►
Description
Data Sources for Developmental Algorithms, GSoC Question and Answer, project updates. Jamming Phase Transitions, Unjamming, and Cell-cell Signaling. Jamming and 1-D Automata. Attendees: Susan Crawford-Young, Karan Lohaan, Jiahang Li, Richard Gordon, and Bradly Alicea
A
B
D
D
C
D
D
D
C
C
Otherwise,
I've
been
talking
to
some
of
you
about
the
gsoc
projects.
I
know
I've
been
talking
about
jihang
about
the
cl,
where
or
sort
of
the
gnn's
approach,
and
he
has
an
approach
that
he'd
like
to
write
up.
So
I'm
going
to
look
over
your
proposal
later
on
today,
I'll
give
you
some
feedback
and
then
the
submission
actually
is
tomorrow
for
this.
So
people,
I
think,
are
kind
of
hitting
the
hitting
the
deadline,
but
that's
okay.
C
We
talked
about
it
and
and
if
you
know
I,
I
know
that,
like
a
lot
of
people
are
planning
to
submit.
So
if
you
don't
get
selected
for
it,
you
know
I'm
still
open
to
working
with
you.
You
know
over
the
summer
and
working
on
a
project
and
getting
you
know
something
really
nice
as
a
product
of
that
now.
That
being
said,
we
have
a
lot
of
data
sets
that
you
can
use,
and
so
this
is
an
opportunity
for
people
to
use
their
skills
and
build
a
project
out
of
it.
C
And
then,
if
you
get
selected
for
gshawk
all
the
better
I've
talked
to
koran
as
well.
He
sent
me
something
I'll
review
that
hurry
krishna
I've
reviewed
his
things,
so
that's
that's
pretty
good.
We
have
a
number
of
people
also,
I
think
we're
maybe
submitting
things,
but
I
haven't
really
seen
it
draft
from
them.
So
we'll
see
what
the
we'll
see
what
we
can
we'll
see
what
happens
from
that
yeah.
C
C
C
So
we
have
cell
tracking
data
where
they
label
the
nuclei
with
some
fluorescent
marker
and
then
they
take
microscopy
images
of
the
organism.
Usually
the
embryo
and
all
the
cells
are,
the
nuclei
have
some
marker
in
them
and
they
contract
those
nuclei
under
a
uv
microscope,
and
they
can,
you
know,
follow
that
through
time,
so
they
just
take
images.
C
You
know
not
so
many
images
per
second
and
then
they
have
these
slices,
these
focal
plane
slices,
which
means
that
they
go
through
the
specimen
and
they
go
through
different
focal
planes
and
they
get
those
images.
So
you
end
up
with
a
stack
of
images
through
the
embryo
at
different
sort
of
levels.
You
know,
so
you
can
see
different.
C
You
know
if
you
go
from
like
the
top
to
the
bottom,
so
you're
going,
you
know
from
the
dorsal
to
the
ventral
in
anatomical
terms
and
you're,
getting
all
these
slices
and
you're
getting
images
from
those
different
places.
So
sometimes
cells
are
towards
the
top,
sometimes
towards
the
bottom.
You
can
figure
out.
You
know
you
can
get
all
the
information
about
the
cells
that
way,
but
you
also
have
a
time
series.
C
So
you
have
you
know
a
number
of
frames
acquired
per
second
and
then
you
just
simply
stitch
those
together
as
like
a
time
you
can
treat
it
as
a
time
series.
You
can
also
just
look
at
like
different
cells
when
they
emerge
and
look
at
their
position
and
do
it
that
way.
So
you
don't
need
to
have
a
time
series,
but
you
can
also
do
like
a
dynamic
analysis
of
it.
C
Now
there
are
a
number
of
data
sets
that
exist
where
they
don't
have
the
images
but
they've
extracted
the
data
from
the
images
and
they've
created
like
a
data,
a
raw
data
set
of
points
and
the
points
are
labeled
by
coordinates
so
when
they
take
an
image
their
coordinates.
You
know
these.
These
points
are
different,
coordinates
on
the
image
and
then
they
give
you.
C
X
would
be
the
anterior
posterior
front
to
back,
and
then
the
sides
would
be
y,
and
so
that's
that's
how
they
usually
do
this
and
that's
how
the
data
is
presented.
So
you
can
get
it
out
of
the
raw
data
or
you
can
get
it
out
of
the
numeric
data
and
then
what
we've
done
in
the
past
in
the
group
is
to
transform
that
into
some
sort
of
coordinate
system
of
that
that
you
can
use
to
compare
across
different
images.
So
the
raw
data-
just
basically
has
you
know
some
coordinates
some
coordinate
labels.
C
They
don't
really
make
much
sense
if
you
compare
them
across
images,
but
if
you
use
a
sort
of
an
averaging
for
the
sort
of
like
a
centroid
for
the
entire
image
or
something
like
that,
you
can
usually
get
a
nice
coordinate
system.
That's
consistent!
There
are
other
ways
to
do
it
as
well.
But
that's
that's
what
you
need
to
do.
Those
are
the
steps
you
need
to
do
to
get
some
of
the
cell
tracking
data
in
so
we've
done
a
lot
for
c
elegans.
C
We've
there's
some
for
drosophila
they're,
like
they're
different
model
organisms
that
people
use.
We
can
get
cell
tracking
data
for
that
for
zebrafish,
and
then
we've
also
worked
with
other
organisms
such
as
diatoms.
So
we
have
data
from
diatoms
which
are
the
single
cell
algae.
Sometimes
they
live
in
colonies
and
we've
been
analyzing
the
colonies.
C
There
are
other.
You
know
organisms
that
we've
used,
like
you
know,
more
exotic
ones
that
aren't
really
usually
model
organisms,
they're,
very
good
images
of
spiders
spider
embryos,
they're
also
some
good
images
of
plant
embryos,
and
so
we
don't
have
any
of
those
in
our.
We
haven't
worked
with
those
too
much,
but
those
exist
hello,
perron.
How
are
you
so?
Let
me
go
over
some
of
the
hey
guys,
hi.
