►
From YouTube: DevoWorm (2021, Meeting 39): Segmenting Diatoms, Topological Analysis, Aspects of Morphogenesis
Description
Overview of progress on segmenting Diatom colonies with pix2pix algorithm. Topological Data Analysis and associated methods, upcoming conferences, overviews, and Github issues. New paper on the C. elegans connectome (sexual differentiation, predation). Aspects of morphogenesis: mosaic (programmatic) vs. regulative (self-organization) development, new perspectives on Turing R-D, genetic manipulation of positional information. Attendees: Asmit Singh, R. Tharun Gowda, Karan Lohaan, Mainak Deb, Susan Crawford-Young, Richard Gordon, and Bradly Alicea.
A
A
C
A
Yeah
sure,
all
right
so
welcome
to
the
meeting
we're
gonna
get
started.
We're
gonna
talk
about
the
digital
basil
area
stuff
and
it
looks
like
the
roon
and
asmeter
here
then
we'll
go
on
to
some
papers
and
some
other
submission
information,
and
maybe
some
other
people
show
up
during
the
meeting
we'll
see
it's
usually
what
happens
so
so
what
do
we
start
with
talking
about
throne
and
asthma
and
your
progress
on
the
digital,
basilary
images.
B
A
D
Yes,
so
like
we
were
like
trying
to
like
make
those
bounding
box
kind
of
figures
on
the
overlapping
objects,
but
I
think
the
model
is
not
performing
well
because
we
have
like
only
one
video
and
most
of
the
images
that
we
got
from
the
video
are
like
blurry
and
a
lot
of
our
focus
and
yeah.
So
I
guess
we
need
to
find
a
bigger
data
set,
so
we
can
train
the
model
on
that
and
improve
the
results.
D
D
B
E
D
B
So,
like
this
mountain
is
a
modified
light,
maybe
that
helps
the
moderate
gap
to
those
boundaries
in
like
a
better
way.
So
yeah,
like
other,
like
from
the
top
of
the
mind
I
can
think
of
like
basic
water
shedding
or
like
this
kind
of
edge
detection
technique,
so
like
just
basically
sharpening
the
edges
of
these
colonies.
So
we
could
try
these
like
different
filters.
We
have,
you
know
drawn
up
with
the
locks
which
I
think
all
these
administrators
have
to
have.
A
E
A
That
we
have
so
that's
yeah,
but
I
I
wouldn't
you
know,
I
wonder
if
there
isn't
some
way
to
take
like
well.
There
might
be
some
way
to
take
the
entire
colony
and
and
sort
of
do
an
ad
hoc
type
of
classification.
Where
you
know
you
just
look
at
the
movement
across
cells.
Like
you,
don't
worry
about
the
single.
A
B
B
Yeah
if
they
still
have
some
sort
of
an
overlap,
or
so
like
two
rectangles
and
along
their
lens,
they're
overlapping.
So
in
that
case
it
basically
captures
it
as
one
long
radar
instead
of
two
going
to
the
top
this
one
over
here,
so
it'll
capture
it
like
this
yeah.
So
it's
something
like
that,
but
in
the
cases
where
it's
not
performing
well,
I
guess
the
thing
is
that
the
second
cell
is
actually
much
more
vertically
like
a
vertically
down.
B
D
B
Right
so
I
guess,
or
the
cases
it's
not
performing
as
well
as
when
this
light
like
these
are
really
left
leg,
they're
distributed
vertically
and
there's
not
as
much
a
horizontal
displacement.
So
in
these
cases,
when
they
so
so
in
this
case
or
like
the
model,
is
not
picking
up
that.
These
are
two
different
rectangles
and
it's
like
basically
considering
this
as
one.
E
A
E
A
You
basically
know
that,
like
the
cell
is
of
a
certain
length,
so
you
can
you
know
you
can
limit
your
rectangle
to
that
size
and
then
it
would
just
like
if
it's
that,
if
it's
twice
the
length
you're
near
twice
the
length,
then
it
might
like
you
know,
be
able
to
find
a
boundary.
It
wouldn't
be
the
exact
size
of
the
cell,
but
it
would
like
differentiate
the
cells.
A
So
you
know
you
modif
you,
you
introduce
some
variation
to
the
data
set,
and
so
in
that
way
you
can
actually
you
know,
maybe
train
in
on
a
number
of
different
cell
boundaries.
In
that
case
like
for
those
cases,
you
know
you
might
have
like
something
where
you
know
you
actually
define
the
boundary.
Somehow
I
know
we
were
drawing
like.
B
A
One
time,
like
I
remember
doing
this,
where
we
would
draw
boundaries
for
a
very
small
training
set,
like
you
know
it
within
a
embryo
we'd
have
like
you
know,
we
could
differentiate
most
of
the
boundaries
of
the
cells,
but
there
were
still
some
trouble
spots,
so
we
just
did
it
by
hand
for
a
very
small
training
set,
and
then
you
know
train
it
on
those
boundaries.
So
kind
of
you
know
you
kind
of
fill
out
what
it's
missing.
E
B
B
A
B
B
B
D
D
B
B
B
B
A
E
B
A
But
I
think
it's
it.
It
sounds
pretty
good
sounds
like
you're
coming
along
a
little
bit
of
a
some
issues
with
you
know,
but
this
is
probably
normal,
because
it's
not
you
know,
you
do
have
a
lot
of
things
that,
where
the
focus
changes-
and
so
you
know-
that's
just
that's
what
happens
when
you're
trying
to
observe
a
colony
like
that,
and
but
that's
that's
the
nature
of
this
data.
