►
From YouTube: DevoWorm (2023, #6): DevoWorm on Huggingface II, GSoC, Tensegrity Networks II, Vesicle Sims II
Description
DevoLearn/DevoWormML on Huggingface and Github. New models and repositories, open data on Huggingface. C. elegans cell modeling and cell geometry. Models of cell-cell interaction and cytoskeletons in development. Models of tensegrity networks, biological material soft physics, and network organization Mathematical modeling of spontaneous autocatalysis in vesicles: adding a toy model of Braitenberg Vehicles, a mixed molecular world, and memory mechanism for spontaneous autocatalysis. Attendees: Richard Gordon, Susan Crawford-Young, Sushmanth Reddy Mereddy, Gautham Krishnan, and Bradly Alicea
B
Well,
how's.
Everyone
today.
B
C
B
C
B
On
in
that,
using
that
tool,
and
then
he's
also
doing
some
things
with
the
GitHub
repository.
C
A
huge
data
set
of
and
I
was
trying
to
just
make
a
model
which
can
predict
breast
cancer.
Is
there
or
not?
So
that's
the
side
project
actually.
A
B
B
C
C
D
B
D
Yeah
yeah
sure
so
I'm,
actually
a
sophomore
studying
for
b-tech
Cs
and
my
interests
are
machine,
learning,
deep
learning
and
stuff.
So
I've
worked
with
various
computer
vision,
model
segmentation
models
and
related
things.
So
I
from
the
past
one
week,
I've
been
working
looking
into
demograph
and
how
it
works
and
I
was
able
to
run
it
locally
and
yeah.
Yesterday,
I
made
a
PR,
also:
okay,
A
Small,
Change,
yeah
and
I'm
looking
forward
to
working
more
on
it.
B
Well,
yeah,
it
sounds
great
thanks
for
participating
and
contributing
yeah
look
forward
to
it
as
well
right,
yeah,
so,
okay
yeah,
let
me
go
to
the
actual
I'll,
go
to
the
hug
and
face
I'll,
just
preview
that
and
then
I'll
go
to
the
GitHub.
So
this
is
the
hug
and
face
site.
It's
huggingface.co,
Diva,
worm
group.
We
have
an
organization,
so
you
know,
hugging
face
is
basically
a
hub
for
machine
learning
models.
It's
supposed
to
stress
reproducibility
and
you
know
people
can
put
different
organizations
have
their
code
up
here.
B
They
have
their
different
platforms.
So
you
know
there
are
a
lot
of
implementations
of
stable
diffusion
on
here,
for
example,
where
you
can
download
and
run
them,
and
so
this
is
for
us
we
have
for
now.
We
can
add
to
this.
If
people
want
to
add,
we
have
the
membrane,
segmenter,
lineage
population
model,
nucleus,
segmenter
and
thanks
to
sushma,
for
putting
those
together.
B
Getting
those
up
on
hugging
face
need
to
create
a
card
for
this,
but
we
might
add
more
as
we
go
along
I,
don't
know
there
I
think
there
are
a
couple
more
models
we
could
add
in
and
then
whatever
you
know,
people
want
to
do.
We've
had
a
lot
of
activity
in
terms
of
machine
learning
and
deep
learning
in
the
group.
So
I
don't
even
know
what
we
have
available.
B
There
may
be
other
things
that
we
could
put
up
here,
but
I'd
like
to
have
it
up
here,
because
it's
it's
a
nice
place
to
have
for
people
to
gain
access
to
it
easily,
and
then
we
can
also
I
think
you
know
put
other
I,
don't
know
if
we
can
host
other
types
of
models
here
as
well,
not
really
sure
what
the
rules
are.
They're
data
sets
you
can
host.
We
have
some.
C
Data
sets,
with
your
permission,
I
just
want
to
host
this
data
assets
and
models
by
today
itself,
because
actually,
last
week
only
I
I
was
about
to
deploy
the
data
set.
Also
I
thought
it
would
not
be
okay,
because
downloading
epic
data
set
and
keeping
in
hugging
places
something
yeah
I
thought
it
may
create.
Some
problem,
even
Minot
told
me
that
to
just
take
a
permission
from
badly
and
then
you
proceed
of
keeping
the
whole
data,
there
can
I
create
this
whole
network.
B
I,
well,
you
mean
put
a
data
set
up,
did.
Is
there
a
storage
limit
for
it
for
the
account.
C
C
C
I
don't
know
like
can
I
host
the
whole
data,
epic
data
set
here
or.
B
Well,
I,
don't
know
about
how's
well,
I
guess
we
could
host
the
version
that
we
use
I
mean
I.
Think
it's
publicly
available.
It's
not
like.
If
they
don't
you
know
it,
does
they
don't
want
it
public,
so
we
I
guess
we
could
host
a
version
of
it
here.
B
We
could
also
you
know,
host
some
of
the
things
we
have
on
GitHub
I
can
put
together
some
things.
Maybe
we
could
put
up
here
as
well.
So.
B
Well,
we've
put
together
different
data
sets
for
like
summaries
of
the
C
elegans
embryo
in
terms
of
like
the
cell
tracking
positions,
so
like
three-dimensional
position
so
approximations
of
cell
size,
and
then
we've
done
some
other
work
with
like
differentiation
codes
and
things
I'll
point
to
the
ones
that
maybe
we
should
put
up.
C
C
B
C
C
I
mean
one
issue
was.
C
C
B
C
B
Well,
if,
if
you
have
an
idea
for
how
we
might
do
that
because
I
don't
know
exactly
well
I'll
I'll
go
to
the
we'll
go
to
the
GitHub
repository
next
and
we'll
talk
about
it.
