►
From YouTube: DevoWorm (2022, Meeting 4): Tensegrity in cells, Sea Urchin and Neural CAs, fitness landscapes
Description
Tensegrity structures in cells and tissues, tensegrity networks, image processing of sea urchin skeleton, Neural CAs and skeletal pattern formation, follow-up on protein structures and hydrophobicity, computation of fitness landscapes. Attendees: Mainak Deb, Richard Gordon, Karan Lohaan, Susan Crawford-Young, Sarkath Jain, and Bradly Alicea.
A
A
C
A
C
A
A
A
A
A
C
C
C
C
C
E
C
A
Yes,
yeah,
that's
it
it's
in
this
review
paper,
oh
okay,
yeah
and
that's
nice,
because
I
just
needed
to
hand
hand
dr
zhang
to
dr
grief,
one
paper
and
I
said
and
they're
saying:
well,
we
don't
really
want
to
get
involved
with
all
this
stuff.
This
is
dr
shree
and
I
said
just
quoting
birth.
This
paper.
C
C
A
A
C
A
So
we
have
to
add
that
that's
why
I'm
trying
to
use
console
multi-physics
curve
for
my
check,
for
my
I
don't
know
how
it
does
with
linear
elements,
though
yeah.
Well,
these
are
not
linear
elements.
This
is
a.
This
is
a
microtubule,
and
this
is
well
microtubules
and
mechatronics
are
usually
linear,
not
always,
but
well,
these
spectrum
enough.
They
are
and
microtubules
tend
to
break,
and
you
know.
C
C
A
A
C
A
C
A
C
A
C
A
A
So
there
we
go
and
I
didn't
come
up
with
this.
I
actually
asked
dr
zhang
what
what
what
should
I
use
for
this
like
there's,
three
of
them,
the
regular
continuum,
model,
integrity
or
active
matter
formulas,
and
he
said
10
seconds
he
just
went
nailed
it.
He
didn't
even
like
finish
it.
He
didn't
think
he
just
went
straight
for
it
yeah
yeah,
so
he
decided.
C
A
C
A
So
this
is
yeah.
This
is
where
I
proposed
my
project.
This
is
the
glassy
dynamics
where
it's
solid
and
then
fluid,
and
then
here
I'm
just
going
to
look
at
this
t2
transition
where
there's
apoptosis
and
if
you
get
apoptosis,
you
get
these
rosettes
happening,
and
this
makes
the
tissues
susan
explain
what
apoptosis
says.
A
A
A
A
A
C
C
A
C
C
A
A
A
A
A
So
that's
that's
my
challenge
right
here
and
it
may
not
work,
but
that's
what
dissertations
are
about
apparently
you're
supposed
to
do
something.
That's
not
your
thesis
project
so
right
and
see.
If
you
can
use
this
technique
for
your
thesis,
so
I'm
yeah
I
get
to
do
some
rather
complicated,
algebra
and
maybe
end
up
with
a
gi,
mungus
matrix.
B
C
B
C
A
A
B
C
A
B
B
B
F
F
A
E
F
F
F
The
value
of
every
pixel
is
basically
the
distance
of
that
pixel
to
the
to
the
nearest
point.
So
if
so,
if,
if
that
pixel
is
far
away
from
the
nearest
void,
its
value
is
higher.
It's
close
like
if
the
pixel
is
only
one
pixel
away
from
the
white,
it's
its
color
is
mapped
as
one
and
if
the
pixel
is
four
pixels
away,.
F
F
F
It's
essentially
the
essentially
the
distance
from
the
nearest
void
or
the
background.
So
if
you,
if
we
over,
if
we
select
pixels
from
it,
which
overlapped
the
with
the
medial
axis
skeleton,
if
you
only
consider
the
skeleton
pixels
from
this
distance
map,
we
get
something
that
looks
like
this
I'll
zoom
into
it
yeah.
So
here
the
skeleton
pixels
are
mapped
to
the
nearest
void
distances,
I'm
not
sure
if
I'm
able
to
put
it
correctly
but
yeah.