So
I
think
that
should
work
though
yeah
we'll
see
how
it
works,
and
you
can
report
back
it's
great.
A
B
B
B
B
So
our
idea
was
that
we
wanted
to
do
segmentation
right
now,
like
the
typical
augmentation
techniques
or
like
these
unit
based
approaches
were
not
really
working
there.
So
so
we
moved
on
move
to
this
particular
approach,
so
it
up.
There
give
the
model
a
set
like
set
of
video
frames,
which
are
like,
obviously,
the
microscopic
images
and
another
expected
output
would
be
a
black
value
with
this
white
rectangles
and
each
white
rectangle.
B
A
A
C
E
C
A
A
Yes,
thank
you,
and
so
you
know,
if
you
want
to
ask
questions
about
anything,
you
can
reason
the
slack.
We
have
a
what
we
call
the
diva
worm
slack,
which
is
the
general
slack
channel.
Then
we
have
the
divo
learn
channel,
which
is
if
you're
interested
in
diva
learn,
which
is
something
that
my
knock
is
involved
with.
Then
you
know
you
can,
and
this
is
like
a
pre-trained
model
for
cell
segmentation
in
embryos.
A
You
can
join
that
group
and
you
know
that's
more
technical
around
the
issues
surrounding
divalern,
but
you
know
you
can
ask
general
questions.
I
guess
about
computer
vision
and
and
about
machine
learning
in
that
channel
as
well.
So
we
have
a
lot
of
people
are
interested
in
machine
learning
and
computer
vision
as
you've
just
heard,
and
a
lot
of
people
using
trying
to
use
different
techniques.
A
We
have
open
data
sets,
so
we
have.
You
know
we
have
a
diatom
species
called
bacillaria,
which
is
this
colony
of
these
rod
shaped
cells
that
move
in
different
ways.
We
have
c
elegans,
of
course,
which
is
the
nematode.
So
you
know
we
look
at
the
embryo
of
that.
So
usually,
you
have
a
relatively
small
embryo
that
has
labeled
data
that
you
can
use.
A
You
know
you
can
connect
the
labels
to
the
cells
as
they're
dividing.
So
that's
actually
a
very
nice
model
for
doing
interesting
things,
and
then
we
have
other
data.
We
have.
We
have
a
things
like
axolotl
data
that
we
know
we
can.
You
know
analyze
and
that's
a
little
bit
more
challenging,
but
that's
something
that
is
available
and
you
can
work
on
that
as
well.
A
So
welcome
to
the
group-
and
if
you
have
any
questions
you
know
put
the
questions
in
slack
or
you
know,
look
over
the
github
repos
and
we
have
the
open
issues.
But
you
know
some
of
those
open
issues
are
pretty
broad,
so
I
don't
know
if
you
can
necessarily
some
of
them.
You
can't
really
address
directly.
You
have
to
sort
of
inquire
about
them,
so
that's
just
kind
of
the
nature.
G
G
G
A
Well,
good
welcome.
So,
let's
see
the
next
thing
I
want
to
talk
about
was
we
have
a
couple
of
things
you
know
ongoing?
I
always
do
this
in
every
meeting.
I
go
through
a
couple
of
things
that
we
have
so
the
submissions,
so
we
know
we're
working
on
different
submissions
and
some
of
these
are
kind
of
still
out.
You
know
outstanding.
A
We
have
a
couple
of
conferences
coming
up
in
december,
there's
neuromatch
4.0.
So
this
is
a
computational
neuroscience
conference,
but
you
know
they
also
do
some
machine
learning
and
deep
learning
in
this
group.
This
is
a
online
conference.
It's
you
know
it
has
a
nominal
fee,
but
it
can
be
free.
You
can
requested
a
waiver
it's
december
one
and
two.
A
So
if
you're
interested
in
that
area,
of
course,
go
to
that
conference,
we're
not
submitting
anything
from
this
group,
but
my
other
group
is
submitting
a
couple
things
and
I
was
thinking
of
having
something
in
conjunction
with
this.
I
don't
know
what
exactly
yet
it
could
be
on
a
regular
time
on
monday.
A
It
could
be
some
other
time
during
the
week,
but
it
you
know
I
was
thinking
about
doing
something
with
you
know,
maybe
inviting
people
from
that
community
to
see
some
of
the
things
we're
doing
in
diva
worms,
so
more
information
about
that.
It
might
be
just
as
simple
as
an
overview
for
people,
but
I
definitely
want
to
take
advantage
of
this
in
some
way.
A
Then
there's
also
nur
ips,
and
this
is
happening
in
early
december
and
the
submissions
for
that
are
closed.
But
I
wanted
to
point
out
that
this
is
also
virtual
and,
if
you're
interested
in
machine
learning
deep
learning,
this
is
something
that
you
can
attend
or
as
jesse
who's.
A
member
of
our
group
has
done.
A
We
have
some
of
these
other
things
like
quantitative
comparison
of
archaea
and
shape
droplets.
This
is
something
that's
kind
of
outstanding
we've.
You
know
had
some
conversations
about
this
and
it's
basically
taking
single
cells
and
looking
at
their
shape
and
classifying
the
shape.
I
think
my
knock
was
working
on
that
at
one
point,
I
don't
know
if
he,
if
he
made
any
progress
on
that
or
not.
A
We
also
have
the
mathematics
of
diva
worm,
which
is
this.
It's
supposed
it's
going
to
be
like
a
paper
of
different
mathematical
models.
We
use
in
diva
worm.
