►
From YouTube: DevoWorm #23: GSoC Updates, Activity Review, Differentiation Trees, Beyond Waddington Landscapes
Description
GSoC Coding period (week 2): Updates on Digital Microspheres. Activity Review (Paper/Conference Submissions Document, Current and Inert Projects). Discussion about differentiation trees in the context of mosaic and regulative morphogenesis. Attendees: Susan Crawford-Young, Harikrishna Pillai, Karan Lohaan, Richard Gordon, Anlan Sun, and Bradly Alicea
C
D
E
D
Okay,
thanks
yeah,
so
the
google
summer
code
activities
and
here's
one
of
the
people
from
google
summer
code
right
now,
hari
krishna,
so
and
but
we
have
an
archive
of
meetings
and
in
some
of
those
archives,
I'll
I'll
respond
to
you
in
in
slack
we'll
have
you
know
we
have
a
youtube
channel.
We
have
some
data
sets
and
I
have
an
onboarding
guide.
I
can
send
you
and
that
should
start
you
off
on
on
some
things.
D
F
Yes,
so
on
the
last
meeting
I
had
mentioned
about
the
algorithm,
which
I
tried,
so
I
wanted
to
show
the
result
from
which,
like
which
I
got
using
that
okay
I'll
actually,
I
tried
to
use
the
embryo
data
set,
but
during
the
algorithm
show
the
error
saying
that
the
images
are
not
much
continuous.
So,
for
the
sake
of
getting
a
result,
I
tested
out
a
different
360
degree
view
data
from
the
internet
and
it
gave
an
output.
So
I'm
now
trying
to
make
it
work
for
this
dataset.
So
I'll
show
you
the
result.
D
D
F
D
F
D
A
A
F
H
H
H
I'll
share
it
in
the
meeting,
so
the
difference
between
those
images
I'll
either
have
to
incorporate
another
step.
You
know
to
unblur
those
images,
especially
when
you
get
to
state
six
state,
seven
and
stage
eight
tax
dot,
embryos
right,
the
number
of
cells
and
the
like.
It
gets
very
hard
for
the
algorithm
to
decide.
You
know
to
find
corresponding
points
between
two
image
sets
like
if
we
have
it
for
wait.
Let
me
just
share
the
link
first
I'll
say
the
google
link
drive
or
I'll
just
try
to
find
it.
H
H
H
H
H
It
won't
give
too
many.
You
know,
dancing
options
as
to
what
to
go.
How
way
to
go
about
you
know
when
we
have
to
finally
get
the
final
model
so
yeah
that
is
there.
Otherwise.
I
think
this
was
the
main
thing.
Otherwise
they
keep
on
checking.
You
know
alternatives
for
two
or
three
of
the
things
like
I
had
mentioned.
You
know
the
especially.
The
second
part
where
we
have
to
you
know
generate
the
3d
model
from
the
3d
point
cloud,
so
for
that
I've
not
found
any
better
alternative.
H
So
far,
so
I'll
be
sticking
with
my
own
method
that
that's
there
and
then
also
about
the
marsh
short
thing,
I'm
still
searching
for
you
know
a
neural
net
that
can,
you
know,
create
a
3d
probabilistic
point
cloud
for
doing
some
of
these
things,
so
I
think
instead
of
you
know,
searching
for
things
like
this
sort.
I
think
I'll
have
to
you
know,
take
a
very
base,
basic
boilerplate
code
for
doing
something
like
this
and
then
you
know
modify
it
specifically
for
the
accelerator
model.
H
So
that
is
also
something
that
I
was
looking
into.
Otherwise
for
the
weekly
goals.
These
are
the
things
that
I
had.
You
know
the
main
part
of
it
was
just
you
know,
finalizing
which
embryo
stages
I
have
to
choose
and
that
I
will
be
going
forward
with
properly
so
that
range
so
far
is
three
to
six
I'll.
Try
to
increase
more.
You
know,
if
I
can
find
bet
like
a
better
way
to
you,
know,
use
more
images.
C
D
Well,
that
sounds
like
we're
coming
along
good
and
both
hari,
krishna
and
karan
are
working
on
this
and
they
have
their
own
looks
like
they're
going
in
complementary,
but
different
directions.
So
that's
good
and
then
from
jiahang
and
wataru
I've
heard
I've
been
talking
with
them
a
lot
on
slack
and
they're
working
on
they're
they're,
making
some
progress
on
their
work.
We
had
some
discussions
about.
D
You
know,
I
think
we
I
think
jia
hong
is
interested
in
topological
data
analysis,
something
that
we've
talked
about
the
meetings
but
not
really
kind
of
dug
into
very
much
so
he
might
be
going
in
that
direction,
but
yeah
they're
working
on
their
projects
as
well-
and
I
don't
you
know
it's
a
little
late
for
them
when
they
when
they
do
make
the
meetings.
I
think
it's
like
11
at
night
for
them
so
but
I'm
gonna
share
my
screen.
D
I
wanted
to
go
over
our
github
repository,
so
this
is
the
github
repository
that
we
have
for
gsoc
and
it
looks
like
actually
looks
like
koran
has
made
his
first
contribution
to
this.
So
he's
put
in
a
read
me
an
initial
draft,
a
python
file
here
and
then
some
write-ups,
so
this
is
kind
of
what
I
was
thinking
about
for
submitting
things
to
the
github
repo.
D
D
If
you
want-
and
then
I
know
that
well
so
this
is
quran's
update
here,
he's
putting
in
details
he's
got
a
python
file,
so
this
has
some
code
in
it
and
then
he
has
this
readme,
which
you
know
you
can
put
in.
We
can
do
this
later.