So
this
looks
like
the.
F
But
the
pixel
values
are
mapped
as
the
distance
to
the
nearest
void
and
this
basically
and
this
when
combined
with
the
node
map
that
enables
me
to
find
the
average
width
of
every
branch
so
moving
on.
I'm
also
exporting
this
data
in
form
of
a
csv
file
which
looks
like
yeah.
So
this
is
the
csv
file
it
it
contains.
So
let's
just
ignore
the
first
two
columns,
because
I
forgot
to
remove
them
and
it
contains.
F
And
it
contains
the
branch,
distance,
okay,
so
I'll
start
from
the
I'll
start
from
the
beginning.
Here
again,
so
this
this
data,
basically
every
row,
it
describes
a
branch
which
has
a
which
has
an
initial
node
and
and
a
destination
node.
So
here
we
have
the
branch
type,
which
is
the
branch
types,
are
mapped
between
values
of
one
two
and
three,
which
are
basically
like
if
it's
an
end
to
node
branch
or
if
it's
a
node
to
node
branch
like
there
are
a
couple
of
branch
types
which
were
there
right
sort
of.
F
I
don't
remember
what
the
branch
type
encodings
are
they're
like
okay
I'll
get
to
that
later.
I
have
the
text
so
anyways,
so
the
next
columns
they
show
the
x
and
y
coordinates
of
the
initial
branch
and
of
the
initial
node
of
that
branch.
And
this
and
these
two
columns
they
represent
the
the
endpoint
xy
coordinates
of
the
branch.
And
then
we
have
the
euclidean
distance
of
the
branch,
which
is
basically
the
straight
line.
F
Distance
between
the
end
point
and
the
end
point
and
the
initial
point:
we
have
the
slope
in
two
dimensions
and
we
also
have
the
torque
to
your
city,
which
is
the
ratio
between
the
uterine
distance
and
the
branch
distance.
So
one
subtle
difference
is
that
the
euclidean
distance.
As
I
said,
it's,
the
straight
line,
distance
between
the
initial
and
the
final
point
of
the
branch
and
the
branch
distance
is
the
curved
distance.
It
traces.
A
E
F
This
is,
this
was
just
one
sort
of
structure
that
we
are
using
as
a
2d
sample,
and
this
exact
procedure
would
be
carried
out
in
3d.
So
this
skeleton
that
we
are
saying
we
would
be
getting
a
3d
skeleton
there
and
you
would
be
running
mechanical
simulations
on
them
from
the
from
the
we'll
be
using
the
parameters
like
the
brass
thickness
and
all
that
would
be
putting
them,
basically
plugging
them
into
the
system
into
the
mechanical
simulation
and
we'll
be
comparing
what
happens
and
yeah.
So
this
was
this
is
about
it.
I
am.
F
But
I'm
not
really,
I
didn't
really
get
very
far
with
it,
so
I'll
be
presenting
that
in
the
upcoming
meetings-
okay,
so
yeah-
that
was
it
actually
for
this
topic.
But
apart
from
this
bradley,
I
had
a
little
more
like
there
was
a.
There
was
another
thing
that
I
just
wanted
to
take
a
couple
of
minutes
to
show:
okay
yeah
yeah.
F
F
This
paper
and
what
this
does
to
to,
to
put
it
very
simply,
it
uses
a
neural
network
to
emulate
textures
in
form
of
a
neural
cellular
automata.
So
what
I
did
is
that
I
used
this
sea
urchin
data.
I
use
the
raw
image
of
this
thresholded
image.
I
use
the
raw
version
of
this
hold
on.
Let
me
see
if
I
can
find
the
image.
F
B
F
F
Yeah
so
yeah,
so
these
are
the
kind
of
textures
that
we
get
after.
We
use
the
neural
cylinder
tomato
model
too
yeah,
so
these
colors
they
actually
came
from
these
colors.
I
basically
sort
of
played
around
with
the
neural
network,
and
I
could
get
these
get
these
interesting
outputs,
which
of
course,
they're
they're
coming
from
the
raw
data.