So
well,
we
should
yeah.
We
should
talk
more
about
that.
Bradley
yeah
I
have
a
general
question.
I
put
a
title
of
a
book
in
that
I'm
reading,
okay,.
B
A
Yet
it
seems
very
much
to
be
a
contribution
to
what
could
be
machine
learning,
yeah,
okay,
you
might
want
to
read
it
yourself
in
any
case
what
he
makes
a.
He
distinguishes
features
okay,
so
there's
there's
two
ways
of
classifying
things:
one
is
to
use
features
and
another
is
to
use
what
he
calls.
I
put
it
in
the
working
the
word
in
the
differentia.
A
And
it
seems
to
me
that
that's
basically
a
difference
between
features
but
yeah
and
though
he
doesn't
say
that
right
and
so
I'm
wondering
whether
he's
got
a
new
Theory
or
he
just
anticipated.
That
features
would
also
become
the
difference.
Between
features
will
become.
B
B
B
Well,
I
think
people
are
using
like
secondary
metrics
I
mean
the
features
are
usually
generating
a
feature
space.
So
it's
kind
of
like
a
first
step,
and
then
people
do
a
lot
of
other
kinds
of
processing.
B
A
Yeah,
okay,
so
I
and
and
he
was
calling
for
a
lot
of
research
on
whether
or
not
people
and
animals
do
this.
D
A
B
A
B
B
Yeah
so
yeah
we
were
looking
at
the
hugging
face
and
it
was
that's
that
that
will
keep
working
on
that
I'll.
Why
don't
we
search
month?
What
don't
we
talk
about
this
more
like
in
slack
about
the
data
sets
and
everything,
and
what
do
you
think
should
be
archived?
B
D
D
C
B
All
right,
yeah,
that's
great,
it
looks,
looks
great
then
we'll
have
yeah,
then
we'll
have
this
kind
of.
So
we
have
the
for
people
who
are
just
coming
to
Divo
learn.
You
know
you
have
the
devil
and
evil
learned
repository,
which
has
this
based.
Evil
learn
software,
and
this
was
our
idea
like
this
originated
from
our
idea
of
building
a
pre-trained
model
for
microscopy
data
for
embryos,
and
we
built
this
in
2020
yeah.
We
did
some
work
a
couple
years
prior
on.
B
You
know
kind
of
working
on
machine
learning
problems
and
then
we,
this
is
sort
of
the
culmination
of
that
and
then
we've
been
working
on,
updating,
Divo,
learn
and
then
you
know
also
incorporating
more
recently
this
Devo
graph
repository.
So
the
idea
of
Devo
graph
is
its
graph
neural
networks
on
top
of
the
pre-trained
models
and
they
they
were
nominal.
We
developed
separately.
B
There
was
a
workflow
developed
last
year
for
devil
graph
and
we're
going
to
be
working
on
updating
that
this
year
and
then
you
know
we
want
to
eventually
make
them
sort
of
fit
together,
so
you
can
use
them
run
them
together.
You
can
run
Devo,
Devo,
learn
or
devograph
in
the
main
Devon
package.
It
would
just
be
kind
of
like
one
thing,
so
it
looks
like
we
have
a
pull
request
here
and
this
is
from
Gotham.
So
this
is
his
pull
request.
So
it's
a
fixed
number
two
and
animation
for
prediction.
B
So
yeah
we'll
have
oh
say:
I
need
I,
get
I,
guess
I'll
review
it
later.
I
haven't
reviewed
it
yet
I
didn't
really
get
to
it,
but
yeah.
B
This
is
the
so
this
is.
This
is
from
like
one
of
the
things
that
gets
segmented
in
Diva,
learn
or
the
well.
The
base
of
Devo
graph
is
Devo,
learn
just
the
segmentation
part,
and
this
is,
of
course,
the
segmentation.
So
this
is
segmenting
the
membrane
edges
and
the
centroids,
which
are
roughly
the
memory
or
roughly
the
nucleus,
but
we
can't
be
entirely
sure.
There's
no
good
question.
Does
anyone
investigated.
A
B
B
Predicts
the
location
of
the
nucleus
yeah,
okay,
yeah
I
know
we
had
a
problem
with
some
of
the
data
that
they
have
and
I
don't
know
what
the
latest
data
is
but
like
in
standing
the
nucleus
getting
a
good
stain
of
like
the
nucleus,
that's
sort
of
localized,
it's
just
kind
of
like
they
put
a
stain
in
the
nucleus
and
it's
kind
of
all
around
it,
and
it's
really
hard
to
get
a
handle
on
where,
like
you
know,
the
nucleus
could
be
if
you're
talking
about
segmenting
a
cell,
it
could
be
quite
a
quite
a
bit
of
the
cell.
B
You
know
up
to
like
five
or
ten
percent
of
it
in
the
middle.
You
know.
A
Yeah,
you
know
so
it's
it's
hard
to
get
a
data
set
where
they,
where
you
can
see
both
the
nuclei
and
these,
the
cell
membranes.
B
Well,
I
mean
I,
guess
the
thing
you'd
need
would
be
a
very
accurate
nuclear
stain
or
to
like
find
a.
You
know,
have
a
nucleus
stain
where
you
could
find
a
centroid
for
the
nucleus
and
then
a
centroid
for
the
whole
cell
body,
which
is
actually
something
could
that
could
be
calculated
from
the
membrane
I.