Actually,
if
we
want
to
build
a
readme,
because
when
we
have
the
finished
product,
maybe
a
readme
would
be
good
to
have
well,
we
have
to
have
a
readme
for
the
submission
to
google,
but
you
know
that's
that's
part
of
it.
D
Okay,
yeah
and
then
the
write-ups,
which
you
know
if
you
want
to
create
write-ups.
I
I
had
asked
people.
I
think
if
there
was
a
requirement
this
year
for
doing
like
a
weekly
blog
post
update-
and
I
don't
think
there
is
this
here-
but
in
past
years,
they've
done
like
weekly
updates
on
their
on
people's
progress.
Now
you
don't
have
to
do
that,
but
it's
nice
to
keep
a
little
diary
of
what
you've
done
so
in
in
koran's
case.
D
D
So
this
is,
you
know,
writing
this
up
in
the
moment
and
then
looking
back
and
so
yeah
week
one
here
he
has
a
some
just
some
points.
You
know
some
bullet
points
that
you
can
follow
up
on
and
so
that's
and
then
I
did
talk
to
jia,
hong
and
wataru
about
working
on
theirs
and
they're,
going
to
try
to
do
they're
going
to
work
largely
from
their
own
repository.
D
D
Let
them
update
us
at
another
time,
but
I
just
wanted
to
go
through
that
gsoc
portion,
so
that
people
were
familiar
with
what
what
you
know,
how
that
would
work
and
how
we're
gonna
go
forward
on
that
and
you
can
work
from
your
own
repo.
It's
it's
fine.
Just
I
just
wanted
to
have
people
give
an
update,
maybe
once
a
week
or
once
every
two
weeks,
so
that
we
can
keep
things.
D
H
H
D
H
D
A
A
D
D
This
is
our
list
of
things
that
we
have
outstanding,
that
we're
submitting
to
different
places,
and
you
know
perhaps
things
that
we
want
to
put
on
the
list
to
target
different
venues
or
things
that
are
sort
of
fallen
by
the
wayside,
and
we
can
pick
them
up
again
so
to
go
over
what
we
have
here.
We
have
the
so
in
green.
These
are
things
that
have
already
been
presented
or
accepted.
D
So
I
have
a
I'm
going
to
be
actually
at
netsci
2022,
which
is
the
network
science
conference,
and
you
know
I'm
going
to
be
participating
in
both
the
main
conference
and
the
the
satellite
sessions
main
conference.
I'm
going
to
give
a
talk
on
divergent
integration
of
embryo
networks,
so
this
is
a
thing
that
we've
talked
about
in
the
meeting
with
respect
to
embryo
networks.
D
The
divergent
integration
aspect
of
it
is
just
where
you
have
the
formation
of
different
networks
based
on
different
parts
of
the
embryo,
differentiating
so
when
the
connectome
differentiates,
for
example,
this
forms
a
sub
network
with
respect
to
the
rest
of
the
embryo
network,
and
you
start
to
get
this
divergence
of
subnetworks
over
development,
but
they're
also
integrated
because
they're
part
of
the
same
organism,
and
so,
let's
kind
of
describe
the
network
theory
terms.
What
that
looks,
that
process
looks
like
so.
This
has
actually
been
well.
D
This
is
the
accepted
one
here
I
have
this.
I
guess
twice.
I
don't
know
why,
but
in
any
case
we
have
I'm
going
to
present
that
next
month,
that's
coming
up,
so
that's
that's
a
actually
green.
I
should
just
put
green
on
that
I'll
fix
this
later,
but
I
just
wanted
to
go
over
some
of
the
things.
The
next
one
is
this
test
of
williamson
symbiosis,
and
this
is
something
that
dick
gordon
has
been
championing,
and
this
is
a
paper
on
looking
at
different
genomes.
D
You
know
it's
a
bioinformatics
heavy
thing
that
if
you
were
to
do
this,
you
know
looking
at
different
genomes
and
looking
for
evidence
of
overlap
in
developmental
programs,
and
you
know,
we'd
have
to
go
through
that.
What
that
is
a
bit
more,
how
that
proceed
process
would
work,
but
it's
basically
looking
for,
like
the
genetic
precursors
of
symbioses
and
what
they
call
compound
development,
which
is
where
an
organism
goes
through
different
stages
of
development.
Sometimes.
E
D
Change
their
they
change
their
phenotype
quite
drastically
in
development,
and
you
know
by
caterpillars
they
start
out
as
they're
crawling
along
on
on
tree
branches.
They
go
into
this
cocoon
and
then
they
emerge
as
a
butterfly,
and
that
process
involves
a
lot
of
d
differentiation
of
cells
and
redifferentiation
of
cells,
and
so
there's
thinking
that
there
might
be
some
precursors
in
the
developmental
program
that
allow
for
this-
and
this
is
the
test-
would
be
to
look
at
genomes
and
compare
their
structure
and
their
content.
D
D
We
also
have
this
diatom
movement
project,
and
this
is
looking
at
these.
We
do
work
with
diatoms,
so
we
do
work
with
these
they're
single
cell
organisms.
Sometimes
they
exist
in
single
cell
form.
Sometimes
they
exist
in
colonies
and
we're
looking
at
their
we're
interested
in
their
movement.
So
we've
done
a
lot
of
image
processing
on
images
of
diatoms.
D
D
So
sometimes,
if
you
have
like
a
stochastic
motion
where
it's
just
generated
by
a
random
process
that
jerky
movement,
you
can
pick
up
jerks
in
the
in
the
movement
trajectory,
meaning
that
it's
not
smooth.
It's
it
kind
of
oscillates
along
a
line,
and
so
this
is
the
kind
of
thing
we're
interested
in
and-
and
you
can
derive
this
from
image
data.
You
can
drive,
derive
this
from
microscopy
movies.