A
F
A
F
And
there
too,
we
got
some
interesting
results,
and
this
this
almost
looks
like
water
bubbles,
actually,
which
was
interesting
because
water
bubbles
they
also
follow
the
voronoi
pattern
and
as
as
dr
gordon
talked
about
so
this.
This
comes
from
the
neural
cellular
automata
model
itself,
and
I
actually
did
this
for
an
upcoming
exhibition,
which.
F
Which
deals
with
everything
that
is
related
to
equinox
and
serious
chin,
of
course,
is
a
big
part
of
it.
So
we
get
these
interesting
results.
It's
sort
of
some
part
of
it
like
parts
of
it,
looks
sort
of
psychedelic
the
colors
they're
very
interesting,
yeah.
Okay,
I'm
sort
of
loading
for
the
entire
video
to
load.
F
E
F
D
C
C
C
C
F
C
F
C
F
C
Could
help
improve
the
integrity
of
the
structure
yeah,
the
spines
can
actually
move
and
they
can
actually
move
the
whole.
They
can
move
together
in
an
organized
fashion
and
move
the
whole
sea
urchin
over
a
surface,
so
they
have
to
be
controlled
from
some
place
either
outside
or
inside
the
that
shell.
C
C
B
Well,
thanks
a
lot,
my
knock
for
presenting
on
that.
That
was
great
stuff,
glad
that
someone's
taking
the
neural
cellular
automatic
and
looking
at
it
a
little
bit
more
because
we
talked
about
that
in
previous
meetings.
It
never.
I
mean
we
had
a
little
bit
of
interest
in
it,
but
it
was
like
I'm
just
interested
to
see
what
can
be
done
with
that.
You
know.
F
A
G
G
G
So
when
dealing
with
the
structure
like
this
like,
I
could
find,
at
least
you
know
when
we're
dealing
with,
let's
say
parallel
parallel
beta
sheets.
We
can,
you
know,
find
out
the
hydro
phobicity
by
checking
out
the
contact
angle
of
by
let's
say,
if
we
have
another
water
droplet
placed
on
that
beta
sheet.
G
E
G
Some
minor
structural
changes,
but
the
overall
structural
changes
of
that
large
protein
would
not,
I
think,
depend
that
much
unless
you're
doing
that
with
the
whole
protein
structure,
because
unless
we
have
like
major
structural
changes,
so
I
was
kind
of
like
if
we
could,
you
know,
go
over
some
examples
that
would
that
would
kind
of
help
me.
You
know
narrowing
down
the
answer
like.
Are
we
dealing
with
like
really
big
peptide
changes,
or
do
we
have
some
sort
of
examples
when.
C
The
length
of
the
membrane
embedded
portion
is
typically
between
19
and
26
amino
acids:
okay,
okay
and
I'm
not
interested
in
the
rest
of
the
mouthful
okay.
Okay,
so
you
can
just
look
at
the
transmembrane
section.
However,
what
I
am
interested
in
is
the
hydrophobicity
properties
of
the
what's
called
the
transmembrane
portion.
G
G
Okay,
yeah,
because
I
could
find
some
papers,
you
know
they
would.
They
were
kind
of
dealing
with
again
short
peptide
change
only,
but
they
had
this.
You
know
general
thought
process
where
you
know
converting
n2d
or
the
stereoisomers
tend
to
have
more
hydrophobic
properties
compared
to
n
stereoisomers.
G
C
C
E
C
B
B
B
B
Waddington
landscape
or
a
epigenetic
landscape,
and
it's
this
landscape
where
you
have
the
surface
and
it
has
like
a
surface
and
it
has
like
a
like
a
topography
to
it
and
it
consists
of
what
they
call
valleys
or
channels,
and
then
you
have,
but
but
it's
a
landscape,
so
things
proceed
in
a
certain
direction.
So
it's
from
like
the
top
to
the
bottom,
and
you
know
it's
kind
of
like
an
energy
minimization
thing
where
you're
trying
to
minimize
the
potential
of
some
type.
