
In
this
case,
you
have
a
random
network.
The
random
network
does
not
have
a
long
tail.
It
has
this
sort
of
random
distribution
where
you
have
an
a
mean
number
of
connections
and
then
a
small
amount
of
variation
less
or
more
than
that.
C
So
the
random
connection
really
doesn't
have
much
structure
as
a
result,
messages
get
passed
from
cell
to
cell,
more
or
less.
You
have
very
few
long-range
connections
across
the
network
in
a
scale-free
network.
However,
you
can
make
it
across
the
network
quite
quickly
by
just
passing
messages
from
a
node
to
a
hub
and
then
to
another
node,
and
so
that's
the
advantage
of
these
skill-free
and
small
world
networks.
C
But
if
we
look
at
so
we
look
at
this
up
here,
we
have
it
on
a
lattice
we
have.
This
is
in
terms
of
this
path,
length
measure
I
told
you
about
so
in
a
regular
network.
You
have
this
manhattan,
what
they
call
manhattan
distance,
which
is
where
you
move.
You
know
from
node
to
node
in
a
linear
fashion
in
a
stepwise
fashion.
So
you
end
up
with
these
path
links
that
require
you
to
always
go
from
node
to
node.
There
are
no
shortcuts
in
a
small
world
network.
C
You
get
this
shorter
path
length,
but
that's
largely
an
artifact
of
just
randomness.
So
the
path
link
there's
no
real
advantage
to
path
length
in
a
random
network
you
get.
Sometimes
it
can
be
shorter.
Sometimes
it
can
be
longer,
but
there's
no
structure
in
the
random
graph
that
allows
you
to
have
a
shorter
path
length.
C
So
this
just
allows
us
to
see
like
what
neighbors
are
sort
of
in
clusters
together
and
what
neighbors
are
not,
and
so
in
small
world
networks,
you
get
clusters,
but
you
also
have
these.
No,
these
hubs
that
allow
for
the
easy
facilitation
of
of
messages
of
signals,
and
so
this
is
the
advantage
of
this
type
of
network
structure
again,
where
you
get
shortcuts
that
form
and
you
get
one
pair
among
neighbors.
So
it's
you
know
in
a
random
network,
you
get
really
non-specific
connectivity
in
a
small
world
network.
C
You
know
the
relationships
between
them
and
then
they're
correlating
that
structure
on
functional
network.
This
is
a
lot
like
what
they
do
in
neuroimaging
for
the
brain,
so
this
just
gives
us
a
number
of
parameters
to
work
with.
This
is
just
an
example
of
how
you
can
build
these
networks
from
a
population
of
cells
and
some
sort
of
the
properties
between
say,
like
a
regular
connection
regime,
a
random
connection
regime
and
then
this
middle
ground,
which
is
the
small
world
or
scale
free
networks.
C
So
this
is
an
example
of
some
of
the
cell,
like
what
they're
doing
with
cells
they're
measuring
they're
looking
at
signaling
itself.
So
these
are
the
things
that
define
the
connections,
these
signals,
and
so,
in
this
case,
they're
looking
at
calcium,
signaling,
they're,
looking
at
the
intensity
of
the
signal
between
cells,
how
they're
exchanging
calcium
and
how
they're
communicating
with
one
another.
C
In
this
case
they
do
network
analysis
of
acetylcholine
and
histamine,
evoke
calcium
responses
in
the
endothelium,
so
they're,
looking
at
neurotransmitters
and
they're,
looking
at
these
calcium,
ions
and
they're
looking
at
their
activity
in
the
epithelium-
and
this
is
another
way
you
can
characterize
connectivity.
So
if
you
have
cells
that
are,
you
know,
activated
together,
this
is
like
an
again
like
in
neuroimaging
if
the
cells
are
activated
together.
C
If
the
signal
shows
up
in
cells
in
different
places,
they
can
be
assumed
to
be
connected,
and
so
this
is
how
they're
building
this
connectivity
map
they're.
Looking
at
this
cell
culture,
where
they're
looking
at
the
cell
population
they're
looking
at
the
intensity
of
the
signal
for
this
specific
marker
that
they
want
and
then
they're
assuming
that
that
consists
of
a
correlation
matrix.
C
So
when
cells
are
expressing
a
certain
amount
of
this
marker
together,
then
that's
that's
a
connection
and
it's
a
strong
connection
depending
on
the
correlation,
and
so
you
can
correlate
these
over
time
as
well,
because
these
processes
are
time
dependent
you
for,
for
example,
you
don't
want
to
just
pick
up
a
spurious
correlation
and
assume
it's
a
connection
between
two
cells.
You
want
to
have
a
stable
connection
over
time,
and
so
this
is
some
of
the
things
that
they're
doing
here.
C
Also,
not
you
know,
homogeneous
across
the
tissue
and
or
across
the
cell
population.
So
you
have
these
opportunities
for
differentiation
of
diff.
You
know
little
it's
a
function.
You
have
this
opportunity
to
connect
different
parts
of
the
network
and
so
forth.
This
local
structure
shows
that
you
know
you
can
have
these
cells
that
are
highly
connected.
C
Now
community
structure
is
different,
because
community
structure
is
where
you're
finding
local
communities
and
they're
broader
than
these,
like
local
clusters,
because
they're
connected
components.
So
you
have
these
local
hubs,
for
example
on
a
small
world
network
that
are,
you
know,
connected
to
its
local
cluster,
and
then
you
have
these
connector
hubs,
which
connect
the
two
local
hubs,
and
this
together
forms
a
community.
C
So
you
can
think
about
this
in
terms
of
hierarchical
signaling.
If
you
have
a
connector
hub
and
then
you
have
your
local
hubs,
those
local
hubs
that
were
going
to
be
organized
by
the
connector
hub
and
these
sub
in
these
clusters
or
sub-networks,
are
going
to
be
organized
by
their
local
hub,
and
so
that
may
be
very
important
for
organizing
some
of
these
signals,
the
diffusion
of
information
and
so
forth.
C
C
So
this,
I
think,
is
a
nice
nice
paper.
They
kind
of
pull
a
lot
of
information
out
of
these
images.
It's
certainly
not
something
that
most
people
do
for
some
of
these
systems,
but
I
think
it's
a
useful
tool
to
sort
of
understand
what's
going
on
especially
over
time,
and
so
that's
that
paper
so
yeah,
so
I
guess
long
we
had
to
leave
as
well
and
susan.
Thank
you
for
attending.