C
So
the
first
one
is
this
systems
biology
database
system,
science
of
biological
dynamics
database
to
be
specific,
so
this
is
run
out
of
reikin
in
japan
and
they've,
and
this
actually
ken
ho
who
used
to
be
used
to
attend
our
meetings
from
time
to
time.
He
was
running
this
database
and
I
I
don't
know
if
he's
still
involved
with
it,
but
it's
run
out
of
reikin
in
japan.
They
do.
They
have
a
lot
of
different
microscopy
images
and
they
have
this.
C
C
If
you
want
to
learn
more,
you
can
read
this
paper
ssbd
a
data
set
of
quantitative
data,
a
spatiotemporal
dynamics,
a
biological
phenomena,
and
that
paper
is
about
everything.
I
just
told
you
about
like
cell
tracking
and
things
like
that,
because
a
lot
of
these
data
sets
are
cell
tracking
data
sets
more
or
less.
C
C
This
is
the
nuclear
division
dynamics
and
c
elegans
wild
type
embryo.
So
this
is
another
type
of
image
that
you
can
get
so
there.
You
know
we
look
at
some
of
the
images
that
we've
we've
seen
in
the
group.
You
know
we
have
this
sort
of
the
ember
whole
embryo
data,
where
you're
looking
at
cells
dividing
and
this
doesn't
really
label
the
the
nucleus,
but
the
nucleus
would
be
somewhere
inside
the
cell
body
and
you
know
if
you're
looking.
C
This
is
a
dic
image,
so
you
don't
see
fluorescence
in
it,
but
you
can
see
fluorescence
in
some
of
these
other
images.
This
is
a
calcium
imaging
protocol
and
then
this
is
another
type
of
imaging.
This
is
a
molecular
dynamics,
imaging
so
there's
a
lot
of
potential
information.
You
can
get
out
of
microscopy
images
if
you're
working
on
a
method
for
machine
learning
or
a
method.
For
you
know
graph
neural
networks.
C
C
C
We
could
just
take
some
image
data
plug
it
in
and
then
get
a
some.
You
know
allow
it
to
segment
the
cells
first
of
all
segment,
the
edges
of
the
cells
and
then
detect
that
centroid,
which
was
the
labeled
nucleus,
and
then
we
could
output
the
data
there.
So
we
could
output
like
the
dimensions
of
the
of
the
cell
and
the
nucleus,
and
so
that
data
could
be
transformed
into
numeric
data.
But
to
do
that,
usually
you
need
to
have
a
good
training
set.
Well.
C
Like
you
know,
if
you're
doing
with
these
graph
neural
network
embeddings,
for
example,
you
might
take
an
embryo
and
you
might
be
interested
in
doing
a
different
thing
with
it
encoding
some
of
the
network
dynamics,
for
example-
and
I
know
that
we've
talked
a
little
bit
about
that
with
some
of
you
on
how
that
might
be
done.
So
this
these
are.
C
This
is
one
source
of
microscopy
data.
This
has
the
advantage
of
being
these
are
published
in
papers.
So
some
of
these
data
sets
are,
you
know
they
may
be
about
10
years
old,
but
they're.
You
know,
good
data
sets
they're,
they
were
published
in
nature
and
science,
and
you
know
they.
They
provide
a
good
and
they're,
probably
newer
ones,
but
the
thing
is
is
that
we
want
good
data.
D
C
B
C
C
C
So
these
are
process
data
sets.
If
you
go
to
diva
worm,
github.com
diva
worms,
backslash
diva
worm
and
then
the
last,
the
second
diva
worm
is
capitalized.
So
it's
it's
case,
dependent
you'll,
see
that
we
have
all
these
folders,
so
we
have
different
types
of
data
sets.
So
this
is
we
worked
with
what
they
call
a
c-squirt
which
is
obsidian
an
obsidian
and
it's
a
type
of
organism.
C
It's
a
marine
invertebrate
that
has
a
number
of
different,
distinct
life
cycles,
and
you
know
it
goes
from
like
being
this
four-fold
symmetric
embryo
to
being
this
like
tadpole
almost
and
then
it
ends
up
being
session
later
in
life.
It
settles
down
on
the
bottom
of
the
ocean
and
it
sits
there
and
filters
the
water
for
its
food.
C
So
this
has
a
very
diverse
life
history,
but
this
this
data
is
from
the
early
embryonic
stage
and
the
reason
that's
important
is
because
this
is
another
organism
that
has
a
very
well
characterized
early
development.
So
this
is
like
c
elegans
c
elegans
has
you
know
a
very
distinct
early
development?
You
can
track
every
cell,
but,
more
importantly,
you
know
the
history
of
that
cell.
You
know
what
the
lineage
tree
looks
like.
So
I
was
talking
about
this
with
jiahung
yesterday
on
slack
that
you
have
c
elegans.
C
Has
this
very
distinct
type
of
development,
where
every
cell,
you
know
what
the
fate
of
that
cell
is
going
to
be
it's
deterministic,
and
so
because
of
that
we
can
track
the
cells
we
know
which
cell
it
is.
We
know
what
the
identity
is
and
we
know
what
it's
going
to
become.
So
it's
going
to
divide
into
two
cells.
We
know
what
those
cells
are
going
to
be,
and
we
know
it's
deterministic
across
c
elegans,
with
some
exceptions
for
like
mutation
that
they're
all
basically
going
to
end
up
in
the
same
place.
C
C
This
is
a
database
for
these
obsidians
and
it
allows
you
you
know:
there's
some
genome
annotations
here,
there's
some
species,
so
I
think
they
have
some.
They
have
some
data
raw
data
sets
for
like
the
lineage
tree,
like
I
think
the
anatomy
tab,
but
they
also,
I
think
they
have
there's
some-
and
I
don't
know
if
it's
in
this
database
there's
some
microscopy
data.
C
So
there's
an
anatomy
tab
here.
Actually
that
kind
of
goes
into
some
of
this
anatomy
and,
as
you
can
see
from
the
anatomy
here,
it's
four
fold
symmetry.
So
when
a
cell
divides
you
have
these
four
poles
here
and
they're
symmetrical
and
that
the
cells
are
organized
in
the
same
way.