D
That
people
have
made-
and
you
know
some
of
the
movies
available-
are
low
at
a
low
sampling
rate,
but
we
have
movies
that
are
of
a
higher
sampling
rate
that
we
can
use
for
this,
and
we
can
compare
those
movies
and
you
can
do
a
lot
of
mathematical
transformations
of
the
data,
and
so
this
involves
a
number
of
things.
This
involves
image
processing.
D
This
involves
mathematical
modeling
and
it
involves
comparative
work
because
you
might
want
to
look
between
different
types
of
diatom
to
get
a
sense
of
what
what
we're
dealing
with
in
terms
of
the
movement
to
interpret
the
movement.
So
there's
a
lot
of
work
there.
We've
done
work
on
diatoms
previously.
D
So
if
you
go
to
our
github
repository
for
diva
worm,
we
have
a
project
called
digital
basil
area
and
this
project
started
a
couple
years
ago.
We've
had
a
number
of
contributors
to
this,
and
the
idea
here
is
that
we're
trying
to
model
these.
So
these
are
diatoms
they're,
marine
microorganisms
or
algae,
and
they
exist
in
these
colonies
or
they
can
exist
as
free-form
single
cells.
Sometimes
they
have.
You
know
they're
shaped
in
different
ways,
so
they
have
sort
of
rounded
phenotypes.
D
These
are
elongated
phenotypes,
so
these
are
different
but
they're.
You
know
they're
arrayed
in
this
case
they're
arrayed
in
a
colony
and
they
move
against
one
another
and
that's
sort
of
why
this
organism
is
interesting
because
they
have
this
unique
mode
of
movement.
It's
like
an
accordion,
and
so
this
is
something
we
want
to
interpret.
D
These
are
the
species,
so
they
exist
in
these
accordion-like
colonies.
They
exist
in
these
chains.
They
exist
in
different
forms,
and
so
we,
you
know
it's
very
they're.
It's
not
very
well
known.
The
diversity
here
is
not
very
well
known.
There's
some
diversity,
that's
well
known
in
terms
of
like
people
studying
certain
species,
but
the
movement
is
definitely
not
really
well
understood.
There
are
a
lot
of
competing
theories
as
to
why
that
they
exist,
but
why
why
they
move
the
way
they
do
so.
D
There's
also
some
work,
that's
being
done
on
archaea
bacteria
and
looking
at
how
they
their
shape
can
be
compared
to
the
shape
of
oil
droplets.
So
oil
droplets,
of
course,
are
inorganic
and
archaea
bacteria
are
bacterial,
so
they're
living
cells.
So
what
are
the
differences
between
those
two,
and
that
again
is
something
that
involves
image
processing,
we're
also
interested
in
doing
work
with
cellular
automata
and
wolfram
what
they
call
wolfram
patterns.
D
D
D
In
other
words,
can
you
produce
things
that
look
like
biological
patterns
through
a
closed
set
of
rules,
meaning
that
there's
like
just
a
finite
set
of
rules
that
always
apply
and
always
produce
that
same
pattern?
Or
do
you
need
new
rules
occasionally
to
maintain
those
patterns?
And
so
you
know
there
are
a
lot
of.
We
would
need
to
talk
about
this
if
someone
wanted
to
do
this,
this
involves,
I
think,
working
with
a
simulation
package
working
with
like
cellular
automata,
but
also
maybe
some
image
processing
of
patterns.
D
If
we
have
images
of
patterns
that
we
can
decompose
and
find
some
of
the
maybe
the
rules
of
some
of
these
patterns
as
they
form
or
as
they
exist,
you
know
there
are
a
lot
of
creative
solutions
to
this,
and
I
don't
really
know
what
they
are.
We
we
talked
about
doing
like
a
windy,
ising
model,
so
there's,
but
you
know
that's
that's
something
that
we've
really
moved
very
far
on,
so
if
people
are
interested,
this
is
something
that
we
have.
D
This
is
the
dgnns.
This
is
the
developmental
graph
neural
networks.
This
is
the
project
that
jiahung
and
we're
working
on,
and
we
want
to
create
graph
neural
networks
of
developmental
processes.
So
we
have
these
embryo
networks
which
are
cells
in
an
embryo
that
form
networks.
They
could
be
proximity
networks,
it
could
be
signaling
networks
and
then
taking
those
networks
and
finding
embeddings.
D
You
know
taking
the
data
from
actual
embryos
and
finding
graph
embeddings
that
describe
those
embryos,
so
the
positions
of
cells,
the
positions
of
different
things,
structural
things
in
the
embryo
and
then
building
graph
graph
embeddings
to
describe
that,
and
then
our
target
here
not
only
is
the
end
of
g
sock
and
successful
end
of
gsoc,
but
this
learning
on
graphs
conference-
and
so
this
is
september,
9th
and
16th.
So
this
is
this
deadline.
I
think
this
is
september.
9Th
is
for
the
abstract.
16Th
is
for
the
paper.
D
D
This
is
on
hard
to
define
events
in
biological
networks,
and
so
what
are
hard
to
define
events?
I
did
a
workshop
about
10
years
ago
on
something
called
hard
to
define
events,
and
these
are
things
that
are
hard
to
define.
You
know
they
might
be
like
rare
events
that
almost
never
happen.
They
might
be.
You
know
hard
to
just
describe
measurements,
so
you
know
you
might
be
dealing
with
something
that
you
can't
really
measure
directly
and
it's
very
hard
to
get
a
grasp
on
what
the
measurement
looks.
D
Like
other
types
of
things
like
non,
you
know
non-normal
distributions,
so
things
of
the
very
long
tail,
and
so
these
are
all
these
hard
to
define
events
and
then
what
I'm
gonna
do
in
this
talk.
Is
I'm
gonna
review
this
review?