Think
because
you
could
find
yeah
so
I
mean
you
could
just
do
it
that
way
and
then
because
yeah
I
I,
don't
remember,
I
think
the
new
the
centroids
are
usually
kind
of
a
safe
way
of
saying
nucleus.
A
C
The
model
is
not
working
well
when,
when
there
are
more
cells,
the
whole
one
is
also
not
segmenting
properly.
When
you
check
for
this
Gautam
Krishna
PR
only
when
you
run
that
slider
when
cells
are
densely
populated,
it
is
not
creating
like
this
is
an
example,
but
Gotham
Christian
have
another
video
of
it.
When
the
cells
are
densely
populated.
We
can't
get
clear
segmentation
of
I
mean
centroids
of
all
video
yeah.
B
Let
me
let
me
stop
sharing
my
screen
and
share
your
screen.
D
Okay,
so
this
is
the
actual
cell
membrane
segmental
that
is
working
and
it's
working,
and
instead
of
using
the
cell
membrane
segmental,
we
are
also
trying
to
use
the
nucleus
segmenter
on
the
Epic
data
set.
So
we
can
see
when
the
number
of
sets
are
less.
It
is
predicting
accurately,
but
after
some
time
the
predictions
are
away.
C
I
mean
see
these
centers
are
not
working.
Fine
and
segmentation
was
not
clear
at
all.
When
cells
are
densely
populated,
yeah
I
think
we
need
to
work
work
this
on
this
year,
I
mean
segment
it
clearly,
we
can
use
development
and
we
can
use
other
methods
also
to
segment
it
for
when
the
cells
are
densely
populated.
C
D
So
also
I
just
want
to
show
another
thing.
So
in
the
cell
membrane
segment
in
the
Epic
data
set,
we
can't
actually
get
a
good
prediction,
so
the
thing
is,
in
the
cell
membrane
segmental,
the
video.
D
C
A
The
green
is
mostly
intensities
of
the
same
wavelength.
B
Oh
yeah,
it's
a
fluorescent,
stain
I,
don't
yeah
I!
Think
it's
just
like
the
in
the
Epic
data
set.
It's
like
I
think
it's
like
a
nuclear.
Well,
it's
not
clearly
localized
in
a
nucleus,
but
it's
basically
a
stain
in
the
somewhat
localized
to
the
nucleus
or
like
gfp
stain
for
some
factor
and
that's
expressed
across
cells.
B
A
B
C
C
Actually,
when
I
was
reading
about
this
data,
they
have
captured
using
Ultra
electrics
I
mean
the
empty
ultraviolet
rays,
using
fluoros
in
something
chemical
called
fluorescence.
They
added
to
that.
Whatever
Clan
is
there
and
they
emitted
ultraviolet
rays
on
it,
and
then
they
captured
this
video
whatever
it
is.
Actually
in
this
process,
it
has
been
captured.
B
Yeah
I
don't
know
if
they're,
like
the
reason
we
use
the
in
the
Epic
data
set,
is
quite
old.
Actually
now
it's
probably
about
15
years
old,
so
it's
been
around
a
while,
but
it
it
has
like
the
most
complete
accounting
of
the
developmental
process.
There
are
newer
techniques
like
spam,
Imaging
where
and
we
use
some
of
those
images
for
the
other,
where
you
see
the
membranes
training
for
membranes
and
those
are
better.
Those
are
a
lot
higher
resolution.
B
The
stains
are
more
distinct,
you
know
they've
stained
the
membrane
and
the
nucleus
or
something
approximating
the
nucleus,
and
then
you
know
you
have
that
those
data,
but
we're
kind
of
putting
those
together
into
one
model
to
get
our
what
we're
doing
their
predictions
for
diesel
learning,
and
so
it's
going
to
be
kind
of
hard
to
get
things
later
on
in
development.
I
mean
they're,
probably
ways
to
do
it,
but
I
couldn't
tell
you
exactly
what
to
do.
B
B
Yeah
I
think
I
saw
this
at
one
time,
so
this
is
yeah
where
they've
kind
of
let's
say
their
kind
of
characterizing
a
lot
of
these
processes.
So
in
the
different
data
sets,
they
didn't
really
characterize
the
processes.
I
mean
they
visualize
them,
but
they
don't
really
characterize
them.
This
is
putting
together
like
an
atlas
of
C
elegans
from
you
know,
for
this
whole
process,
and
then
they
create.
So
here
we
report
an
automatic
pipeline
CS
paper
which
combines
automated
segmentation
of
fluorescently
labeled
membranes
with
automated
cell
lineage
tracing.
B
Well,
that's
what
you
have
with
the
with
the
Epic
data
set,
there's
actually
separate
metadata
that
tells
you,
the
different
cells,
like
there's,
an
identity
of
the
cells
that
that's
characterized
by
the
lineage
tree,
so
every
cell
that
gets
generated
in
the
developmental
process
when
there's
a
cell
division,
it
has
a
certain
identity,
and
so
this
is
what
they're
doing
here:
they're
putting
this
together
they're
generating
a
time-lapse
3D
atlas
of
cell
morphology,
and
so
they
have
this
I,
don't
know
kind
of
what
the
outcome
of
this
was
I.
B
Been
okay,
so
they
just
have
some
benchmarks
and
for
segmentation
results
they
have
yeah
volume.
Inconsistency
which
is
not
I
mean
we
had
this
brand
of
this
problem.