So
it
kind
of
you
know
they
have
like
we
as
humans
or
c
elegans
or
axolotl.
C
That's
two-fold
symmetry
now,
four-fold
symmetry
is
where,
if
you
had
an
arm
coming
out
of
your
front
coming
out
of
your
back
and
they
were
all
identical,
that
would
be
four-fold
symmetry.
Actually,
you
probably
wouldn't
have
a
front
and
a
back
anymore.
You'd
have
like
four
sides,
but
that
you
know
so.
C
I
don't
know
why
the
data
set's
not
labeled
on
this,
but
this
is
the
one
of
the
the
data
sets
that
they
have
on
ssbd
in
in
terms
of
the
microscopy
images.
This
is
just
the
raw
data
of
this.
This
is
like
the
numeric
raw
data,
so
there's
a
lot
of
numeric
raw
data
here,
much
more
than
what
we
have
for
obsidians,
and
this
is
like.
I
think
this
series
is
from,
like
the
single
cell,
all
the
way
out
to
almost
to
the
hatch
stage
of
you
know
where
the
worm
hatches.
C
It's
you
know,
the
the
embryo
keeps
growing
in
terms
of
cells
and
then
there's
a
hatch
where
it
comes
out
of
the
egg
and
then
there's
some
post
postnatal
development,
which
is
where
they,
you
know
some
there's
more
cell
division,
but
it's
not
essential
to
the
function
of
the
organism.
So
you
know
they
have
this
sort
of
the
cell
division
that
happens
outside
the
egg,
but
this
this
data
set
gives
you
everything
within
the
egg.
C
Then
there's
like
cell
birth
and
death
timing,
data
for
c
elegans.
This
is
where
we
have.
We
can
take
those
cell
tracking
data
and
annotate
them
with
timing
differentiation
times
for
each
cell,
so
you
have
like,
for
example,
you
know
a
time
a
cell
is
observed
a
time.
The
cell
divides
timing
of
other
things.
C
C
C
I
can
tell
you
exactly
what
those
are
offline,
because
I
have
to
see
what
they
are.
I
mean
I
have
to.
You
know,
go
back
and
look
because
I
haven't
looked
at
this
in
a
while,
but
the
point
being
is
that
I
can
you
know
we
have
a
lot
of
data.
That's
been
processed,
there's
a
lot
of
data
available
and
if
you
need
data,
that's
not
a
problem.
C
Finally,
there's
this
devo
zoo-
and
this
is
something
I
mentioned
in
the
group
and
we
haven't-
maybe
we
haven't
gone
through
it
enough,
but
just
one
more
time
for
people
who
may
not
be
aware.
This
is
d.
This
is
this
view
on
github
link.
This
just
takes
you
back
to
this,
so
that's
that's
what
that
is,
and
but
we
also
have
different
data
sets
organized
by
organisms,
so
this
is
for
nematodes.
C
C
We
also
have
microscopy
movies,
so
these
spin
images,
these
spin
data
sets,
are
actually
really
good
for
finding
you
know,
being
able
to
segment
cells
and
finding
nuclei
if
you're
doing
some
sort
of
machine
learning
technique
to
segment
cells.
These
are
very
easy
to
work
with,
and
the
nuclei
as
well.
C
Some
of
these
other
data
sets
are
not,
you
know,
are
maybe
a
little
bit
more
challenging,
but
they
have
different
types
of
features
in
them.
Then
this
is
the
for
the
c-square
for
the
acidian.
We
have
micro,
microscopy
movies
and
we
have
this
data
set
of
the
cells
and
their
identity
and
they're.
I
think
some
sort
of
timing
data,
I'm
not
sure
if
it's
by
seconds,
but
it's
definitely
useful
for
organizing
the
lineage
tree.
For
that
then
there's
drosophila.
We
have
a
couple
of
data
sets
for
that.
C
C
C
Then
we
have
even
something
called
morphozoans,
which
are
artificial
organisms,
and
we
have
some.
You
can
run
a
simulation
of
these
morphozones
and
see
what
they
look
like
and
then
some
other
things
as
well.
So
that's
our
devo
devozu,
and
so
I
actually
I'm
looking
towards
improving
upon
that
in
the
coming
months,
but
I
haven't
had
time
in
the
recent
past
to
do
anything
with
it.
C
I'll
probably
talk
about
it
in
coming
meetings,
but
I'm
thinking
that
there's
you
know,
I
think,
there's
some
use
in
terms
of
model
organisms
and
there's
a
lot
of
interest
in
sort
of
comparative
studies
where
people
take
different
model
organisms
and
look
at
them
and
look
at
the
differences
between
them
and
the
similarities,
so
that
there's
an
opportunity
there
and
of
course,
if
you're
working
on
something
you
know
some
sort
of
software
platform
that
can
be
incorporated
as
well.
We
haven't
full,
we
have
the
actual.
C
We
have
the
platforms
for
the
different
machine
learning
data
sets.
But
you
know
we
don't.
We
have
those
in
a
different
in
a
different
area
of
the
website,
so
it's
like
diva
worm
ai.
So
we
have
a
library
there,
but
that's
something
that
you
know
might
be
integrated
more
closely
with
the
model
organisms.
F
C
Yeah
so
diva
learn
has
a
lot
of
the
development
of
that
was
based
on
spim,
like
the
spin
microscopy,
which
is
the
it's
a
type
of
light
microscopy
and
we're
using
specifically
a
lot
of
fluorescent
images.
So
the
fluorescent
images
will
have
like
the
it'll
have
a
label
on
the
nucleus
and
then
it'll
also
have
like
labels
along
the
edges
of
the
cell,
and
so
you
can
actually
use
those
type
of
images
to
segment
cells
and
define
the
center
of
the
cell.
C
You
know
so
you
have
like
this
tracking
of
the
center
of
the
cell
and
you
can
actually
segment
the
cell
out
now
there
are
other
automated
tracking
methods
that
have
been
used.
The
data
sets
that
I
showed
you
before
about
the
c
elegans
cell
tracking
data
sets
those
are
actually
also
used
in
diva
learn
and
that
was
used
as
sort
of
like
a
pre-training
step.