What
we
did
at
that
workshop-
and
I
also
did
some
follow-up-
I
used
to
run
a
sort
of
sort
of
an
annotated
list
online
for
about
a
year
after
the
workshop,
and
I
would
add
in
things
that
looked
interesting.
D
So
there
are
all
sorts
of
mathematical
approaches
to
these
kind
of
problems,
all
sorts
of
different,
their
approaches
using
network
theory
and
but
no
one's
ever
really
integrated
them
into
the
same
top
under
the
same
topic.
And
so
that's
what
we
were
doing
back
then
now
I
want
to
review
that
those
materials
and
bring
them
forward
into
what's
going
on
in
network
science
today.
D
Finally,
we
have
this
differentiation
tree
of
the
brain,
which
is
where
dick
gordon
has
derived
a
lot
of
work
of
his
work
on
differentiation,
trees
and
differentiation
waves,
and
it
involves
looking
at
our
lineage
trees
are
looking
at
like
the
line
of
descent
in
an
embryo
in
a
little
bit
different
way
than
we're
used
to
so
we're
used
to
what
we
call
lineage
trees,
which
are
trees.
That
branch
you
know,
cells,
divide
and
you
get
cell
lineages.
D
So
one
cell
has
you
know:
one
mother
cell
has
two
daughter
cells
or
maybe
a
collection
of
daughter
cells
depending
on
the
type
of
embryo,
and
then
you
can
trace
out.
You
know
different
identities
for
these
cells
with
a
differentiation
tree,
it's
a
little
bit
different
because
you're
talking
about
you,
know
the
differentiation
of
tissues
and
the
tissue
precursors.
D
D
And
so
you
end
up
with
this
tree
of
differentiated
tissues,
and
so
this
is
something
that
we
were
talking
about
doing.
This
may
need
to
be
refined,
but
I
want
to
bring
back
that
we've
done
some
papers
on
this
work
in
the
past.
I
want
to
turn
our
attention
back
towards
this
a
little
bit
more,
and
so
I
want
to
go
over,
maybe
some
examples
of
this.
D
This
is
from
one
of
our
papers,
actually
a
figure-
and
I
was
asked-
I
think,
gia
hung
asked
me
about
this,
so
I
want
to
bring
it
up,
so
this
is
sort
of
in
c
elegans.
This
is
how
we
envision
a
different
like
a
differentiate,
what
I
call
a
differentiation
map
and
it's
a
differentiation
tree
in
c
elegans,
but
it's
embedded
in
the
embryo
and
it's
you
know
at
least
two-dimensional
topology
on
the
surface
of
the
embryo.
So
this
is
so.
D
D
Are
differences
so
this
shows
like
this
differentiation
map,
where
you
have
going
from
p0,
which
is
the
single
cell
state
to
a
b
and
p
l
or
a
b
and
p
one.
So
these
are
two
d.
This
is
a
two
cell
state
and
then
you
go
down
these
different
lineages,
so
you
have
a
b,
a
b,
a
a
b,
a
l
and
then
down
to
a
b
a
r.
D
Now
what
are
these
green
and
red
arrows?
The
green
and
red
arrows
represent
different
asymmetries
in
the
cell
size.
So
when
you
have
a
division
generally,
you
know
you
might
assume
that
it's
a
symmetric
division
where
it's
roughly
about
the
same
size,
50
50.,
so
you
take
your
your
mother
cell.
It
has
a
certain
volume
and
the
daughter
cells,
if
they're
symmetrical
will
be
roughly
half
of
that.
So
it's
like
half
of
that
size,
and
so
then
you
know
this
continues
down
the
way.
D
But
in
a
lot
of
cases
you
have
asymmetric
divisions,
which
are
where
the
size
is
different.
So
it
could
be
that
your
daughter's
one
is
sixty
percent
of
of
the
original
mass
of
the
mother.
One
is
forty
percent
of
the
massive
mother:
that's
considered
an
asymmetric
division,
so
this
differentiation
map
actually
shows
that
they're
different.
D
But
if
you're
using
good
markers
for
the
for
the
membrane
around
the
edge
of
the
cell,
you
can
get
a
pretty
accurate
measurement
of
size,
and
so
these
size
differences
are
important.
So
that's
why
it's
important
to
have
not
only
sort
of
a
centroid
position
when
you're
doing
cell
tracking
but
a
sort
of
a
volumetric
measurement
as
well,
and
there
are
different
ways
you
can
approach,
volumetric
measurements,
so
the
difference
between
lineage
trees
and
differentiation.
Trees
are
as
follows.
D
So
for
lineage
trees
we
have
a
mother
cell
and
we
have
two
daughter
cells.
Sometimes
we
have
multiple
daughter
cells.
If
you
have
cell
proliferation,
usually
not
in
c
elegans
but
in
other
species,
then
one
of
those
daughter
cells
in
terms
becomes
the
mother
cell
and
gives
birth
to
new
daughter
cells.
And
again
this
can
be
a
proliferative
process
where
you
have
more
than
two
daughter
cells,
but
for
the
case
here
we're
just
kind
of
interested
in
the
c
elegans
lineage
tree,
where
we
have
two
daughter
cells
per
mother
cell.
D
D
D
It's
going
to
have
the
same
tree
structure
and
each
of
these
nodes
is
a
cell,
but
we're
interested
instead
of
in
the
heredity,
in
other
words,
the
mother
cell,
giving
rise
to
two
daughter
cells,
and
then
this
axis,
being
the
anterior
posterior
axis
and
in
different
species.
It's
defined
in
different
ways,
sometimes
there's
radial
symmetry,
in
which
case
these
endpoints
of
the
tree
are
correspond
to
different
parts
of
the
symmetry,
so
in
sea,
squirts
there's
a
four-fold
symmetry.