So
this
is
something
that
is
currently
kind
of
in
a
poster
form
and
needs
to
be
flushed
out
into
a
paper
yeah.
I
haven't
worked
on
it
in
a
while,
but
it's
something
we'll
come
back
to
this.
Where
is
it?
There's
a
oh
yeah,
the
diva
learned
paper.
A
This
has
been
like
very,
very
behind
I'm
very,
very
behind
on
this,
but
we
so
for
the
divo
learn
platform
we're
developing
this
paper
just
like
kind
of
like
a
technical
paper
on
it.
We
need
to
fill
it
out.
I
promised
that
I
would
do
it
after
the
end
of
last
google
summer
of
code,
but
I
have
yet
to
get
to
it.
A
So
this
is
something
that
will
be
kind
of
I'll
try
to
make
a
point
to
work
on
this
a
bit,
and
this
is
something
that
my
knock
and
and
a
couple
other
people
are
working
on.
Okay
thirun
says
he
has
to
go
I'll.
Try
some
data,
augmentation
techniques
start
with
finding
coordinates
of
the
objects
and
get
back
have
a
nice
week.
Thank
you
for
attending
through
and
so
that.
That's
that's!
So
that's
good,
and
so
yeah
we'll
be
working
on
that
as
well.
A
So
there
are
a
lot
of
yeah
there's
still
some
things
that
are
outstanding.
If
you're
interested
in
working
on
some
defined
problem,
we
have
some
things
here
on
diatoms,
we
have
some
other
types
of
computational
biology
projects.
Let
us
know
where
what
you
might
want
to
work
on
if
there's
something
of
interest
here,
we
also
have
our
major
tasks,
so
this
is
still
technically
hacktoberfest
and
this
is
for
2021.
A
A
We
also
have
some
other
things
that
are
ongoing
some
other
papers.
This
is
in
the
group
meeting.
So
if
you
go
to
diva
worm
your
readings
repository,
this
is
the
project
board
there,
and
so
there
are
all
these
issues
and,
like
I
said,
the
issues
can
be
very
broad.
They
can
you
know
there.
Some
of
them
are
very
broad,
so
these
are
things
that
you
need
to
sort
of
inquire
about.
A
E
A
A
E
A
A
So
this
is
something
we've
talked
about,
and
topological
data
analysis
is
this
area
of
analysis
that
basically
you're
not
analyzing
shapes
and
you're
you're,
applying
mathematical
techniques
and
models
to
different
shapes
and
you're
using
shapes
to
characterize
data?
And
it's
it's
a
really
interesting
field
and
we
haven't
really
gotten
very
far
in
it
in
general
in
terms
of
research,
but
there
are
some
tools
that
that
exist
on
the
web.
You
know
this
is
sort
of
an
alternative
to
machine
learning.
Where
you
know
you
use
a
training
set
to
analyze.
A
A
You
know
with
numbers,
so
topological
data
analysis
doesn't
really
involve
too
much
training.
It
involves
using
tools
to
sort
of
quantify
the
shape
and
structure
and
data,
so
so
yeah.
So
this
is
a
collection
of
powerful
tools
that
can
quantify
shape
and
structure
and
data
in
order
to
answer
questions
from
the
data's
domain.
This
is
done
by
representing
some
aspect
of
the
structure
of
the
data
in
a
simplified,
topological
signature,
so
topological
means
something
to
do
with
shape.
A
So
if
you're
familiar
with
you
know
like
a
romanian
surface
like
in
theoretical
physics,
they'll
often
use
that
and
it's
just
a
sort
of
a
spherical
projection
of
space,
and
so
you
know
there
are
different
ways
that
you
can
do
analyzes
different
models.
You
start
with,
and
you
analyze
things
that
unfold
on
the
space.
A
So
you
know
a
lot
of
times
what
they'll
do
is
they'll
project
data
to
a
space
and
then
they'll
analyze
it
in
that
space
and
there's
a
reason
why
they
do
this,
and
so
it
sometimes
it
reveals
the
structure.
Actually,
if
you're
familiar
with
things
like
principal
component
analysis,
a
lot
of
times,
you're
projecting
data
points
to
sort
of
a
manifold
and
you're,
you
know
rotating
that
manifold
and
you're
finding
the
best
fit.
So
that's
not
really
a
topological
data
analysis,
but
it
uses
the
same
set
of
principles.
A
First,
the
persistence
diagram
represents
loops
and
holes
in
the
space
by
considering
connectivity
of
the
data
points
for
our
continuum
of
values,
rather
than
a
single
fixed
value.
So
this
is,
they
have
a
persistence
diagram
which
is
one
tool
so
you're
representing
different
things
like
loops
and
holes
in
a
space
you're
considering
connectivity.
A
So
we've
talked
about
this
with
respect
to
graph
theory
where
or
complex
networks
where
things
are
connected
and
if
they're
connected,
then
they
have
this,
you
know.
Sometimes
things
aren't
connected
fully
they're.
You
know
different
gaps
and
the
connectivity,
and
so
these
are
the
things
we
want
to
know
where
those
are
and
then
so
that's
one
way.
The
second
topological
signature,
the
mapper
graph,
returns
a
one-dimensional
structure
representing
the
shape
of
the
data
and
is
particularly
good
for
exploration
and
visualization
of
the
data.
Well,
these
techniques
are
based
on
very
sophisticated
mathematics.
A
A
And
so
yeah
this
is
kind
of
the
same
thing,
and
so,
let's
see
this
little
article
serve
as
an
introduction
to
the
standard,
tda
methods
for
domain
scientists
and,
let's
see
so
data
with
distance.