So
the
idea
was
that
we
had
like
you
know.
C
We
would
take
the
data
from
this
automated
cell
tracking
and
put
that
in
as
a
pre-training
step,
so
that
model
would
know
what
that
looked
like
and
those
images
are.
You
know:
they're
they're,
good,
but
they're
also
pretty
raw
they're,
not
very
high
resolution,
and
so,
but
they
give
you
information
about
the
where
the
cells
are
in
the
embryo
and
then
we
also
use
the
spin
data
to
sort
of
confirm
it
at
a
higher
resolution.
C
So
we
ended
up
with
this.
You
know
pre-trained
model
that
had
a
pretty
good
sense
of
like
how
to
segment
a
cell
when
it
saw
something
in
a
c
elegans
embryo
take
out
the
centroid
and
then
use
that
as
a
center
point,
and
then
you
know
put
you
know,
give
you
automated
output
of
the
numeric
data,
so
yeah
we
use
those.
Two
data
sets
well
c
elegans
bim
data
sets,
I
mean,
there's
probably
more
than
one.
I
think
we,
I
can't
remember
exactly
which
one
we
used,
but
those
are
higher
resolution.
C
So
we
have
like
these
two
data
sets
that
we
use
automated
cell
tracking
and
then
the
spam
data
set,
which
was
segmented
by
the
system
and
then,
if
you
use
diva,
learn
it
works
best
with
those
sorts
of
high
resolution
images.
If
it's
a
lower
resolution,
image
it'll
still
work,
but
it's
not
as
accurate.
C
F
Yeah
understanding
points,
and-
and
in
my
proposal
I
I
think,
maybe
our
project
to
use
the
development,
because
we
need
to
build
the
graph
structure
from
the
given
videos
and
images
and
we
need
development
to
extract
some.
Some
information
from
these
images,
such
as
the
coordinates
and
such
as
the
centroids,
such
as
the
distance
between
the
centroids,
and
it
will
help
us
with
building
the
graph
yeah
so
yeah,
yeah
yeah.
F
That's
why
I
asked
this
question
and
I
have
another
question:
is
that
actually,
in
my
proposal,
I
designed
the
project
which
can
construct
some
high
level
embedding
of
graphs
using
such
as
graph
sage
and
gcn
jt,
some
other
graphene
network
models,
but
we
haven't,
because
I
have
a
little
understanding
of
the
biological
analysis.
So
I
have,
I
didn't
know
which
task
what
task
you
will
focus
on.
F
I
was
doing
some
works
related
to
graphene
natural.
I
mean
the
algorithm
part
of
june,
we'll
focus
on
some
tasks
such
as
the
node
classification,
graph
classification
and
some
link
prediction,
and
this
test
will
help
us
to
evaluate
the
how
the
models
perform
and
how
our
methods
perform,
but
if
all
projects
only
generates
embeddings
without
without
using
tasks
to
evaluate
the
embedding,
I
think
in
a
way,
maybe
we
lack
a
method
to
evaluate
our
projects
and
our
methods.
F
Yes,
because
there
are
a
lot
of
ways
to
build
graphs,
to
build
note
and
each
features,
but
if,
if
we
don't
have
appropriate
tasks
to
evaluate
how
the
the
model
performed
or
is
the
no
feature
and
each
feature
good
or
not?
Maybe
yeah.
That's
my
point.
Yes,
I
think
maybe
tests
are
important,
but
I
didn't
mention
that's
in
the
in
the
project,
because
I
didn't
think
we
have
enough
time
to
cover
the
whole.
The
all
of
the
points
yes
yeah.
C
So
yeah,
so
I
think
that's
an
important
point
with
in
terms
of
tasks
and
you're,
going
to
have
data
and
you're
going
to
be
extracting
data
and
there's
actually
in
c
elegans.
For
example,
there
is
a
lot
of
annotation
so,
like
I
said,
the
cell,
the
process
is
deterministic,
so
we
know
what
the
cells
are.
We
know
what
the
labels
are.
We
know
what
they're
going
to
become,
but
it's
interesting
to
you
know
just
do
these
tasks
independent
of
that
and
say
what
is
this?
C
You
know
because
the
the
the
annotations
that
they
made
when
they
made
the
lineage
tree
of
c
elegans,
it
was
actually
hand
drawn
and
they've
they've
been
able
to
validate
it
since
then,
of
course,
but
like
the
idea
was
that
you
know
you'd
sit
there
and
observe
it.
You
draw
out
like
a
mat,
a
tree
based
on
what
you're
seeing
under
your
microscope.
It
was
pretty.
It
was
there's
no
computation
involved
in
that.
So
you
know
one
of
the
things
that's
interesting.
There
is
that
we
have
data.
C
We
have
like
all
this
cell
tracking
data
and
microscopy
data.
We
can
segment
out,
but
we
also,
you
know,
have
this
annotation
data,
but
we
only
have
like
you
know.
This
is
something
that
was
kind
of
done
through
observation.
It
wasn't
you
know
there.
Some
people
have
done
things
with
molecular
like
molecular
techniques
like
bar
coding
and
things
like
that.
But
there's
still
you
know,
there's
still
relationships
in
there
that
we
don't
know
about
so
there
may
be
relationships
in
there.
C
That
might
be
very
interesting
using
this
sort
of
graph
embedding
and
then
doing
a
task,
and
then
we
can
actually
take
what
we
get
out
of
that
and
match
it
up
to
what
we
know
from
some
of
the
other
techniques.
So
that
might
be
actually
interesting
because
we
do
know
about
the
biology
and
I
don't
wanna-
I
don't
want
you
to.
C
I
don't
expect
you
to
know
that
much
about
the
biology
to
start
with
that's
something
we
can
build
up
towards,
because
it's
it's
something
that
you
have
to
sort
of
learn
a
little
bit
about
and
then
once
you
learn
about
it,
you
can.
You
know
I
mean
in
past
years
we've
had
people
kind
of
put
the
lineage
tree
use
that
as
like
a
training
step.