So
you
know
there's
this
potential.
B
It
goes
down
to
the
bottom.
You
know
from
the
beginning
of
development
to
the
end
of
development
and
those
valleys
that
you
have
as
your
canals
branch
off
so
that
when
there
are
different
points
of
like
differentiation,
they
branch
off
and
you
end
up
at
the
other
end
with
a
bunch
of
wells
that
are,
you
know,
embody
that
landscape,
and
so
you
know
there
there's
a
lot
more
to
it.
You
can
look
at
the
underbody
of
it
and
see
that
you
know
maybe
there's
a
tree.
E
B
Or
that
there's
gene
expression,
that's
undergirding
all
of
these
pathways,
and
so
you
know
this
is
mostly
a
conceptual
thing,
and
people
have
done
like
quantitative
analyses
of
them
in
evolution,
though
they
have.
These
things
called
fitness
landscapes
which
are
similar
and
they
have
a
similar
origin
as
these
metaphors.
B
B
B
B
A
B
Started
out
as
a
paper
by
sewell,
wright,
1932
and
the
paper
in
the
original
paper,
they
had
soil
writed
done
drawings
of
different
two-dimensional
landscapes,
so
the
two-dimensional
landscapes
had
these,
like
you,
know,
hills
and
valleys
that
were
sort
of
like
shaded
air
regions
and
non-shaded
regions,
and
then
that
was
the
idea.
You
know
it
was
very
simple,
and
people
have
thought
about
this
more
and
they
thought.
Well.
You
know
fitness
is
actually
multi-dimensional.
B
So
you
know
you
have
this
multi-dimensional
landscape
that
we
can
reduce
down
to
three
dimensions.
So
when
you
often
you
see
fitness
landscapes,
they'll
look
like
three-dimensional
maps
like
this,
so
you
can
see
that
in
the
fitness
landscape.
Here
you
have
this.
This
set
of
hills,
and
then
you
have
this
genotypic
space
here,
which
is
every
point
on
this.
Landscape
is
a
point
in
genotypic
space,
so
it's
some
distinct
genotype,
and
so
then
what
happens
is
that
populations
have
to
hill
climb
to
increase
their
fitness.
B
You
know
you
can
have
multi-dimensional
spaces
for
just
down
to
three
dimensions
and
the
like,
but
in
this
case
they're
actually
talking
about
using
mathematical
techniques
to
understand
the
structure
of
these
fitness
landscapes,
so
fitness
landscapes
are
being
analyzed
by
constructing
genotypes
with
all
possible
combinations
of
small
sets
of
mutations
observed
in
phylogenys
or
in
evolution.
Experiments
in
turn.
These
first
glimpses
of
empirical
fitness
landscapes,
inspire
theoretical
analysis
of
the
predictability
of
evolution,
so
you
can
actually
predict
evolution
by
sort
of
understanding
these
landscapes
and
what
these
genotypes
look
like
as
they're.
C
C
B
B
So
as
you
like
evolve
and
you're
interacting
with
you
know,
like
you,
have
these
predator
prey
dynamics
where
you
have
some
interaction
with
the
environment,
the
landscape
is
changing,
and
so
your
position
on
the
landscape
is
always
in
flux,
and
it's
hard
to
really
capture
that
on
something
like
this,
and
you
know
it's
a
common
criticism
of
the
type
of
model.
So
there
are
a
lot
of
assumptions
you
have
to
make
here
about,
like
you
know,
isolating
the
the
evolutionary
dynamics,
whether.
B
Right,
whether
that's
like
something
that
we
should
be
doing,
I
don't
know
because
I
mean
you
know
it
may
be
that
that
doesn't
make
much
sense,
but
it
you
know
so
yeah,
so
so
all
right.
Actually,
this
is
an
example
of
the
type
of
map
that
soul
sulrite
drew
back
in
his
original
paper,
and
so
these
are
just
kind
of
like
these
two-dimensional.