So
you
know.
The
first
thing
you
need
to
understand
is
that
there's
this
idea
that
there's
a
distance
between
data
points
and
so
when
you
have
data
points
for
the
distance?
A
E
A
Thought
we
were
going
to
go
through
a
bit
more
here,
but
I
think
the
idea
here
is
that
you
know
you
have
your
data
has
a
distance.
You
can
configure
the
data
on
a
surface.
You
can
find
the
distance
between
points.
Then
you
can
find
sort
of
the
structure
between
those
points
and
you
can
look
for
different
ways
of
you
know
different
geometric
relationships
between
them
and
so
there's
a
it's.
You
know
I
don't
know.
A
A
The
next
thing
I'm
going
to
talk
about
is
this
paper.
I
actually
did
a
review
on
this
paper
and
this
is
another.
This
is
actually
going
towards
a
complex
graph
or
complex
network
theory
or
graph
theory,
and
this
is
called
resilience
and
nematode.
Connectors
based
on
network
dimension,
reduced
method,
and
so
this
is
a
I.
I
was
a
reviewer
on
this
paper,
so
I
this
is
an
interesting
paper.
They
talk
about
the
connectome
of
c
elegans,
but
they
also
talk
about
the
connectome
of
christianca
specificus,
which
is
another
nematode.
A
There
are
certain
differences
in
you
know,
there's
certain
cells
and
certain
circuits
that
exist,
that
are
you
know,
predator
specific,
and
so
so.
The
abstract
on
this
is
the
whole
map
of
nematode
connectomes,
provide
important
structural
data
for
exploring
the
behavior
mechanism
of
nematodes,
but
to
further
reveal
the
functional
importance
and
resilience
pattern
of
nematode
neurons.
It
is
necessary
to
effectively
couple
the
regulatory
relationship
between
neurons
in
their
topology,
so
this
is
where
they
actually
use
a
signal.
A
More
generally,
you
know
open
worm
is
using
software
they're
using
things
like
neuron
and
some
other
types
of
software
to
model
the
electrical
activity
of
neurons
in
the
connecto,
and
so
you
can
do
this
for
different
circuits,
like
behavioral
circuits
and
even
get
out.
You
know
you
can
get
different
behaviors.
You
can
simulate
different
behaviors
and
so
they're
doing
a
similar
thing
here.
A
They're
using
this
excitation
function,
they're
using
this
a
differential
equation
to
depict
the
activity
of
a
neuron,
so
they're
not
actually
looking
at
the
ion
channel
activity,
which
is
what
they
do
in
open
worm,
but
they're
looking
at
a
more
sort
of
collapsed
description,
which
is
this
differential
equation
model,
and
so
we
need,
because
we
have
a
lot
of
neurons.
We
need
a
high
dimensional
differential
equation
to
capture
the
neural
network
with
mean
field
theory.
A
We
decouple
this
n-dimensional
question
into
a
one-dimensional
problem
and
then,
in
this
framework
we
emphatically
analyze
the
characteristics,
similarities
and
differences
of
the
structure
and
dynamical
behaviors
of
the
neurons
neuronal
system
for
these
two
species,
and
then
they
talk
about
comparing
the
results.
A
simulating
method
and
theoretical
approach
shows
the
most
important
homologous
neurons
between
these
two
species
or
i2
and
nsm
so
they're,
looking
at
the
different
connectives
for
these
two
species,
and
they
find
that
there
are
some
two
neurons
which
serve.
A
You
know
that
are
basically
the
same
in
both
species
and
i2
and
nsm,
which
may
lead
to
their
different
behavior
characteristics
of
predation
and
prey.
So
that's
interesting.
They
did
this
experiment
where
they
took
this
data
set
from
a
group.
You
know
like
a
biological
group
and
they
analyzed
it
using
this
method.
So
they
got
the
connectoms,
which
is
you
know.
Usually
people
will
take
like
they'll.
Do
microscopy,
they'll
take
what
they
call
micrographs
and
they'll
look
for
gap,
junctions
or
they'll,
look
for
synaptic
connections
between
cells
and
then
they'll
mark.
A
You
that
those
cells
are
connected
somehow
now
what
they've
done
here
is
they've
used
a
model
to
simulate
the
behavior
of
the
neuron,
the
activity
of
the
neuron,
and
then
you
know
they're
connected
together,
so
there's
an
assumption
that
they
work
together
in
a
circuit,
and
then
you
know
you
have
these
through
these
simulations
they're
able
to
show
that
c,
elegans
and
pacific
pristianca
specificus
have
these
two
different
connectomes.
They
behave
in
certain
ways
and
that
these
two
cells
are
sort
of
homologous,
i2
and
nsm.
E
A
A
So
that
means
that
they
have
both
eggs
and
they
have
sperm,
and
so
they
fertilize
they
sell
fertilize,
and
they
have
they
lay
eggs.
They
don't
need
a
partner
to
have
to
lay
eggs
that
are
fertile,
so
they
can
produce
their
own
offspring,
and
so,
but
this
leads
to
a
lot
of
what
they
call
clonal
selfing
and
it's
not
good
for
the
species
diversity.
A
So
one
percent
of
the
population
is
composed
of
males
and
males
are
very,
you
know
you
can
detect
males.
They
have
this
interesting
tale.
It
looks
like
a
little
shovel
and
that's
the
where
they,
you
know,
send
their
sperm
out
to
some
hermaphrodite.
That's
they
mate
with,
and
so
this
is.