Where
there's
inconsistency
in
the
volume
of
the
cells
that
are
so
when
you
generate
a
cell
through
segmentation,
you
get
inconsistent
volumes,
you
get
lost
cells,
it's
just
what
happens,
then
you
you
know
they
have
so
they
have
this
morphological,
Atlas
I'm
just
wondering
if
this
is
something
that
is
online,
it
doesn't
I,
didn't
see
the
link
but
yeah
I
remember
this.
A
B
Yeah
they
they
do
reconstruct
the
tree
sort
of
from
these
images.
So
they
have
this
tree
sort
of
in
the
three-dimensional
space
here.
So
it
kind
of
looks
like
a
network,
but
it's
really
just
connecting
the
different
cells
of
different.
So,
like
you
know,
you
have
the
AV
lineage
here.
B
B
A
B
B
Is
yeah,
so
this
is
the
tree.
It's
interesting
that
they
have
this
cell
shape
irregularity
here,
as
we've
talked
about
I,
think
cell
shapes
in
oh
yeah,
so
you
have
these
different
shapes.
My.
A
B
B
No,
no
it's
just
that
I
mean
you
can
clearly
see
the
divisions
you
just
it's
like.
You
can
ID
the
cell
based
on
its
position.
It's
not
like
you
know.
C
elegans
embryos
are
just
unique
in
that
way,
so
you
can
just
basically
say
this
is
what
that's
all
supposed
to
be
I.
Suppose
in
some
case
you
know.
Maybe
we
haven't
done.
Maybe
the
tree
isn't
correct.
I
don't
know,
but
in
most
cases
it
is
because
it's
we've
been
able
to
observe
this
a
lot.
B
B
B
A
Look,
we
have
the
data
that
identical
twins
are
never
identical
in
humans,
right,
yeah,
okay,
so
when
you
get
twinning
in
C
elegans,
if
you
get
exactly
the
same
thing
every
time
where
something
or
is
there
a
variation.
B
B
So
this
is
I'm
gonna
go
over
this
graph
or
this
figure
I'm
going
to
move
on,
but
I
just
wanted
to
show
this
I
thought
this
is
interesting.
So
this
these
are
these
different
idealized
shapes
that
they
have
here
and
then
they
have
the
tree
and
then
what
they
find
is
they
find
these
different
cells?
This
may
speak
actually
to
your
question
when
they
segment
the
cells
for
like
these
different
cell
identities,
they
have
different
kind
of
slightly
different
shapes
and
sizes.
B
So
for
abpl,
which
is
a
cell
in
the
a
b
lineage
ABP
sub
lineage,
which
is
I,
think
it's
up
here,
avpl
you
get
these
different
shapes,
so
you
don't
always
get
the
same
shape.
It's
a
little
bit
different
each
time
and
they're
Composites
of
these
different
types
of
shapes.
So
these
are
the
sort
of
the
platonic
shapes
that
you
would
expect
a
sphere
icosahedron
and
so
forth,
and
then
these
are
like
the
combinations
here
where
you
get
different
things
that
come
out
so
you're,
just
segmenting
membrane,
okay,.
B
I
think
AB
well,
yeah
abpl
does
abpl
splits
an
abpo
a
and
a
b
p
l,
p,
I,
think
anterior,
posterior
and
they're
identical.
They
just
move
to
different
parts
of
the
one
to
the
anterior
and
one
of
the
posterior
end,
and
then
they,
oh
okay,
okay,
yeah.
B
C
B
A
B
I
think
prob
well
I
think
there
is
global
control
for
Clans,
because
you,
you
know,
you
have
cells
that
go
well.
That's
a
good
question!
Actually,
because
you
have
this
initial
anterior
posterior
Division
and
then
you
have
differentiation
there
and
then
you
have
divisions
within
those
hemispheres,
anterior,
posterior
and
they're,
going
either
left
or
right
or
anterior
posterior,
and
so
there's
local
division
and
there's
localization
there.
So
you
don't
have
things
that
move
all
the
way
across
the
embryo,
but
there
is
cell
migration.
A
A
B
Okay,
well
I'm
going
to
move
on
from
that
paper,
but
that's
thank
you
for
bringing
that
up
and
then
I
just
wanted
to
go
over
quickly
if
people
are
interested
in
this
project.
This
is
the
dsoc
project
for
this
year.
This
is
on
the
incf
neurostars,
which
is
their
discourse.
This
is
where
incf
our
sponsor
has
posted
all
the
projects
that
they're
sponsoring
this
is
3.1.
So
if
you
go
here,
you
can
ask
questions
about
the
project
and
then
I've
also
provided
a
link
to
the
slack.
B
So
if
you're
interested
in
joining
our
slack
here,
you
can
join
by
a
launch
pass.
The
slack
is
a
better
place,
probably
to
discuss
some
of
this,
but
this
is
also
a
good
place
because
you'll
get
feedback
from
incf
as
well.
If
you
need
so,
we
have
some
interests
already
from
people.
Our
NAB
actually
is
working
for
incf
as
a
sort
of
a
coordinator
for
their
gsoc
project,
and
he
was
a.
B
He
was
an
intern
in
this
group
back
in
2018.,
so
he's
he
was
one
of
one
of
our
students
and
then
he
became
involved
in
incf's
vsoc
program,
and
so
you
know
he's
spreading
the
good
word
of
Diva
worm
to
the
rest
of
the
world.
I
guess
great,
so
I
think
that's
great!
Thank
you
Botham
and
such
month
for
that
how's
Susan
doing.
E
Okay,
there
we
go
I'm,
doing
fine
I,
have
a
presentation.
E
E
E
Okay,
I'm
not
sure
this
is
Select
app
select
window
there.
Okay.