D
D
This
is
the
differentiation
tree,
and
so
here
you
have
a
differentiation
tree
with
cells
that
divide,
and
they
have
mothers
and
daughters
just
like
this,
but
we're
interested
in
the
difference
between
these
two
cells
we're
interested
in
the
difference
in
their
volume,
and
so
now,
if
you
know,
we
want
to
have
the
smaller
cells
to
the
left,
the
larger
cells
to
the
right.
So
we
order
these
by
the
volume
of
the
daughter
cells.
So
these
the
the
tree
is
organized
a
little
bit
differently.
D
Now
in
a
regulative
embryo
which
is
not
c
elegans,
so
c,
elegans
is
actually
a
mosaic
embryo
and
you
can
use
this
method
for
most
mosaic
embryos
for
a
differentiation
tree
in
a
regulative
embryo.
You
still
have
this
tree-like
structure,
except
instead
of
cells.
Now
you
have
tissues
and
so
now
you're
differentiating
different
tissues.
D
The
reason
why
the
different
the
lineage
and
differentiation
trees
in
c
elegans
is
basically
the
same
is
because
we're
really
you
know
in
a
c
elegans
tree,
all
the
cells
are
deterministic
with
respect
to
their
fate.
So
you
know
every
one
of
these
sublineages
has
a
certain
set
of
cell
types
that
will
emerge
from
it.
Every
cell
has
sort
of
an
identity.
So
if
you
go
through
this
lineage
you'll
end
up
at
the
end
of
development
with
cells
that
have
specific
functions.
D
We
have
classified
all
of
these,
the
they
have
a
nomenclature
for
this
and
they're
precursors
to
define
cell
types
and
it's
very
hard
to
change
those
with
cell
signaling
or
with
environment
they're.
Basically,
there
in
a
regulative
embryo
cells
will
branch
like
this
and
their
fate
will
be
dependent
on
where
they
are
in
the
embryo
and
the
signaling
mechanisms
that
they
get
exposed
to
so
in
c
elegans
in
this,
in
this
mosaic
form
of
development.
D
D
Here
you
might
have
like
a
muscle,
fiber
or
a
muscle
set
of
muscle
cells
that
might
start
to
differentiate,
and
then
here
you
might
have
different
types
of
muscle,
different
specific
muscles
in
in
the
emerging
embryo.
So
you
can
see
that
there's
this
differentiation
tree,
that
things
are
differentiating.
These
might
be
neurons.
D
Okay,
we
don't
necessarily
do
this
in
c
elegans,
because
we
don't
have
to.
We
know
what
the
fates
are
going
to
be,
so
we
can
actually-
and
there
are
other
methods
that
we
can
use
for
this-
we're
working
on
a
categorical
theory
methodology
for
this-
a
sort
of
a
hybrid
hyper
graph
and
a
category
theory
method
for
this.
But
it's
different
than
this
sort
of
differentiation
tree
where,
when
cells
differentiate
into
a
new
tissue,
they
get
a
branch.
So
you
can
follow
back.
D
These
signals
from
the
very
specific
types
of
cell
tissues
that
get
generated
to
a
very
general
set
of
cell
types
or
a
very
general
population
of
cells
that
emerge-
and
this
is
important
when
you
have
different
regions
of
the
embryo,
forming
tissues
and
maybe
cells,
changing
their
fate
as
they're
exposed
to
different
signals.
And
then
there
are
things
called
differentiation
waves
which
are
waves
of
differentiation
that
occur
in
space,
and
so
that's
something
we
can
talk
about
at
another
time.
D
D
So
if
you
have
like
cell
tracking
data-
and
you
have
like
the
mother
cells
position-
and
you
have
the
daughter
cell
positions,
you
can
interpolate
and
you
can
have
this
in
three
dimensions.
So
you
can
interpret
an
angle
at
which
differentiation
occurs
and
so
or
division
occurs.
So
you
have
the
a
b
cell.
It
divides
into
a
aba
and
abp
aba,
going
towards
the
anterior
end
or
the
head
abp
going
towards
the
posterior
end
in
the
towards
the
tail
and
then
there's
this
angle
of
division.
That
is
maybe
important.
D
We
don't
really
know,
I
don't
think
anyone's
written
about
this
as
a
paper,
but
this
is
something
you
observe
in
these
maps
that
show
great
variation
in
the
angle.
So
this
has
a
pretty
wide
angle
here:
between
aba
l
and
abar,
aba,
aba
and
abp
have
a
small
a
little
bit
smaller
angle,
and
then
you
know
a
b
to
aba
abp.
That
has
a
different
angle,
and
so
you
can
like
calculate
these
angles
out.
Sometimes
they're
almost
180
degrees,
sometimes
they're
a
lot
narrower.
I
I
I
Okay,
now
that
might
be
you'd
have
to
sort
of
do
local
strains
of
markers
on
of
some
kind
of
markers
to
see
if
you
can
figure
out
what
the
direction
of
minimal
strain
was
or
or
easiest
way
to
get
easiest
way
to
double,
because
after
all,
when
a
cell
doubles,
it's
also
going
each
cell,
each
daughter
cell
may
or
may
not
go
back
to
the
original
volume,
especially
if
it
does
increase
in
volume.
Then
it's
got
a
squeeze
somewhere
yeah.
I
D
I
D
D
I
I
D
Yeah
yeah,
so
I
mean
there
are
ways
we
could
ex
like
estimate
the
volume
and
that's
actually
there's
a
set
of
interesting
questions.
I
know
people
have
like
kind
of
taken,
like
you
know,
done
segmentations
of
cells
around
the
membrane
and
and
tried,
but
I
don't
think
people
have
actually
looked
at
the
question
of
like
volume
distributions
and
things
like
that.