This
male
strategy
exists
so
that
the
c
elegans
populations
that
do
all
the
selfing
don't
run
out
of
genetic
diversity.
A
A
But
a
couple
years
ago
there
was
a
group
that
mapped
out
both
connectomes
c
elegans
or
hermaphrodite
and
male
and
so
they're
different
a
little
bit
of
difference
in
the
connectome.
But
you
know
they
wanted
to
map
it
out
to
make
sure
that
we
had
both,
and
so
they
took.
Actually,
this
group
took
those
two
data
sets
and
did
some
further
analysis
on
it.
A
So
in
the
hermaphroditic
and
male
c
elegans,
we
test
the
control
level
of
the
intermediate
neuron
group
over
the
output,
neuron
group
and
the
single
neuron.
These
results
suggest
that
our
theoretical
approach
can
be
used
to
reveal
the
effects
of
bioconnectivity
groups
potentially
enabling
us
to
explore
the
interaction
relationship
of
neural
networks.
So
this
is
an
interesting
paper.
They
have
these
two
different,
so
they
have
a
they
start
with
this
neurodynamics
model.
A
E
A
Neuron
I
I
is
a
capital,
I
is
the
basal
activity
of
the
neuron,
which
is
at
rest.
J
is
the
maximal
interaction.
Strength
between
a
pair
of
neurons,
which
is
where
they're
coupled
r
is
the
inverse
of
the
death
rate.
Alpha
is
the
firing
threshold,
and
so
you
can
calculate
this
sort
of
activity
and
then
have
that
activity
in
two
different
neurons
that
have
some
connection
leads
to
their
coupling
their
their
act
activation
coupling,
and
so
there
are
a
lot
of
ways
you
can
do.
A
A
This
is
something
you
could
apply
to
a
neural
network
as
well,
and
in
fact,
a
lot
of
the
activation
functions
for
neural
networks
are
not
this
complex,
but
they
have
different.
You
know
a
structure,
that's
not
too
dissimilar
from
this.
They
don't
usually
use.
You
know
you
will
use
differential
equations,
but
in
any
case
so
then
they
have
this
framework
for
network
resilience.
A
A
So
basically
the
connectum
comes
in
this
matrix
of
you
know,
values
you
have
all
it's
sort
of
like
an
all
to
all
matrix
where
you
have
pairwise
comparisons
and
the
pairwise
comparisons
have
some
value.
You
know
it
could
be
like
a
correlation
coefficient.
It
could
be
some
sort
of
strength
coefficient,
so
you
know
a
weight,
so
these
are
all
like.
This
is
depending
on
the
data
set
that
you
get
now.
A
A
Christiankus
is
a
little
bit
more
diffused
than
c
elegans
c
elegans
more
tightly
connected
in
this
area.
I
don't
know
what
that
means.
That
could
mean
that
it's
you
know
that
could
have
some
difference
between
predation
and
being
a
prey.
It
could
just
be
because
that
species
you
know
there
are
different
behavioral
differences
in
terms
of
what
they're
doing
their
behavioral
repertoire,
but
they
say
here
that
they
can
detect
these
differences.
A
You
know
that
are
due
specifically
to
predator
and
prey,
but
there
are
other
differences
as
well
in
these
connectomes
and
just
to
show
an
example
of
how
the
connectomes
are
connected.
You
have
these
hubs
and
you
have
these
peripheral
nodes
and
the
there's
a
functional
distinction
where
the
hubs
are
taking
in
a
lot
of
connections
from
different
places,
or
they
have
you
know
very
strong
connections
with
certain
neurons
and
others
are
not
as
strongly
connected.
So
you
know,
that's
something:
that's
functionally
you
know
has
some
functional
distinction
too.
A
Some
cells
are
interneurons
where
they're
integrating
a
lot
of
things
from
motor
neurons
and
sensory,
neurons
and
other
you
know.
Other
cells,
are,
you
know
just
sensory
neurons,
sensing,
one
part
of
the
environment
or
something
like
that
and
they're
feeding
forward
their
behavior
so
or
their
activity.
So
that's
why
they're
structured
in
that
way
so
yeah
we
have.
A
I
think
this
is,
I
don't
think
there's
so
this
is
the
hermaphrodite
versus
the
male.
This
is
the
hermaphrodite
where
you
have
this
sensory
neuron
and
then
you
go
to
the
muscle
and
in
this
case,
which
is
the
male,
I
believe
this
is
where
you
have
a
model
of
the
interneurons
where
you
have
sex
specific
cells,
sensory
neurons,
and
they
go
to
the
muscles.
So
there's
this
difference
in
the
hermaphrodite.
You
just
have
the
sensory
neurons
in
males.
A
A
Okay,
so
I'm
going
to
get
into
some
other
a
couple
other
papers.
I
just
wanted
to
talk
about
some
of
this.
Let's
see
that
was
one
okay.
I
just
want
to
make
sure
I
go
through
some
of
these,
so
this
part
of
the
meeting
I'm
going
to
just
go
through
a
couple
more
papers
and
then
we'll
wrap
up,
but
I
want
to
get
through
this
list.
You
can
see
we
have
this
huge
list.
A
So
the
first
one
is
this
program
flow
paper,
and
this
is
the
title
is
programmed
flow
programmed
and
self-organized
flow
of
information
during
morphogenesis,
and
this
is
about
morphogenesis,
and
so
the
abstract
reads
how
the
shape
of
embryos
and
organs
emerges
during
development
is
a
fundamental
question
as
fascinated
scientists
for
centuries.
Of
course
we
talk
about
that.