B
B
D
E
Yeah,
okay
and
then
pansegrity
the
definition
of
it
epithelial
tissues,
because
that's
what
I'm,
modeling
and
then
my
modeling
of
epithelial
tissues
and
then
some
examples
and
I
showed.
E
E
And
then
you
make
a
map
estimating
the
mechanical
properties
of
what
you've
just
taken.
Images
of.
E
All
right,
you
can
pull
off
it
as
well,
but
most
of
it
is
compression
image
Imaging
great
anyway.
This
is
Optical
towards
tomography,
setup
in
Dr,
Sharif
slab,
so
it
says,
sweat,
Laser
Source,
and
this
is
all
infrared
light
and
the
central
frequency
is
1100,
nanometers
and
I.
Think
it's
plus
or
minus
100
nanometers.
Don't
quote
me
on
that:
I
need
to
look
it
up
and
because
the
result
is
here
in
glue
yeah,
it's
that's
your
raw
data
and
then
it's
made
into
a
signal
here,
I
think
using
a
GPU.
E
E
Okay
and
this,
this
is
the
heart
of
how
it
works.
You
have
a
sensor
there
and
a
tissue
layer,
and
this
sensor
there
is
made
up
of
fine
layers
of.
E
Let's
see
gelatin
or
cellular
elastic
material,
that
kind
of
matches
matches
the
tissue
and
when
you
get
your
image,
follow
these
this
blue
layer
in
the
sensor
matches
this
blue
layer
in
the
tissue.
So
you
can
say
that
the
elasticity
of
this
little
this
this
layer
matches
this
layer
in
the
tissue.
So
this
tissue
elasticity
is
the
same
as
those
blue
blue
layer.
You've
got
here
in
your
sensor,
the
same
with
the
red
layer
and
a
red
here:
okay,.
E
E
E
And
what
is
teaching
anyway,
pseudo-colored.
E
Anyways-
and
this
is
an
example
of
something
that's
optically
similar-
the
inclusions
are
optically
similar
similar.
So
you
have
a
optical
Aquarius
tomography
image
and
you
don't
see
anything.
But
if
you
use
Optical
coherence
elastography,
then
you
can
see
the
stiffer
inclusions.
E
Resolution,
it
depends.
Let
me
go
back
here
on
Oh,
you
mean
yeah,
I,
don't
have
a
scale
bar
in
this.
They
didn't
have
scaled.
A
Arc
but
well
it
says
it's
about
a
millimeter.
What's
the
millimeter
across,
but
but
the
edges
of
the
objects
look
a
little
bit
boring.
So
obviously
there's
some
limitation
on
spatial
limitation.
E
E
That's
this
is
this
is
maybe
part
of
why
they
have
me
searching
for
tensegrity
to
see.
If
I
can
the
more
refined
model
of
this
you
might
be
part
of
it.
Maybe
I'm
supposed
to
be
making
a
better
sensor
there,
somehow
I'm,
not
sure.
E
Anyways
so
then
I
go
on
into
the
integrity
and
I
think
I've
shown
that
here
before.
E
10
segerty
is
a
structure
that
has
compression
elements
and
here
they're
in
a
green
intention
elements
in
red,
so
I've
shown
you
some
of
this
before
this
is
tensegrity
is
multi-layered
in
living
systems,
so
this
is
acting
as
a
cell
cytoskeleton
and
they
appear
to
be
a
tensegrity
and
cells,
have
tensegrity
components
and
tissues,
and
also
when
you
get
into
larger
structures
such
as
the
lumbar
spine,
the
bones
and
ligaments
act
as
an
Integrity
system.
It
doesn't
hold
that
side.
A
second
Bradley
I
want
you.
A
B
D
E
E
I'm,
just
this
isn't
long,
it's
like
30,
slides
anyway.
This
is
the
pansegrity
I'm
using
to
model
my
systems,
and
here
is
an
example
of
stress
strain
curve
that
is
for
a
living
tissue
and
they
have
sort
of
the
J
curve
where
you
have
a
toe
region.
That's
a
different
slope
from
the
elasticity
region.
They've
got
in
a
here
and
then,
of
course,
it
gets
into
a
plastic
region
where
it
starts.
E
E
A
I
can
look
it
up
many
decades
ago,
where,
if
you
put
slides,
you
know,
you
know
box
and
beat
it
up.
You
get
a
sudden
phase
transition.
A
E
I,
don't
know
but
I
don't
know,
you've
done
some
experiments
with
axolotlings,
where
you
compress
them
oh
yeah,
and
then
they
were
permanently.
E
In
some
of
tissues
there
is
hysteresis,
but
in
something
like
ligament.
Usually
there
isn't,
so
it
just
depends
on
the
tissue.
I
have
a
breast
cancer
curve
that
has
a
very
small
hysteresis.
Okay,
so
I
just
I'm
just
gonna
go
with.
It
depends
on
the
tissue
after
five
dollars.
E
That's
on
a
membrane
because
it's
a
basement
membrane
or
extracellular
Matrix,
depending
on
what
you
want
to
call
it.
So
these
can
be
flat
or
in
between
or
tall
and
polymer,
and
then
I
could
get
into
stratified
layers,
because
there
are
some
some
things
like
I
guess:
urinary
tract
where
they
have
stratified
epithelial
tissue.
A
E
Yeah
gratified
and
they
there's
Mark
cuboidal
three
layers.
E
Yeah,
it
just
depends
where
it
is
in
the
body.
I
know
in
the
ear
you
get
this
type
of
a
thing
where
you
have
actually
bone
here,
which
is
different
from
skin
yeah
and
I'm.