I
mean.
D
D
I
D
D
So
this
is
something
that
I
know
we
come
back
to
from
time
to
time,
and
this
is
not
maybe
some
of
you
aren't
familiar
with
it,
but
this
is.
I
think
this
might
be
an
accessible
way
to
talk
about
it.
So
waddington
landscapes-
and
we
don't
start
with
this
one,
so
waddington
landscapes.
Are
these
structures
that,
if
you're
familiar
with
like
energetic
landscapes
or
radiant
descent
landscapes,
they're
somewhat
similar?
D
It's
the
similar
concept,
I'm
trying
to
find
a
picture
in
here
of
having
a
landscape
that
describes
this
process,
and
so
in
in
development.
You
have
a
process
of
differentiation
of
cells,
differentiation
of
embryos,
and
you
have
this
landscape
that
describes
sort
of
the
path
of
least
resistance,
and
so
so,
let's
see
I
don't
know
if
they
actually
have
one
of
these,
because
they
say
it's
revisiting
them,
but
they
don't
say
anyways.
D
I
wanted
to
show
a
picture
of
this
landscape,
but
just
imagine
a
landscape
where
you
have
these,
they
serve
the
surface.
That's
you
know
not
flat,
but
has
some
undulations
to
it,
and
some
of
the
undulations
include
valleys
or
canals,
and
so
what
they
mean
by
cannabization
is
that
development
tends
to
follow
those
canals
from
a
point
of
like
a
single
cell
embryo
to
the
point
of
adulthood,
and
so
this
is
a
very
conceptual
map
where
you
know
there
are
these
different
branch
points
down
the
landscape?
D
D
So
you
know
it's
like
thinking
about
how
many
paths
development
can
take,
and
so
sometimes
it
can
take
alternate
paths
and
sometimes
it
sticks
to
the
same
path,
and
there
are
different
reasons
for
that.
But
it's
a
it's
a
conceptual
model
of
how
development
kind
of
unfolds,
and
so
in
this
paper,
they're
revisiting
what
they're
re-examining
waddington
so
they're.
D
So
this
is
where
you
have
plastic
traits
that
become
part
of
the
part
of
the
developmental
process.
So
you
would
have
these
things
originally,
that
were
plastic
from
an
from
a
source
in
the
genome,
and
then
you
have
the
traits
become
part
of
the
developmental
process
and
they're,
not
plastic
in
the
descendant.
So
this
is
what
genetic
assimilation
is.
It's
assimilating
this
plasticity
into
the
into
the
developmental
process.
D
Genetic
assimilation
evolves
rapidly
assumed
to
be
in
large
part
due
to
segregating
genetic
variation
only
expressed
in
rare
novel
environments.
So
this
is
where
you
have
would,
but
otherwise
phenotypically
cryptic.
So
when
organisms
encounter
these
rare
novel
environments,
this
is
often
the
product
of
this.
Where
you
get
this
genetic
association.
G
J
D
D
The
landscape
here
this
is
like
waddington
when
he
was
writing
his
books
in
the
early
to
middle
20th
century.
He
did
he
kind
of
drew
these
by
hand,
and
it
was
very
conceptual,
and
this
ball
here
represents
an
embryo.
So
it's
you
know
it's
a
in
the
in
the
form
of
a
ball,
and
then
you
have
these
canals
that
come
down.
So
there's
this
branch
point
here,
so
this
ball's
rolling
down
this
canal
and
then
it
either
goes
down
this
path
or
this
path.
D
D
So
this
is
a
very
complex
process
in
this
nice
drawing
and
so
then
you
know
you
get
to
the
end
and
you
get
these
four
alternate
paths
for
development.
Where
does
the
ball
end
up?
One
way
to
model
this?
If
you're
familiar
with,
I
don't
know
some
of
you,
don't
know
the
game,
the
price
is
right
or
the
tv
show
the
price
is
right,
but
there's
this
game
called
plinko-
and
I
know
disc-
has
talked
about
this,
some
in
in
his
books
on
plinko
and
how
you
drop
a
like.
D
This
has
its
origins
in
what
they
call
the
galton
board
in
statistics,
which
is
a
similar
thing.
Where
you
have
this,
it's
a
problem.
It's
a
probability
tool
where
you
have.
You
know
you
drop
something
in
the
top
and
it
bounces
around,
and
you
know
it's
probabilistic
it'll
end
up
in
any
one
bin.
So
this
is
that's
a
totally
probabilistic
example,
but
in
development
you
know,
sometimes
things
are
probabilistic
and
sometimes
things
are
deterministic.
D
I
Bradley
I
I
always
have
had
a
problem
with
waddington's
picture
yeah.
If
the
ball
represents
the
embryo
and
the
embryo
becomes
called
one
kind
of
cell,
if
every
brain,
if
it
I
don't
know
what
else
could
be,
because
there's
no
cell
division
here
and
the
embryo
really
should
go
back
down
all
the
paths
right.
D
D
D
D
Right
right
and
the
idea
well
the
idea
in
like
a
gradient
descent,
landscaper
energy,
minimization
and
in
a
fit
in
a
fitness,
landscape
and
evolution,
it's
maximizing
and
minimizing
fitness,
and
so
in
this
case
they
don't
really
have
those
criteria.
Like
you
don't
have
these
usual,
you
could
call
them
optimization
criterion.
D
K
D
There
we
go
all
right
now.
I
share
mine
all
right
there
we
go.
So
that's
that's
what
this
paper,
that's
the
context
for
this
paper
and
they
they
re-evaluate
this
and
so
they're,
actually
using
data.
So
they
talk
about
how,
despite
previous
work,
suggesting
a
substantial
cryptic
genetic
variation
contributing
to
the
evolution
of
this
assimilation.