All
the
time
in
the
group
tissue
dynamics
arise
from
a
small
set
of
cell
behaviors,
including
shape
changes,
so
these
aren't
like
neurons
and
their
interactions
with
other
neurons.
A
These
are
cells
that
are
undergoing
morphogenesis
in
an
embryo,
and
so
there
are
a
whole
different
set
of
interactions
here
so
shape
changes,
cell
contact,
remodeling
cell
migration,
cell
division
cell
extrusion.
So
these
are
things
that
our
cells
move.
They
divide,
they
call
extrusion,
which
means
that
they're
things
that
are
coming
out
from
like
a
flat
surface,
so
cells
are,
you
know,
taking
on
sort
of
a
unique
shape,
as
opposed
to
being
a
sphere.
A
So
they
talk
about.
Last
week
we
talked
about
gene
regulatory
networks,
and
so
in
this
case,
what
they're
doing
is
they're
linking
gene
expression
and
biochemical
cues
to
some
of
the
mechanics
and
geometry
that
act
as
a
source
of
morphogenetic
information,
and
so
this
actually
works
at
different
time
and
lengths
get
what
they
call
length
scales,
which
are
spatial
distances.
A
So
the
first
model
is
a
program
which
is
like
a
differentiation
program,
and
so
this
you
could
think
of
this
as
like
something
that's
programmed
into
the
genome
and
then
expressed
or
something
that
is,
you
know
like
follows
a
clock
in
every
embryo.
This
kind
of
model
follows
deterministic
rules
in
this
hierarchical.
A
The
second
model,
however,
follows
the
principles
of
self-organization,
which
rests
on
statistical
rules,
characterizing
the
system's
composition
and
configuration
local
interactions
and
feedback.
So
the
first
one
is
like
a
program.
That's
you
know
a
top-down
sort
of
system,
so
you
know
imagine
that
you
know
you
have
some
gene,
that's
a
like
a
clock,
gene
that
tells
the
embryo
when
it's
time
to
sort
of
you
know
differentiate,
or
you
know
just
you
know,
exhibit
some
sort
of
symmetry
breaking.
A
That's
that's
a
very
seductive
model,
but
it
doesn't
really
follow
that
this.
All
this
is
programmed
into
the
embryo.
Now
in
c
elegans
there
seems
to
be
a
lot
more
of
this
programming
programmatic
behavior
than
in
other
embryos,
where
you
have
a
lot
of
cell
to
cell
signaling.
But
the
second
rule
follows
the
principles
of
self-organization,
so
in
some
forms
of
development,
this
actually
is
much
more
common.
A
Where
you
have
cells
that
are
signaling
one
another
and
you
can
have
differentiation
occur
spontaneously,
as
opposed
to
like
at
some
point
in
development,
like
you
see
more
with
c
elegans.
Actually
so
these
two
modes
to
the
mechanisms
of
four
very
general
quests
of
tissue
deformation
are,
you
know,
discussed.
So
you
have
these
four
different
types
of
tissue
deformation
and
you
have
the
two
models
that
so
you
have
deterministic
self-organization
and
then
you
have
these
four
different
types
of
tissue
manipulation.
A
A
I,
like
you,
know
a
lot
of
these
reviews
that
kind
of
go
through
some
of
these
ideas
that
that
influence.
What
they're?
Talking
about
so
here
in
box,
one
is
the
mosaic
and
regulative
theories
of
development.
So
we've
talked
about
this
in
the
meetings
where
there's
this
mosaic
theory
of
development,
which
is
based
on
this
programmatic
approach,
where
you
know
you
have
these
cells
that
sort
of
think
the
events
that
unfold
in
development
and
they're
likely
encoded
in
the
genome.
A
So
you
know
you
have
these.
You
know
where,
if
you
take
a
frog
embryo
and
you
kill
with
a
hot
needle
half
of
the
two
cell
stage
or
the
four
cell
stage,
you
end
up
with
half
an
animal.
So
you
end
up
with,
like
you
can't
like
form
the
entire.
We
talked
about
flat
worms
last
week
and
we
said
that
every
cell
in
the
flatworm
is
totipo,
meaning
that
any
cell
can
form
an
entire
ember
and
form
an
entire
organism
on
its
own.
A
In
this
case,
this
is
the
opposite
of
that.
It's
that
every
cell
has
to
be
in
place
and
if
you
knock
out
half
of
the
cells
very
early
in
development,
you
end
up
with
half
an
animal
half
of
rock.
It's
an
interesting
way
to
you
know
it,
but
this
happens
in
c
elegans
too.
If
you
take
the
two
cell
stage
embryo
and
you
divide
the
you
take
the
cells
apart
and
you
grow
them
independently
in
a
culture.
You
end
up
with
really
nothing
the
cells
divide,
but
they
don't
really
form
anything.
A
A
A
You
know
they
can
form
different
adaptive,
phenotypes,
and
so
that's
what
that
is,
and
then
they
kind
of
get
into
morphogenetic
information
and
some
other
things
as
well.
So
two
two
papers
that
follow
up
on
that
is
the
self-organized
tissue
regulation
paper
and
susan
I'd
be
interested
in
these
references.
So
I
want
to
send
them
along
to
her,
but
this
paper
focuses
on
amnio
development.
A
It's
just
just
amniotic
development,
which
is
where
you
have
live
births,
any
animal
that
can
give
live
birth
and
this
is
a
highly
regulative
and
self-organized
process.