So,
like
you
know,
you
they'd
use
the
data
set
they'd,
you
know
build
a
classification,
but
then
they'd
also
have
the
lineage
tree
to
help
them
along.
C
F
C
F
F
Their
expandability
or
graphical
natural,
such
as
why
this
tour
can
solve
some
some
tasks
and
I'm
wondering
would
you
do
you
all?
I
mean
the
team.
Does
that
didn't
care
about
this
question
like
if
this
model
can
be
explained
by
some
mathematical
way?
Yes,
because
I
think
if
we
lack
explain
its
explanability
or
models,
we
can
really
trust
our
model.
But
if
we
care
much
about
the
explainability
we
can.
We
cannot
design
some
effective
or
efficient
model
structure,
yeah,
so
yeah,
I'm
interested
in
it.
C
Oh
yeah,
we
we're
interested
in
that
we
we
actually
talk
about
it
occasionally,
like
you
know,
we're
interested
in
kind
of
the
pure
mathematics
of
it,
I
guess,
or
the
the
pure
machine
learning
of
it.
So
you
know
we've
done
that
in
the
past.
Quite
a
bit
I
mean
it's.
Our
main
focus
is
on
sort
of
the
biological.
C
You
know
how
you
apply
it
to
biology,
but
we're
also
interested
in
the
mathematics
of
it
as
well,
because
a
lot
of
the
things
we
talk
about
here,
you
know
we
have
to
understand
the
math
to
understand
how
the
model
can
be
applied.
So
it's
not
not
something
we
can
just
throw
on
to
the
biological
data
set.
We
have
to
know
kind
of
what
it's
doing
and
and
optimizing
you
know
mathematically
so
yeah.
I
think
that's
something
that
we
could
talk
about
as
well
or
or
we're
working.
C
No,
I
don't
know
if
yeah
I
don't
know
if
it
would
be
a
matter
of
just
taking
out
the
cells
or
if
it's
because
I
mean
like
a
lot
of
the
defined
mutants,
you
know
kind
of
what
the
cells
are
missing
like
in
in
some
mutants,
they
miss
you
know,
they're
missing
certain
cells
and
development
yeah.
So
I
don't
know
if
it's
a
matter
just
taking
those
out
or
what.
C
Well,
you
get
like
you
get
the
mutant,
but
you
also
get
like
if
you
do
a
cross
say
between
two
mutants.
You
get
like
different
combinations.
I've
done
this
to
make
like
a
cross
where
the
you
want
to
have
like
a
double
mutant
and
you
can
make
double
mutants
and
even
triple
mutants
doing
cross
breeding
experiments,
so
you
might
yeah,
so
the
wild
type
would
be
similar.
You'd
have
like
that
mutation
in
the
wild
type.
C
Yeah,
well,
it
would
be
yeah
I
mean
you
would
have.
I
wonder
if
we
could
do
that,
just
kind
of
like
computationally,
I
don't
know
but
yeah
I
mean
it
seems
like
it
would
be
very
similar,
because
the
only
thing
you're
doing
in
a
like
a
cross
where
you
have
say
two
defined
mutants
and
you're
just
making
one
with
that
has
like
both
defined
mutations.
C
Is
that
you're,
just
you
know,
combining
them
and
then,
when
they're
combined,
sometimes
the
the
phenotype
isn't
like
the
phenotype
of
both
of
them?
It's
like,
if
you
know
well,
sometimes
it
is
sometimes
it
is
the
ones
I
did
were
actually
a
combination
of
both
so
you'd
get
like
both
sets
of
traits
in
in
the
organism.
So,
okay.
A
C
D
C
D
B
E
Yeah
yeah,
I
just
had
some
more,
you
know
general
doubts,
because
you
know,
as
I
was
going
through
my
proposals,
you
know
I
was
thinking
of
things.
You
know
that
I
can
add
to
it.
So
one
of
the
major
things
you
know
after
the
model
you
know
late,
should,
I
add,
like
volumetric
representations
of
individual
cells
or
like
should
I
label
each
section.
E
E
So
I
was
thinking
of
maybe
you
know
labeling
each
cell
in
that
particular
stage
as
well.
As
you
know,
getting
the
projected
surface
area
of
that
particular
cell,
then
getting
the
volume
based
on
that
projected
surface
area.
So
like
this
was
my
thought
process.
You
know
the
kind
of
features
that
can
be
added
to
the.
E
E
Okay,
so
I
think
there's
a
big
tendency
right
most
of
them
having
a
lot
of
like
most
of
their
projected
pictures
are
usually
in
the
darker
side,
like
like
the
percentage
you
know,
is
kind
of
skewed
more
towards
the
daughter
side
of
the
embryo
embryos.
Is
it
because
it's
more
heavier
on
that
side,
because
it's
a
perfect
microscope?
C
D
Yeah,
is
it
correct
that
the
top
side
is
more
pigment
yeah?
The
top
side
usually
has
more
pigment
pigments
in
it.
There
is
a
sharp
one
between
the
darkest
cloth,
pretty
pretty
sharp
yeah.
It
is
a
definite
divide
there.
Okay
and
it
changes.
If
you
change
gravity,
if
you
put
the
animal
in
a
centrifuge
and
then
you'll
get
a
change
in
the
amount
of
animals
section
that
you
have
okay
yeah,
it
causes
abnormalities
and
development.
D
D
D
C
C
C
And
yeah
it's
something:
we've
tried
to
do
in
some
of
the
other
projects,
but
it's
really
not
something
easy
to
pull
off.
I
think
in
this
kind
of
visualization,
where
you
have
you're
navigating
a
sphere.
You
know
it's
kind
of
like
google
earth
where
you
have
the
earth
and
you're
zooming
into
different
places,
and
the
maps
have
annotations
on
them
like
different
pieces
of
information
and
usually
the
way
they
do.
That
is
your
layers,
so
you
can
turn
layers
on
and
off
as
you
go
through.
D
D
C
D
C
C
E
E
C
C
C
C
E
C
All
right,
so
I'm
gonna
finish
the
meeting
with
a
couple
of
things
following
up
from
last
week,
we
talked
about.