D
Some
have
argued
for
a
prominent
role
for
new
mutations
of
large
effect,
concurrently
selection.
So
this
is
something
that
it's
a
different
mechanism.
So
if
we
go
back
to
waddington,
he
was
less
concerned
by
the
relative
contribution
of
cgv
or
new
variants,
but
aimed
to
test
the
role
of
canonization,
which
is
this
genetic
assimilation.
It's
very
similar
idea.
E
D
Canalization
is
where
your
take,
you
know,
you're,
taking
these
different
paths
based
on
what
you're
encountering
in
terms
of
plasticity-
and
this
is
also
an
evolved
form
of
robustness
while
canonization
has
been
extensively
studied.
Its
role
in
the
evolution
of
genetic
assimilation
is
disputed
in
part
because
explicit
tests
have
evolved,
robustness
are
lacking,
so
they
recreated
waddington
selection
experiments
on
an
environmentally
sensitive
change
in
drosophila
wing
morphology.
So
this
is
the
fruit
fly.
D
There
is
no
evidence
for
evolved
changes
in
canalization,
contributing
to
genetic
assimilation
and,
finally,
there's
a
potentially
what
they
call
pleiotropic
or
when
you
have
a
single
genetic
background.
You
have
many
different
variants,
many
different
phenotypes
that
emerge
and
fitness
consequences
in
natural
populations.
D
It
may
not
be
entirely
cryptic,
so
cryptic
variation
is
where
you
don't
necessarily
know
the
source
or
it's
hard
to
characterize
the
variation.
So
if
we
go
down
to
the
figures
on
this,
we
have
this
example
here
of
the
this
is
a
drosophila
wing,
and
these
are
the
cross
beans.
You
see
here,
and
so
you
know,
the
people
have
analyzed
these
quite
a
bit.
They've
done
different
things
with
image,
processing
and
they've.
Looked
at
like
different
aspects
of
the
cross
of
the
cross
veins
in
the
wing.
D
D
This
shows
like
this
graph
shows
where
you
have
up
selection,
which
is
in
blue
genetic
assimilation,
which
is
here
and
down
selection
in
green,
and
then
it
says
the
wheels
associated
with
this
show
strong
response
to
selection.
So
up
selection
works
for
this
trait.
This
is
cross
vaden's
frequency,
so
there
are
more
cross
veins.
D
You
know
per
square
millimeter
than
in
the
ones
that
were
down
selected,
so
selection
actually
plays
a
heavy
role
in
defining
the
cross
vein
frequency
here
and
then
this
genetic
assimilation
shows
that
this
process
here
is
delayed
for
about
17
generations
and
then
matches
what
you
see
with
selection.
So
this
is,
I
guess,
they're
doing
some
experimental
evolution
where
they're
raising
different
generations
of
drosophila
under
different
conditions
and
looking
at
them
and
they're
plotting
out
the
results.
D
D
D
Yeah,
so
when
you
raise
them
at
a
higher
temperature,
they
end
up
with
you
know
these
increased
number
of
defects,
and
so
in
this
case
you
didn't
see
any
differences
between
macro
or
micro,
environmental,
canalization
or
wing
defect,
and
so
this
is
interesting.
It
has
a
lot.
You
know
it.
It
kind
of
shows
that
there's
that
they're,
overcoming
this
effective
temperature,
somehow
and
but
that
they're,
you
know
they're
kind
of
mimic
what
you
see
with
canonization
environmental
canalization.
D
Okay,
so
that's
that's
this
paper,
I'm
gonna
go
to
the
next
one,
so
I
think
that
sets
up
this
paper
pretty
well,
and
this
is
actually
a
lot
more
accessible
and
probably
brings
us
home
a
little
bit
more.
So
this
is
it's
a
feature
that
was
in
cell.
I
don't
know
if
it
was
in
the
journal
cell,
but
it's
by
cell
press
and
the
question
is:
is
how
can
waddington
like
landscapes,
facilitate
insights
beyond
developmental
biology,
and
so
this
is
like
giving?
D
They
asked
a
bunch
of
people,
a
bunch
of
scientists,
this
question
and
they
wanted
to
get
a
sense
of
how
can
waddington
like
landscapes,
go
beyond
developmental
biology.
So
this
is
where
you
know
we're
talking
about
embryos
here,
but
maybe
talking
about
other
things
as
well,
that
are
related
to
maybe
other
biological
questions
or
in
engineering.
So
this
person
here
nika
shakiba.
D
She
mentioned
engineering
beyond
landscapes,
so
the
waddington
landscape
is
often
an
intuitive
framework
to
conceptualize
cell
fate,
grounded
in
our
innate
understanding
of
the
physical
rules
that
guide
balls,
rolling
down
a
hill.
So
there's
a
connection
here
between
gravit
gravity
and
the
waddington
landscape.
D
The
hills
and
valleys
of
the
landscape
are
molded
by
the
cellular
processor,
the
interregulatory
network
of
dna,
rna
and
proteins
that
govern
cell
states.
So
in
this
case
you
know
you
can,
and
there
are
different
ways
you
can
characterize
this.
You
know
internal
network
of
things.
This
is
a
lot
of
stuff
going
on
here,
dna,
rna
and
proteins.
D
So
you
know
we
want
to
be
able
to
break
out
of
this
metaphor
and
actually
apply
it
to
some
problems,
and
so
now
you
know
we
can
reimagine
the
landscape
using
predictive
computational
models
and
use
it
in
melee
and
synthetic
biology
and
use
it
as
a
tool
for
designing
landscapes.
That
guide,
sulfate
long
prescribed
paths
so
in
in
in
synthetic
biology,
often
times
you're
putting
genes
into
a
cell
to
get
it
to
do
specific
things
like
create
some
enzyme
or
create
some
protein,
and
so
you
can
synthesize
chemicals
using
these
things.