Not
skin
tends
to
have
well
it's
complicated.
E
No,
it
has
some
of
these
well
I,
don't
know,
I
get
which
one
you
would
say
skin
is
I,
think
it
depends
on
where
the
skin
is,
but
it's
at
least
five
layers
thick
and
I
had
a
an
image
of
where
each
of
these
was
in
the
body
and
I.
E
This
is
the
dermis,
and
this
is
our.
This
is
the
epidermis
pardon
me,
and
this
is
the
dermis
underneath.
Oh
underneath,
okay
I
would
think
so.
Yeah
Okay.
A
The
reason
I
asked
is
because
I
used
to
study
melanoma
and
one
of
the
big
factors
was
that
if
you're
quite
early,
you
could
cure
the
patient
95
percent.
But
if
this
is
melanoma,
parts
are
penetrated.
E
Ecm
extracellular
Matrix
this
thing
yeah
the
right
basement
membrane.
Then
then
it
causes
problems.
Does
this
mean
the
blood
vessels
don't
go
above
that
layer?
E
It
looks
like
there's
blood
vessels
in
the
dermis
and
not
in
the
epidermis
yeah.
It
does.
D
E
E
I,
don't
know
I'm
not
sure
how
these
are
kept
alive.
They're
not
exactly.
Some
of
them
die
off
at
the
top
to
help
keep
your
skin
secure,
yeah,
and
if
you
rip
that
off
it
hurts
this
upper
layer.
E
A
E
E
I'm
not
sure
but
I
what
they
were
showing
was
that
there
was
acting
at
the
apical
end
here,
and
then
they
showed
that
there
was
microtubules
and
intermediate
filaments
holding
the
nucleus
to
other
parts
of
the
cell,
to
the
desmosomes
and
to
the
integrals.
These
are
adult
cells,
yeah
yeah.
These
are
cells
that
aren't
changing
that
much
they're
adult
cells.
Yes,
so
from
the
previous
slide,
you
can
see
that
there
are
microtubules.
E
B
Yeah
sure
I
imagine
in
development
it's
much
more
Dynamic
like
the
the
network,
is
shifting
a
lot
more,
especially
yeah
when
cells
are
dividing
and
things
like
that.
My
question,
I
guess,
is
in
these
adult
cells.
To
what
extent
can
you
regard
vehicle.
A
E
Okay,
well,
there's,
certainly
any
people
acting
active
eyes
and
an
act
of
myosin
at
these
Junctions.
E
Might
okay
sure
I,
don't
I,
don't
mind!
This
is
interactive,
minus
and
there's
your
myosin
and
here's
your
actin
and
actinin
I.
Don't
know
how
you
pronounce
that
without
my
passion,
focal
adhesion
which
in
turn
is
attached
to
your
integrands
JX,
this
extracellular
Matrix.
E
So
what
you
need
for
your
well,
your
model,
you
have
to
have
pre-stress
in
your
cytoskeleton
and
you
need
a
and
the
extracellular
Matrix
adhesion
through
through
these
integrins.
E
And
it
was
pointed
out
to
me
by
someone
named
Stephen
Levin
that
I
was
forgetting
Kai
and
I
was
I'm
aware
of
this.
You
know
relaxation
and
creeping
tissues,
and
that's
often
measured
rather
than
this
stress,
strain
diagram
and
tissues.
E
E
So
it
could
be
microtubules
as
well,
because
there's
a
stiff
element
as
well
as
the
strings
and
then
I've
got
posts
around
the
outside
and
when
I
attached
things
together,
like
the
posts
to
the
cell
model,
I
tried
a
plane,
attachment
and
I
tried
a
tensegrity
type
attachment
and
guess
what
the
tensegrity
type
attachment
works
better
and
the
plane
attachment
I
got
minus.
What
is
that
six
times,
ten
to
the
minus
five?
E
But
if
you
have
a
tensegrity
attachment,
you
have
6
times
10
to
the
minus
three
Newton
was
that
it
will
take
before
it
falls
apart
and
then
six
Newtons
in
tension
that
that
that's
quite
an
inferences
yeah,
these
factors
of
10.,
so
no
100,
I,
guess
in
in
this
in
the
compression
foreign
ly,
get
this
to
work
with
three
cells,
and
this
structure
tends
to
put
the
largest
stresses
on
the
outside
of
itself
on
the
post.
B
Yeah,
it's
fine,
probably
about
25
minutes.
30
minutes!
This
is
fun.
It's
great!
That's
great!
Actually,
I
had
a
question
about
the
forces.
If
you
like,
I,
think
it's
27
so
slide.
27.
E
B
Are
yeah
so
you
you
say:
compression
Point
forces
in
Z,
Direction,
Max
and
then
negative
point:
zero,
zero,
zero,
six
Newtons,
so
negative
Newtons
would
be
like
compression
force.
Attention
force
would
be
positive
or
yes,.
E
Okay,
so
it
it's
less
flexible
in
compression
than
it
is
in
tension,
but
that's,
maybe
partly
because
they're
the
model
well
really
attached
to
its
base
and
his
base
is
not
flexible.
B
B
Yeah
so
it
goes
down
and
you
can't
push
it
down
very
much,
but
you
can
pull
it
up
a
lot
so
just
from
yeah.
B
E
A
E
A
E
E
B
E
E
Stem
cells
are
amazing,
Timeless
stem
cells
into
either
a
neural
tissue,
which
is
blue
Mayo,
which
is
muscle.