So
here
we're
not
talking
about
organisms
that
have
this
programmatic
type
of
development,
but
other
types
of
you
know
amniotic
organisms
that
have
like
us
or
like
mice
that
have
this
regulative
and
self-organized
type
of
development,
and
so
this
paper
is
this
kind
of
talks
about
self-organized
tissue,
mechanics
and
regulation.
A
They
talk
about
gene
expression,
but
only
sort
of
obliquely,
so
they
talk
about.
You
know
in
this
case
they're
talking
about
analyzing,
intact
and
mechanically
perturbed
avian
embryos.
We
show
that
mechanical
forces
that
drive
embryogenesis
self-organize
in
an
analog
of
turns
molecular
reaction,
diffusion
model
so
they're
using
this
model
of
turns
molecular
reaction,
diffusion
they're,
using
an
analog
of
it
with
contractility,
locally
self-activating
in
the
ensuing
tension
acting
as
a
long-range
inhibitor.
A
These
mechanical
feedbacks
govern
persistent
patterns
of
tissue
flows
that
shape
the
embryo
and
steers
the
contaminant
emergence
of
embryonic
territories
by
modulating
gene
expression.
So
they
talk
about
gene
expression,
only
obliquely.
They
talk
about
it
in
terms
of
what
the
forces
are
doing
and
how
they're
affecting
gene
expression.
A
So
you
know
there
are
a
lot
of
ways
you
can
affect.
Gene
expression
by
you
know
the
cell
doing
behaving
doing
things
I
misspoke
about
amniotes
they're
not
actually
live
births.
They
have
this
amniotic
sac,
so
birds
that
lay
eggs
also
have
this
their
amniotes
as
well.
So
just
a
clarification
there
I
didn't
want
to,
but
in
any
case
it's
it's
not
the
same
thing
as
like
the
programmatic
development.
A
So
this
is
the
yeah
this.
This
is
kind
of
a
nice
paper
that
kind
of
lays
out
a
model
for
sort
of
this
mechanical
aspect
of
development
and
then
kind
of
getting
into
how
they
relate
to
gene
regulation,
but
not
so
much
so
that
it
focuses
on
genes
just
kind
of
how
it
affects
some
of
these
regulators
and
then
their
activity.
A
So
here
you
can
see
an
example.
Their
model
is
that
you
have
these
forces
contractility
tension.
You
have
the
expression
of
gdf1
here
in
this
case
where
contractility
is,
is
driving
it
tension
is
inhibiting
it,
and
then
you
have
embryo
formation,
and
you
can
see
here
that
there's
this
relationship
between
tension
and
contractile
contractility,
which
is
where
you
have
two
different
mechanical
forces
interacting
and
then
it
you
know
it
describes
this
reaction
diffusion
model
where
you
have
regulation
of
what
the
tissue
is
doing
in
this.
In
this
space.
A
Yeah
I'll
send
you
all
these
in.
You
know
at
the
end
of
the
meeting
and
then
finally,
this
dynamic
transmission
of
positional
information
in
drosophila
embryogenesis.
So
this
is
moving
now
back
to
drosophila,
which
is
the
you
know.
Our
fruit
fly
model,
and
this
is
this
type
of
development
is
more
programmatic
and
there's
a
mix
a
lot
of
times
in
organisms.
There's
a
mix
between
the
programmatic
type
of
development
and
the
regulative
type
of
development.
A
A
Approach-
and
it's
you
know
so
you
can
like
track
cells.
They
you
know,
there's
this
cellularization
phase,
so
it's
an
interesting
model
from
rheogenesis,
it's
very
different
from
c
elegans,
but
also
very
different
from
see
like
humans
and
birds
and
even
fishes.
So
so
this
is
the
abstract
here.
It
is
suggested
that
stuff
and
style
is
key
in
controlling
the
variability
of
the
posterior
boundary
of
the
hb
anterior
domain.
A
A
You
know
they
don't
know
what
controls
this
process,
but
they
know
that
there's
this
sort
of
spatial
restriction
in
this
part
of
the
posterior
embryo-
and
they
know,
there's
some
sort
of
control
by
this
factor,
but
they
don't
know
what
the
mechanism
is.
So
here
we
quantified
the
dynamic
3d
expression
of
segmentation
genes
and
drosophila
embryos.
A
So
again,
this
is
where
you
have
genes.
We
talked
about
hox
genes,
where
you
have
like
the
spatial
restriction
of
their
expression,
and
so
they
get
expressed
in
different
segments
of
the
embryo
and
drosophila
is
a
very
good
model
for
this,
because
you
can
actually
see
this
process
unfold
very
clearly
and
so
the
but
people
don't
really.
A
A
You
know
express
one
gene
over
another,
or
is
it
this
that
there's
a
lot
of
gene
expression
product?
That's
you
know,
sort
of
produced,
and
then
you
know
in
a
gradient
and
then
the
gradient
is
sort
of
restricted
by
interactions
amongst
genes
or
amongst
you
know,
cells
are
making
different
concentrations
of
it
and
there's
this
sort
of
order
that
happens
at
the
level
of
the
embryo,
so
that
you
know
that's
that's
the
question
that
we're
asking
and,
of
course
people
will
develop
models
for
this.
A
So
there's
the
reaction
diffusion
model
which
can
create
striping
and
there
are
other
types
of
models
that
can
do
the
same
sort
of
thing
in
this
case
they're
interested
in
something
called
positional
information,
which
is
where
each
cell
has
like
some
sort
of
positional
information.