B
You
know
density
plots
where
you
have
like,
where
they're
actually
topographic
plots,
if
you've
ever
seen
a
map
where
they
show,
like
you,
know,
elevation,
but
this
is
basically
it.
You
know
these
are
the
peaks
with
the
plus
sign.
These
are
valleys
of
the
minus
sign,
and
then
these
are
the
neutral
areas
and
the
idea
would
be
it's
a
you
know:
it's
a
topography,
you
climb
or
dip
down
into
a
valley,
and
so
this
is
and
then
this
is
a
development
of
the
fitness
landscape
concept.
B
B
So
this
figure
c
here
is
the
16
genotype
network
and
it's
going
from
the
zero
zero
zero
zero
state
to
one
one,
one
one
state
and
you
can
see
that
there's
this
sort
of
resistance,
so
you
have.
This-
is
wild
type
and
mutant
amino
acids
respectively.
So
the
all
ones
are
a
mutant,
phenotype
of
amino
acids
and
all
zeros
is
this
wild
type.
B
If
you,
you
know,
if
they
want
to
look
at
say
something
like
amino
acid
phenotype,
so
use
a
hypercube
which
is
where
they
start
at
this
zero
state
and
they
try
to
evolve
the
phenotype
to
this
one
state
and
they
have
this
structure
that
gives
you
all
possible
combinations
between
zero,
zero,
zero,
zero
and
one
one
one.
And
then
the
idea
is
to
see
how
many
mutational
steps
it
takes
to
get
from
one
to
the
other.
B
So
you
can
actually
look
at
like
what
they
call
evolvability,
which
is
you
know
how
easy
or
hard
is
it
to
go
from
one
state
to
another?
And
if
you
have
information
about
your
phenotype
space,
you
know
if
you
have
like
mutation
rate
information
or
some
other
information.
If
you
don't
assume
it's
all.
Basically,
you
know
just
equal
problem
equi-probable.
B
B
There
are
all
sorts
of
things
that
can
play
a
role,
and
so
this
is
just
a
model
to
understand
this
better.
It's
not
you
know
perfect,
and
there
are
a
lot
of
problems
with
representing
gene
or
phenotypes,
with
a
binary
string,
for
example,
and
that
you
know
it
works
well
with
with
rna
molecules
or
amino
acids.
B
B
So
there's
this
you
know
these
additive
effects
and
these
multiplicative
effects
of
genes
acting
together,
so
most
phenotypic
traits
are
not
just
one
gene
making
some
phenotypic
trait.
It's
usually
an
interaction
of
genes
making
that
trait.
So
you
know
you
might
have
your
some
trait,
that's
very
seems
very
simple
and
straightforward
and
it
might
have
10
genes
underlying
it
and
then
also
the
environment.
B
B
B
B
B
B
Minus
1,
which
is
l,
is
the
total
number
of
loci,
and
then
this
just
shows
you
how
those
all
sort
of
interact
and
then
there's
the
rough
mount
fuji
model,
which
is
obtained
by
combining
a
house
of
cards
model
with
an
additive
landscape.
So
they
use
this
additive
epistasis
to
model
this,
so
there
are
all
different
ways
that
people
use
these
landscapes
with
respect
to
epistasis
and
looking
at
random
effects
or
quasi-random
effects
and
interaction
effects.
B
B
B
I
think
just
like,
I
think,
they're
using
like
experimental
evolution
here,
and
this
is
where
you
put
like
say
like
you:
can
you
know,
look
at
bacteria
or
something
you
can
culture
bacteria
over
a
number
of
generations,
and
you
can
use
c
elegans
or
this
as
well,
although
it
takes
longer
and
you
just
of
all
phenotypes
or
genotype.
Well,
you
evolve
genotypes
and
they
end
up
with
different
phenotypes.
B
So
in
bacteria
you
would
evolve
this
for
a
large
number
of
generations
and
it
generates
a
lot
of
variation.
You
can
look
at
the
then
look
at
the
genotypes
and
then
map
them
to
phenotypes
and
then
build
landscapes
and
you.