Maybe
this
green
and
an
Osteo
which
is
bone
in
this
orange
orange
color
here,
depending
on
the
substrate
elasticity,
great
for
your
basement
membrane
or
your
extracellular
Matrix,
depending
on
what
you
want
to
call
it
yeah,
that's
very
unsigned.
Scientific.
Do
I
need
to
be
more
excited
when
I
present
this.
B
E
E
B
E
E
A
Oh
okay,
see,
if
you
can
I
mean
I'd,
be
especially
interested
if
anyone
you
know
a
lot
of
businesses,
modeling
balls
of
cells
and
the
cells
are
usually
of
uniform
type.
A
D
A
Base
control
at
least
two
cell
types
from
for
the
modeling
yeah,
like
the
onion
structures,
you're
on
the
old
found
the
instructions
from
the
differential
adhesion
days.
E
No
yes,
differential
adhesion
and
the
cells
tend
to
go
around
each
other.
A
A
Oh
okay,
it
was
all
based
on
differential
adhesion
with
the
the
physicists
are
getting
much
more
complicated
models,
but
they.
E
Only
go
to
one
cell
type,
oh
cool,
all
right.
What's
the
differential
vacation
and
surface
tension
and
see
if
somebody's
got
something
going
with
more
than
one
cell
type
right,
okay,
yeah,
yeah,.
C
B
A
After
after
cleaned
up
the
slides
a
bit
like
putting
in
an
explanation
for
GT,
why
don't
you
give
a
copy
of
the
ground
we
put
in
the
archives,
yeah.
B
E
Okay,
I'm
I'm,
not
sure
what
Dr
Sharif
would
think
of
that
or
if
it
would
be
okay,
if
I
could
share
it,
I'll
ask
him
if
I
can
share
it
and
then
I'll,
I'll,
okay,
fine,
yeah,
yeah
I
need
to
ask
him,
and
he
says
he
has
a
lot
of
Notions
about
things.
A
B
Right,
yeah,
well,
yeah,
okay,
I!
Think!
That's
it
for
today,
any
other
things
you
need
to
mention,
or
yes,.
D
B
So
I've
put
this
into
this
is
the
simulation
that
I've
been
working
on
and
I
put
this
into
the
main
paper
I'm
not
done
yet,
but
I'm
I'm
getting
there.
So
this
is
in
the
main
paper.
B
This
is
the
paper
that
we
have
been
working
on
and
you
know
I
did
an
earlier
contribution
to
this
extend
some
things
at
droplets:
oil
droplets
toy
model
for
the
origin
of
life,
so
the
whole
point
of
this
exercise
is
simulations
is
to
model
the
origin
of
Life
and
so
from
spontaneous
by
spontaneous
means,
and
so
there's
some
literature
review
here,
a
model
of
Auto
catalysis,
which
we
talked
about
last
week.
B
Let
me
explained
what
that
was
and
then
getting
into
you
know
some
tables
for
that
and
then
coming
down
to
the
so
yeah
I
put
in
or
there's
a
part
here
on
applying
breitenberg
vehicles
or
greatenberg
like
vehicles
to
this
problem,
so
I.
Actually,
this
is
the
part
that
I
contributed
earlier,
where
you
would
create
breitenberg
vehicles.
B
Breitenberg
vehicles
are
these
Toy
models
where
you
would
put
together
a
very
simple
organism
with
the
nervous
system,
but
it's
using
this
analog
of
vehicles
with
wheels
and
and
a
body
and
and
light
sensors,
and
things
like
that.
So
this
is
sort
of
a
diagram
of
the
model.
B
You
have
these
viable
and
inviable
rate
bird
Vehicles.
You
can
put
together
the
pieces
and
the
idea
is
that
they
assemble
like
a
like.
You
know,
assembling
a
747
and
a
junkyard
using
a
tornado,
it
just
kind
of
comes
through
and
it
pushes
the
pieces
together.
And
if
you
do
this
enough
times,
you
get
like
things
that
are
functional
so
that
that
was
I.
Think.
B
So
yeah,
so
this
is
just
kind
of
I
worked
this
out
as
a
set
of
viable
and
inviable
vehicles,
giving
examples
here
using
the
toy
model
and
then
coming
up
with
so
having
like
a
parts
list
with
a
number
of
different
states
that
the
parts
list
can
take
and
then
the
probability
that
that
will
result
in
a
spontaneously
functional
vehicle.
B
There
are
actually
a
lot
of
configurations
which
end
up
being
viable,
and
this,
of
course,
is
just
based
on
my
toy
model,
so
it
could
vary,
but
yeah
yeah,
it's
it's
I,
think
it's
yeah
I
think
it's
surprisingly
high,
so
yeah
so
anyways
doing
using
a
focused
methodology.
You
know
you
can
get
that
result
and
then
the
second
part,
which
is
the
part
that
I've
been
working
on.
So
this
is
computer
simulation
of
a
vesicle,
so
I
need
to
change
some
of
this,
but
this
is
basically
the
parameter
list.
B
I
need
to
finish
this
up,
I
think
if
this
is
most
of
them.
This
is
just
taken
from
the
I'm
running
this
in
scilab,
which
is
sort
of
a
version
of
Matlab,
that's
open
source.
So
it's
it's
that
type
of
language.
It's
a
symbolic
language.
It
does
a
lot
of
Matrix
calculations
and
that
sort
of
thing.
So
this
is
just
setting
up
the
parameters
and
then
the
graphs.