They
know
in
the
embryo
where
they
are,
and
so
from
that
positional
information
you
can
say
this
is
you
know
what
I
should
be
expressing
and
then
there's
the
striping
that
occurs
so
they're
able
to
quantify
the
3d
expression
of
segmentation
genes
with
improved
control
of
measurement
error.
A
A
A
This
very
is
this
variability.
Over
short
time,
windows
is
comparable
to
that
of
the
wild
type.
Moreover,
style
negative
mutants,
the
upstream
liquid
gradients,
which
are
these
gradients
of
expression,
show
equivalent
relative
intensity
noise
to
that
of
the
wild
type
and
down
in
the
downstream.
Even
skipped
and
cephalic
furrow
show
that
the
same
positional
errors
as
these
factors
in
the
wild
type.
So
these
even
skipped
and
cephalic
furrow
or
other
genes
that
are
being
expressed
in
different
parts
of
the
embryo
and
different
stripes.
A
And
if
you
you
know,
if
you
modify
one
gene
one
factor
here,
these
other
genes
will
be
shifted
in
different
ways
and
so
there's
this
positional
information
that
is
essentially
disrupted
and
shifted
across
the
embryo.
So
this
is
interesting.
This
is
an
interesting
link
between
positional
information
and
the
expression
of
different
factors
that
are
being
expressed
in
the
cell
and
actually
just
one
factor,
because
this
negative
mutant
is
showing
all
of
these
different
positional
shifts.
A
A
I
don't
know
if
they
show
any
images
here,
but
this
is,
you
know,
a
good
model
of
embryogenesis,
and
you
know
this
is
a
nice
model
because
it
allows
you
to
investigate
a
lot
of
this.
You
know
gene
expression
and
this
connection
to
positional
information,
and
so
that's
basically
what
they
found
there.
So
you
know
that's
that's
a
sort
of
support
for
this
positional
information
hypothesis
and
you
know
how
it's
modified
when
you
change
the
location
of
genes.
You
also
so.
A
A
You
know
this
gene
should
be
expressed
in
this
location
and
so
forth
in
a
self-organized
model.
You
could
imagine
that
there's
more
this
regulative
mode,
where
you
know
you
don't
really
have
a
lot
of
positional
information,
at
least
it's
local,
it's
not
global,
so
I
mean
you
know.
There
are
different
ways
that
you
can
look
at
embryo
development.
Oh
nick
is
here
hello,
dick
and
I
know
I've
been
wearing.
A
I
just
haven't
gone
back
to
the
throne.
Okay
did
that
estimates
that
I
had
another
meeting,
would
they
have
to
drop
off?
This
looks
really
interesting,
be
great
if
you
could
share
the
papers
on
slack
I'll.
Do
that.
I
also
share
them
with
susan
because
she
requested
them,
and
then
richard
says,
sharks
and
guppies
of
live
births.
Okay,
yeah,
so
I
yeah.
H
F
A
A
I
know
there
was
a
lot
of
papers
and
things
I
wanted
to
go
through
some
of
those
and
crunch
through
them.
So
I
know
last
week
we
talked
about
the
sort
of
the
gene
regulatory
networks
of
eric
davidson
and
we
kind
of
got.
We
didn't
really
wrap
that
up,
we
kind
of
said
well
what
I
said.
I
proposed
that
we
need
some
sort
of
experimental
system
for
computational
gene
expression
networks,
and
so
this
is
something
that
I
might
talk
about
in
future
meetings.
But
that's
you
know
there's
this.
A
Good
computational
models
for
and
now
experimentalists
are
developing
computational
models
for
like
their
model
system,
but
to
have
like
a
first
principles.
Type
models,
which
means
first
principles,
means
you're,
developing
it
from
some
principle
that
you
propose
by
looking
at
all
these
organisms
and
saying.
I
think
this
is
how
it
works.
Let's
do
this
and
then
you
try
it
and
you
know
it
works
well
or
it
doesn't,
and
so.
G
A
Yeah
yeah,
I
think
I've
heard
of
this
like
where
they
do
this
in
regenerative
medicine
where
they
have.
You
know
they
have
like
the
cell
and
it's
in
in
some
they'll,
put
like
a
stem
cell
in
some
location
in
like
an
organ
and
like
it
has
to
be
in
that
niche.
They
call
it
a
niche
and
then
like,
if
it's
not
in
there,
it
won't
diff
if
it's
in
the
niche
it'll,
maybe
differentiate
into
the
cell
type
and
it'll
maintain
its
phenotype,
but
if
it
doesn't
have
all
the
cues
that
it
needs.
A
E
A
F
A
All
right,
yeah,
dick
earlier
we
were
talking
about
some
of
the
it's
actually
esme
and
uzuwal.
Who
did
this
original
bacillaria
paper
that
we
did
and
then
thirune
who's?
Another
person
from
who
joined
the
group
this
year,
they're
working
on
the
basilaria
data
they're
taking
another
shot
at
it
and
they're
trying
to
find
like
they're,
trying
to
build
a
good
model
for
motion
so
they're
looking
at
they're,
you
know
trying
to
do
a
they're
using
a
certain
algorithm
that
should
pick
up
a
lot
of
the
motion.
Cues.
E
H
F
A
Okay,
I
think
that's
it
for
today,
thanks
for
attending
and
welcome
karam,
and
if
you
have
any
questions
you
know
we
can
talk
about
an
email
or
slack
and
I'll
send
those
papers
out
to
susan,
and
you
know
maybe
yeah.
We
can
continue
this
discussion
a
bit.
I
yeah
we'll
have
to
look
into
those
papers
on
the
steve
smith's
papers
and
maybe
some
stuff
on
blood
vessels.