Let's
see
this
is
yeah.
We
talked
about
jamming
phase
transitions
last
week,
so
I
have
a
whole
bunch
of
I'm
going
to
kind
of
quickly
go
through
these
and
and
talk
about
like
some
things
that
follow
up
on
last
week.
So
I'm
not
going
to
spend
a
lot
of
time
on
it,
but
it
for
each
paper,
but
I
can
send
these
out
after
so.
C
The
first
one
is
to
follow
up
on
this
idea
of
jamming
phase
transitions,
which
are
these
changes
in
phase
that
involve
density
of
particles
and
the
idea
that
some
sort
of
like
cooling
or
heating
process,
or
maybe
some
sort
of
simulated
annealing.
So
this
is
a
paper
non-linear
dynamics.
I
don't
remember
what
journalist
is,
but
it's
jamming
is
just
not
cool
anymore,
and
so
this
is
kind
of
goes
through.
It's
like
one
of
these
previews
of
different
papers
that
are
out
there.
C
So
all
around
us
things
seem
to
be
getting
jammed.
We
travel
on
a
highway
and
we
are
caught
in
traffic
jams
and
it
kind
of
goes
through
this.
We
are
usually
so
irritated
that
we
not
really
notice
that
a
jam
state
in
all
these
situations
as
common
properties,
so
the
vibrations
from
the
pounding
actually
do
some
good
and
re-initiating
flow.
So
this
is
where
they
in
this
example
in
factories,
powdered
raw
materials
clogged,
the
conduits
that
were
designed
to
carry
them
smoothly.
C
Our
recourse
in
all
these
situations
is
to
pound
on
our
conduits
and
until
the
jam
miraculously
disappears.
So
that's
the
idea
that
these
phase
transitions
happen
and
then
there's
some
process
to
unjam
things,
and
then
the
jamming
can
will
reoccur
again.
Maybe
and
there's
this
process
of
jamming
and
unjamming.
C
C
These
systems
resemble
solids
because
the
particles
are
driven
into
a
jam
state
by
an
externally
applied
stress
when
jam.
The
disordered
system
is
caught
in
a
small
region
of
phase
space,
with
no
possibility
of
escape,
so
this
is
where
you
have
this
phase
diagram
for
jamming.
This
is
not
exclusively
biological.
This
is
for
all
different
types
of
jamming
phase
transitions,
so
you
have
temperature
load
and
density,
which
are
your
three
parameters
or
one
over
density,
which
is
the
inverse
of
density.
C
And
then
you
have
the
glass
here.
This
is
a
glass
state.
This
is
a
liquid
state.
So,
along
this
temperature
and
density,
reciprocal
axis,
you
have
glass
that
turns
into
a
liquid
along
the
temperature
load
area.
You
have
these
jammed
grains,
turning
into
a
load
and
then
loose
grains
bubbles
droplets
turning
into
jammed
greens
on
the
load,
reciprocal
density
plane.
So
you
can
see
that
there
are
all
these
different.
You
have
this
central
region,
which
is
the
jam
state.
C
It
could
be
glass,
it
could
be
jammed
grains
into
a
solid,
and
then
there
are
all
these
things
that
are
kind
of
you
know
other
areas
that
predominate.
So
the
approach
of
kate's
is
a
start
from
a
pile
of
complex,
completely
non-deformable
particles,
for
which
strain
is
not
obviously
useful.
Variable.
Their
simple
model
of
the
chain
is
of
hard
particles
of
a
chain
of
hard
particles
insists
that
the
gm
system
cannot
be
considered
as
an
elastic
body,
so
they
kind
of
go
through
this
model
and
they
talk
about.
C
There
is
because
there
is
no
obvious
relation
connecting
stress
to
strain
throughout
the
pile
kate
said
I'll
bypass
the
strain
all
together
and
propose
a
relation
between
different
components
of
the
stress
tensor.
This
continues
to
be
a
hotly
debated
assumption,
so
people
are
building
models
of
this
with
particles,
things
like
shaving,
foam
and
they're.
You
know
evaluating
the
state
and
the
transition,
so
shaving
foam,
for
example,
is
jammed
because
the
bubbles
are
tightly
packed
together
under
an
isotropic
stress,
namely
atmospheric
pressure.
C
Greens
are
interesting
because
they
actually
also
use
sand
piles
to
demonstrate
a
first
order:
phase
transition
where
you
drop
sand,
grains
onto
a
pile
and
then
that
pile
at
random
intervals,
it
it
releases
stress
by
deforming,
and
you
get
a
large
avalanche
of
sand
greens,
and
so
this
is
this
is
the
type
of
thing
is
the
first
order.
Phase
transition,
in
this
case
they're
talking
about
sand
greens
being
involved
in
a
in
a
jamming
transition.
C
C
C
C
C
A
I
worked
on
a
membrane
which
is
an
ising
lattice
and
what
would
happen
if
you
looked
at
transport
across
this
membrane
once
the
concentration
inside
the
membrane
was
high
enough.
It
was
an
ising
lattice,
the
particles
would
attract
one
another
and
it
would
jam,
and
the
transport
across
the
membrane
would
go
down
enormously.
A
B
C
A
We've
thought
of
in
terms
of
jamming
in
the
one-dimensional
case,
at
least
because
it's
analogous
to
what
have
you
ever
have
you
ever
been
at
a
a
single
lane
bridge?
Yes,
yes,
okay,
they're
they're
common
in
new
england,
yeah
they're,
often
covered
structures,
and
anyway
the
traffic
can
only
go
one
direction
or
the
other.
And
so
you
end
up
with
jamming
interactions
inside
the
bridge.
C
A
C
C
So
again,
this
this
paper,
which
is
jamming
phase
diagrams
for
attractive
particles,
this
kind
of
goes
through
this
is,
I
think
I
showed
this
well.
This
is
the
one
we
just
saw
on
this
paper,
but
I
think
people
in
last
week's
papers.
I
think
at
least
two
of
them
use
this
diagram,
so
I
was
trying
to
track
down
the
source
of
this
diagram.