B
This
is
the
idea
of
this
molecular
world
with
a
lot
of
infinite
number
of
molecules
of
certain
a
certain
type
and
then
having
a
cell
of
a
certain
diameter
where
these
molecules
get
put
into
that
cell
or
they
they're.
The
cell
membrane
is
the
vesicle
membrane,
is
more
or
less
permeable,
but
you
can
only
fit
so
many
molecules
into
the
cell
at
a
time
and
so
the
chances
of
getting
K
types
is,
you
know,
as
you
increase
the
diameter
and
increases.
B
B
Can
you
bootstrap
this
in
a
very
small
vesicle
from
something
that
has
fewer
types
to
more
types,
so
I
I
showed
this
last
week
and
the
code
and
I've
been
working
on
this
I've
been
working
on
as
sort
of
another
version
of
this,
where
there's
like
this
longer
duration
of
simulation.
So
this
is
a
simulation
for
10
to
the
fifth
days.
B
So
everything
in
terms
of
this,
this
graph
actually
shows
how
this
goes
for
like
two
two
days
where
there's
a
day
night
cycle
and
how,
basically
you
get
a
lot
of
variation
during
the
day
at
night.
You
know
things
kind
of
close
off
and
and
there's
this
resting
period,
and
then
you
get
this
fluctuation
back
and
then
so,
but
that
doesn't
necessarily
result
in
any
sort
of
increase
in
the
number
of
molecules.
B
It's
just
very
static
for
a
vesicle
of
a
certain
size,
but
one
thing
you
can
do
is
to
actually
I
did
two
things
in
this.
So
this
is
the
first
piece
of
code.
That's
not
really
that
important
right
now.
The
second
piece
of
code
is
the
more
important
thing.
This
is
the
one
that
I
I
was
working
on
this
week.
This
is
for
10
to
the
five
days,
so
I
can't
run
I
can't
run
10
to
the
sixth
like
or
maybe
could
run
10
to
the
sixth.
B
But
it's
it's
I'm,
hitting
the
upper
bounds
of
my
computational
limits
here
for
what,
where
what
resources
that
I
have,
because
you're
generating
a
lot
of
characters
or
a
lot
of
variables
for
this,
but
so
you
get
up
like
you,
can
just
take
10
to
the
five
days.
Each
point
is
one
day
that
has
like
a
mixed
day,
night
cycle.
So
it's
like
you
know,
a
half
of
the
time
is
day.
Half
the
time
is
night
and
you
can
it
can
change
that
in
the
model.
B
What
we
do
is
we
draw
from
a
mixed,
uniform,
poisson
distribution,
so
that
cloud
I
showed
you
in
the
previous
figure
that
cloud
it
it
had.
It
like.
It
was
a
a
bunch
of
Articles
or
a
bunch
of
molecules
of
a
uniform
distribution.
B
So
that
means
that
any
one
type
is
is
just
as
likely
as
any
other
type.
And
so
that's
that's
what
you
have,
but
in
this
case
some
types
are
a
little
bit
more
likely.
So
you
have
this
song
process,
that's
mixed
with
the
uniform
distribution.
The
poisson
process
introduces
this
non-regularity
where
it
brings
in
more
types
at
different
frequencies
and
in
the
in
the
base.
You
know
the
base
sampling
pool.
So
that
means
that
the
higher
this
arrival
rate
is
the
more
the
Richer
your
sample
becomes
in
terms
of
non.
B
B
It
actually
increases
the
chances
that
you
get
36
types
for
any
one
vesicle
size
and
then
I
added
a
memory
component,
which
is
like
basically
carrying
over
the
content
from
the
previous
day
at
a
certain
rate,
and
so
you
can
change
this
number
I
put
it
at
0.05,
which
means
it's
retaining,
like
maybe
1
20th,
of
what
it
remember
what
it
had
from
the
previous
day
and
it's
carrying
over
to
the
next
day.
And
so
this
is
the
code
for
that.
B
There's
a
loop
that
kind
of
goes
through,
and
actually
there
are
two
Loops
here
that
one
for
the
sampling
from
this
distribution
and
then
the
other
loop
being
the
memory
that
carries
over
and
then
you
could
plot
them
and
I
plotted.
B
So
this
this
is
what
I
got
so
this
this
mixed
distribution
I
was
able
to
get
it
to
have
sort
of.
You
know
it's
it's
for
this
vesicle
size
where
you
never
get
above
36
kinds,
but
in
this
case
I
I
gave
it
this
mixed
distribution
of
draw
from
it's
actually
exceeding
36.
occasionally.
So
this
red
line
is
36
types
and
it's
exceeding
that
occasionally
in
this
series.
So
this
is
a
pretty
long
Series.
This
is
10
to
the
fifth
days.
You
can
see
that
every
once
on.
B
B
It's
ten,
is
it
10
to
the
fifth?
Well
exactly
yeah
yeah,
but.
A
A
36.,
there's
a
get
rounded
down
it.
B
B
A
B
Yeah,
it's
just
really
hard
to
visualize.
It
like
I,
can't
pull
it
apart,
but
actually
that
that
brings
up
an
interesting
point
because
I
can
take
out
pieces
of
it
and
I
did
that
for
the
second
graph.
So
this
is
the
first
100
days
and
as
you
can
see
here,
you
can't
really
see
it
here
at
all,
but
it
doesn't
exceed
36.
It
actually
maybe
gets
up
to
about
34.
and
then
or
33..
So
you
can
see
that
there's
variation
over
time.
B
You
can
see
it's
really
non-stationary
when
you
add
in
the
memory-
and
you
add
in
this
mixed
sample,
yeah.