The
reason
you
might
want
to
use
experimental
evolution
is
because
it
provides
you
better
control
over
that
whole
process.
B
B
Yeah
yeah
yeah
there's
a
lot
of
there
are
a
lot
of
different
model
systems
for
experimental
evolution.
People
have
done,
I
think,
also
with
with
mice.
People
have
done
a
lot
of
interesting
things
and
like
with
running
behavior,
and
things
like
that
sometime.
I
should
give
a
presentation
on
experimental
evolution,
because
it's
really
fascinating
some
of
the
things
people
are
doing
with
that.
B
B
Can
you
you
know,
figure
out
kind
of
the
the
path
of
evolution
from
the
current
or
the
observed
state
of
of
genotypes
and
how
it's
evolving,
so
people
can
use
landscapes
to
do
these
kind
of
predictive
models
and
it's
a
little
bit
harder
than
you
know,
maybe
like
predicting
a
trajectory
of
a
baseball,
because
you
have
these
historical
contingencies
that
are
important
in
evolution.
So
in
other
words,
once
you
go
down
one
pathway,
you
can't
just
start
going
down
another
pathway,
just
randomly
you
know
somewhere
else.
B
You
have
to
kind
of
follow
what
you
have
and
then
the
options
that
you
have
in
front
of
you
are
come
from
your
evolutionary
history,
so
that
may
actually
make
it
easier
in
some
ways
to
predict,
but
also
you
know
it's
something
that
you
you
don't
typically
use
like
a
standard
statistical
model
to
do
that
yeah.
So
this
is
a
lot
of
trends
in
the
ruggedness,
so
ruggedness.
B
Yeah,
I
think
they're
yeah
they're
using
I
don't
know
what
they're
using
here
they're
yeah,
but
you
can
use
like
in
general
terms.
You
could
do
something
like
what
would
be
a
markov
model
with
memory.
That
would
be
good
and
then
this
ruggedness
issue
is
where
you
have.
These
landscapes
are
the
one
that
I
showed
you
up
in.
This
figure
is
smooth
where
it's
just
kind
of
these
smooth
peaks
but
oftentimes.
B
If
you
get
a
really
high
level
of
mutational
change,
you
get
these
rugged
landscapes,
which
are
kind
of
like
where
the
mountains
are
rugged,
and
you
have
a
lot
more
of
them
more
densely,
packed
into
the
space
and
kaufman
stu.
Kaufman
talks
a
lot
about
rugged
landscapes
and
how
those
are
sort
of
stand
in,
for
you
know,
increasing
complexity.
B
So
a
population
can
get
stuck
in
local
minima,
for
example,
or
it
can,
you
know,
have
to
cross
a
valley,
but
it
you
know
it's
hard
to
cross
a
valley
because
you
have
to
decrease
your
fitness
to
increase
your
fitness
later.
So
those
are
things
that
are,
you
know,
kind
of
difficult
you
you
get
into
these
rugged
landscapes.
When
you
get
really,
you
know,
complex
populations,
but
then
again
it's
harder
for
the
population
to
traverse
these
rugged
landscapes.
So
it's
an
interesting,
open
problem
there
too.
B
B
Well,
thank
you,
susan,
for
presenting
on
the
tensegrity
stuff.
I
know
it
wasn't
finished,
but
it's
pretty
interesting.
I
enjoyed
that
and
I
have
some.
I
think
I
have
some
ideas
about
networks,
but
I
don't
know
if
they're
really
relevant
or
not.
You
know
that
when
you
start
talking
about
these
networks
of
actin
filaments
and
things
in
cytoskeleton,
you
get
into
tensegrity,
but
you
also
get
into
things
in
graph
and
network
theory.
C
A
And
I
think
it's
sort
of
pre-stressed,
because
if
you've
got
something
that
undergoes
goes
a
cramp,
it's
it's
sort
of
like
the
cells
from
the
beginning
of
a
beating
heart
like
they.
They
wreck
their
microtubule
structure
when
they,
when
they
contract
like
they
do
it
regularly,
and
it
was
right
in
that
paper
from.