It's
from
this
nature
paper
back
in
2001,
so
you
know
this
is
where
they
kind
of
laid
out.
C
C
B
A
C
So
that's
and
then
there's
this.
Then
we
get
into
some
of
the
things
that
are
more
biological.
Actually,
this
one
is
universal
scaling
laws
of
glass
rheology.
So
rheology
is
just
this
study
the
material
and
you
know
scale.
We've
talked
about.
Maybe
universal
scaling
laws
a
bit
in
other
contexts,
but
you
know
with
phase
transition.
First
order
phase
transitions.
They
often
use
scaling
laws
to
describe
the
behavior
of
these
phase
transitions.
C
You
know
with
massive
amounts
of
sand
particles
at
random,
so
but
the
distribution
of
those
displacements
observe
the
scaling
law
where
you
get
a
lot
of
displacements
of
single
greens
after
a
grain
has
been
dropped
on
the
top
and
then
a
very
few
number
of
large
displacements,
and
so
those
can
be
scaled
according
to
a
scaling
law
according
to
a
power
law
function.
So
that's
that's!
Where
scaling
laws
are
often
used.
They
often
use
the
muscle
to
analyze
complex
networks
in
this
case
they're.
Looking
at
glass
reality
they're.
C
Looking
at
this
universal
scaling
law,
so
the
similarity
in
atomic
molecular
structure
between
liquids
and
glasses
has
stimulated
a
long-standing
hypothesis,
but
the
nature
of
glasses
may
be
more
fluid
like
rather
than
the
apparent
solids
so
they're
talking
about
this.
That
glass
behaves
like
a
fluid,
but
it
appears
to
be
a
solid
and,
of
course
it
behaves
like
a
solid.
If
you
put
liquid
inside
of
a
glass,
it
doesn't
leak
out.
C
C
Here
we
report
the
dynamic
response
of
shear
stress
to
the
shear
strain
rate
of
metallic
glasses
over
a
time
scale
of
nine
orders
of
magnitude
equivalent
to
hundreds
of
years.
So
they
simulate
this
process
of
of
pressure
on
on
these
glasses
over
hundreds
of
years,
and
then
they
use
this
type
of
stress
relaxation
experiment
to
do
this.
C
The
dynamic
response
of
the
metallic
glasses,
together
with
other
glasses
in
quotes
because
there
are
different
types
of
glass.
Of
course,
if
you
study
the
material
that
you
know
they
take
different
forms
like
at
a
molecular
level
follows
the
universal
scaling
law
within
the
framework
of
fluid
dynamics.
C
This
provides
a
comprehensive
validation
of
the
conjecture
on
the
jamming
dynamic
phase
diagram
and
that's
that
same
phase
diagram
we
saw
before
which
the
dynamic
behaviors
of
the
wide
variety
of
glasses
can
be
unified
under
one
rubric
parameterized
by
the
thermodynamic
variables
of
temperature
volume
and
stress
in
the
trajectory
space.
So
this
is
that
same
diagram
that
we
saw,
I
don't
know
if
they're
going
to
have
it
in
the
paper.
C
This
is
a
scaling
law
here,
where
you
see
the
strain
versus
the
stress.
So
this
is
the
scaling
and
you
can
see
that
it
follows
this
law
where
the
strain
and
stress
are
related,
and
so
let's
see,
if
there's
anything
else
in
here,
I
don't
think
they
actually
show
that
diagram,
but
they
show
some
other.
Oh.
C
C
C
and
then
this
links
a
lot
of
this
to
biology
into
embryos,
so
many
biological
tissues
are
viscoelastic
behaving
is
elastic
solids
and
short
time
scales
and
fluids
on
long
time
scales.
So
again
we
have
this
this
transition
and
matter
type
that
were
how
the
biological
matter
behaves.
So
it's
elastic,
like
an
elastic
solid
on
a
short
time,
scale
fluid
on
a
long
time
scale.
This
collective
mechanical
behavior
enables
and
helps
to
guide
pattern
formation
and
tissue
layering.
So
this
is
something
that's
used
in
the
process
of.
C
You
know
these
sorts
of
things
here
we
investigate
the
mechanical
properties
of
three-dimensional
tissue
explants
from
zebrafish
embryos.
So
these
are
the
zebrafish
embryos
we've
talked
about
there.
I
don't
know
if
they're
using
it
for
any
specific
reason,
they
have
a
lot
of
they
they're
well
characterized
and
they
know
kind
of
how
these
process.
These
processes
are
well
characterized
in
terms
of
single
cells
and
single
cell
behavior.
C
So
so
they
do
this
by
analyzing,
individual
cell
tracts
and
macroscopic
mechanical
response,
so
they
use
a
sinus
cell
tracking
where
they
can
track
the
embryo
or
they
can
track
the
nucleus
of
the
cell
and
they
can
look
at
where
it's
moving.
They
can
also
look
collectively
because
in
zebrafish
you
start
to
get
a
large
number
of
cells
that
divide
from
single
cells.
C
C
So
I
don't
know
I'm
not
familiar
with
caging
behavior,
but
we
develop
a
minimal
three
parameter
mechanical
model
for
these
dynamics,
which
we
calibrate
using
only
information
about
cell
tracks,
so
they're,
using
just
the
cell
track
information
which
doesn't
have
any
inherent
mechanical
information.
It
just
shows
you
where
the
cell
was
moving
and
maybe
the
velocity
you
can
derive
that
from
the
cell
tracks
as
well.
This
model
generates
predictions
about
the
macroscopic
bulk
response
of
the
tissue
that
are
verified,
experimentally,
providing
strong
validation
of
the
model.
C
So
this
is
where,
if
you
change
these
parameters
slightly,
you
can
change
the
physical
elastic
properties
of
the
tissue.
I
don't
know
if
that
relates
to,
like
maybe
like
in
mutant
some
mutants.
You
might
have
this
issue
where
you
know
the
viscoelastic
properties
of
the
tissue
change
or
whether
there
are
developmental
perturbations
that
can
occur.
That
can
do
this,
but
it's
the
idea
is,
is
that
it's
very,
very
close
to
this
phase
transition.
C
C
C
This
is
actually
d-
is
a
two-dimensional
cross-section
of
three-dimensional
tissue
simulation.
So
this
is
a
voronoi
tessellation
of
this,
so
you
can
see
that
this
tissue-
this
is
the
microscopy
image.
This
is
a
simulation
where
they've
taken
what
they
call
a
voronoi
algorithm,
which
is
where
they
find
from
a
centroid
like
the
most
optimal
shape
of
the
cell,
and
so
you
can
see
some
of
these
cells
have
centroids
like
nuclei
and
some
of
them,
don't
it's
just
a
product
of
that
voronoi
tessellation,
and
then
they
show
some
of
these
scaling.
C
So
this
is
two
cell
sample
cell
tracks
extracted
from
experimental
data
illustrating
caging
behavior.
So
that's
in
b
and
c:
that's
here,
this
caging
behavior,
where
it's
along
this.
I
guess
along
this
path
here
and
then
two
dimensional
cross-section
of
the
three-dimensional,
that's
the
one
I
just
showed
msd
data
from
experimental,
ectoderm,
ectoderm,
thin
dashed
and
misendo
dura
solid
explants.
C
So
basically,
what
that
means
is
that
your
cells
are
packed
together,
pretty
densely
and
the
only
way
to
get
those
cells
to
move
around
or
move
position.
So
you
can
see
these
this
purple
cell
in
this
green
cell
are
in
this
position.
The
only
way
to
get
these
cells
to
flip
their
position
is
through
some
some
sort
of
high
energy
collective
rearrangement.
C
So
this
means
that
you
know
again
this.
We
can
observe
this
jamming
phase
transition,
the
cells
migrate
into
this
confluent
state
and
then
the
only
way
to
break
them.
Out
of
that
state
to
flip
things
around
is
to
have
some
high
energy
displacement,
so
you
have
to
sort
of
displace
that
jamming
transition,
like
we
mentioned
with
simulated
annealing
before
you
have
this
like
in
simulated
kneeling.
C
A
C
C
Yeah,
so
I
don't
see
any
reference
to
that,
but
they
they
have
some
nice
pictures
in
here
where
they
show
yeah.
They
show
some
of
these
physical
experiments
and
how
they
do
the
quantification.
It's
kind
of
nice
paper
from
that
standpoint
too,
but
I
don't
think
they
mentioned
the
leafy
flight
at
all
or
anything
about
nonlinear
diffusion.
C
So
then,
that's
finally,
this
paper,
the
interplay
between
cell
signaling
and
mechanics
and
developmental
processes.
So
this
one
talks
about
cell
signaling
in
mechanics,
so
it
takes
this
idea
of
the
jamming
phase
transition
and
puts
it
in
sort
of
a
context
of
cell
signaling
and
some
of
the
things
going
on
there.
So
again,
this
this
relates
to
things
like
positional
information
and
other
types
of
things
that
we
find
in
pattern
formation.
C
So
in
positional
information,
it's
the
position
of
a
cell
that
is
either
you
know,
sort
of
coordinated
by
within
the
larger
context
of
the
embryo,
where
that
the
cell
has
positional
information
as
to
where
it
should
be
in
the
embryo.
So
this
positional
information,
then
they
talk
about
this
in
the
context
of
biomechanics
and
other
types
of
physical
stuff.
So
this
talks
about
the
spatial
and
temporal
regulation
of
gene
expression
and
protein
activity
guides
cell
physiology
behavior,
but
there's
also
some
role
for
interactions
with
the
physical
and
mechanical
constraints
on
development.
C
So
these
are
early
studies
that
sort
of
tested
the
plausibility
of
physical
laws
to
morphogenesis.
So
there's
there's
a
rich
history
here
of
this.
They
also
talk
about
some.
They
have
this
box
of
engineering
principles
and
terms
for
people
who
are
reading
this.
I
imagine
this
is
more
developmental
biologists,
so
they
have
this
glossary
here
and
then
they
talk
about
cell
shape,
change,
cause
or
effect.
So
you
know,
one
of
the
questions
is:
is
do
these
changes?
C
Are
they
causal
or
is
there
an
effect
of
something
else,
and
so
they
kind
of
get
into
some
of
this
in
so
like
these
physical
forces,
are
they
causal
or
the
effect
of
something
else?
The
same
thing
goes
for
like
the
effects
on
cells
themselves
so
yeah.
This
is
a
nice
review
for
this
sort
of
bringing
together
these
themes,
especially
in
terms
of
cell
signaling
and
some
of
the
developmental
biology.
C
So
that's
all
I
have
to
say
about
that.
I
think
that's
a
good
set
of
papers.
I
just
wanted
to
follow
up
on
some
of
the
things
we
were
talking
about
last
week,
because
I
thought
it
was
pretty
interesting
stuff.
So
susan
actually
left
a
couple
of
comments
here.
Lisa
manning
in
a
group
has
written
about
jamming
yeah
that.
B
C
D
D
C
I
know
that,
like
even
in
in
like
stem
cells,
they
do
a
lot
of
studies
where
they
look
at
like
putting
stem
cells
in
different
substrates
and
introducing
forces,
and
they
can
actually
sometimes
affect
the
differentiation
potential
with
the
cell
so
like
if
they
put
a
stem
cell
on
a
certain
substrate
one
substrate
versus
another.
Sometimes
they
have
to
pattern
it.
They
can
get
it
to
differentiate,
or
sometimes
it
is
a
cancer
cells
where
they
put
it
on
the
substrate
and
they
can
get
it
to
differentiate
down
a
certain
pathway.
C
A
C
A
B
B
C
B
C
All
right:
well,
let's
wrap
it
up
for
today
again,
gsoc
applications
are
due
tomorrow,
so
be
sure
to
submit
by
the
deadline
submit
to
their
portal
and
I'll,
be
happy
I'll,
be
looking
forward
to
reading
through
and
evaluating
them.
And
then,
if
you
have
any
other
things
you
want
to
bring
up
during
the
meeting.
Let
me
know
or
bring
them
to
the
meeting
next
time,
and
we
can
talk
about
them.