D
D
As
a
an
aside
here,
she
mentions
mentioned
stochastic
fluctuations,
which
is
interesting,
because
oftentimes
molecular
networks
exhibit
a
lot
of
stochastic
fluctuation.
Sometimes
this
is
a
mechanism
for
robustness,
which
means
it
allows
it
to
function
in
a
lot
of
different
settings
where
you
might
experience
perturbations
that
might
knock
you
off
track
other
times.
It's
just
a
product
of
it,
not
being
a
deterministic
thing.
You
know
it's
just
something
that
happens
irregularly
and
it
happens
in
parallel.
So
you
end
up
with
this
signal
that
looks
like
it's
stochastic.
D
D
D
I
I
D
So
again.
Well,
they
pay
in
this
case.
In
the
landscape
view,
cell
types
are
characterized
by
basins
or
attractors.
So
this
is
where
those
canals
are
now
considered
to
be
basins
of
attraction,
and
so
I
don't
know
if
it
suffers
the
same
issue
that
the
stu
kaufman
work
did.
I
imagine
it's
probably
similar
to
that.
D
Directly
I
mean
the
distance
is
processed.
It
moves
from
one
place
to
another,
but
this
is
you
know
this
is
again
something
that
we
think
about
in
terms
of
energy
and
energy,
minimization
and
energy
function.
So
a
neuroscience
hope
field
pioneeringly
proposed
the
concept
of
energy
function
to
explore
computational
properties
of
neural
circuits,
such
as
associative
memory.
So
maybe
that
you
know
something
like
a
spiking
neural
network
model
might
be
integrated
with
landington
landscapes
and
that
can
drive.
You
know
drive
some
insight,
so
jordy,
garcia,
jojovo.
D
D
D
So
you
know,
I
said
that
the
waddington
landscapes
aren't
really
about
optimization,
but
they
can
be
about
optimization.
You
have
to
be
a
little
bit
more
specific
in
how
they're
you
know
how
they
map
to
the
developmental
system.
You
can't
just
use
it
as
a
metaphor,
but
the
idea
here
is
that
the
idea
that
a
single
dimension
can
account
for
the
state
of
a
cell
as
decisions
are
made
is
clearly
insufficient.
D
So
these
these
channels
aren't
deterministic
they're
kind
of
pushed
they're
kind
of
carved
out
as
the
as
the
cell
moves
along
these
state
pathways,
but
so
this
just
talks
about
how?
If
we
incorporate
more
of
these
things
that
we
know
about
the
biology,
you
know
it
may
be
a
tool
for
optimization
or
understanding
optimization
in
a
biological
sense.
This
is
cell
fate
reprogramming.
D
D
D
But
yet
the
quest
to
distill
the
discovery
and
dynamics
of
biological
systems
into
its
basic
organizational
principles
is
far
from
complete.
So
this
is
where
geometric
perspectives
and
metaphors
can
play
a
fundamental
role.
So
we
know,
for
example,
like
I've
mentioned
that
we
have
these
convex
spaces
that
can
describe
optimization.
D
And
so
this
is
links
to
complex
systems,
theory
which
we've
also
talked
about.
But
this
just
basically
you
know
it
helps
us
structure
our
debates
about
what
the
basic
principles
of
biology
are
and
so
and
then
you
know
we
can
also,
but
we
can
also
look
at
attractors
as
linking
directly
to
data
and
so
alexandra
walczak,
who
is
at
french
university
sort
of
argues
that
you
can
actually
look
at
data
in
this
way
as
well.
D
So
you
know
we
build
so
from
her
perspective
cell
differentiation
pathways
have
diverse
implementations
that
lead
to
a
set
of
finite
states,
and
so,
but
as
these
two
authors
have
shown
a
few
years
back,
it
can
also
simplify
molecular
description
and
lead
to
quantifiable
experimental
predictions.
These
models
also
build
on
the
intuition.
We
see
in
many
living
systems.
D
D
But
it's
something
that
we
need
to
have
an
intuitive
model
for
at
the
level
of
the
luddington
landscape,
but
indeed
waddington
talks
about
this,
but
it
doesn't
have
the
the
data
to
show
that
you
know
how
this
this
unfolds,
and
so
there's
a
linkage
here
between
some
of
the
a
lot
of
the
data
that's
been
collected
out
there
and
bringing
this
back
to
a
landscape
model
so
in
protein
folding.
The
idea
that
proteins
differ
from
other
polymers
by
having
a
funneled
folding
landscape,
which
helps
them
avoid.
D
Misfolded
structures
explains
how
they
can
fold
in
a
finite
time,
and
so
this
is
something
you
can
also
see
in
protein
folding,
where
you
have
these
challenges
of
this
process,
and
you
know
it.
It
ends
up
in
this
sort
of
state
space
from
a
large
number
of
possibilities,
and
so
you
know
using
examples
from
the
data
that
we
have
is
is
actually
quite
useful.
D
Then
there
are
other
examples:
I'll
go
I'll
kind
of
go
through
the
rest
of
this
just
kind
of
like
pointing
out
the
the
application.
So
this
is
human
microbiome
landscapes,
so
applying
microbiomes
to
this
kind
of
landscapes,
learning
from
natural
variation,
which
is
where
you're
learning
you
know,
observing
natural
variation
talks
about
canalization
being
something
like
the
wild
type,
which
is
like
a
a
normal
genotype
instead
of
a
mutant
genotype.
D
So
you
know,
if
you
have
your
your
mutants
actually
generate
more
channels
during
canalization,
so
you
can
learn
a
lot
about
natural
variation
by
these
models
by
adding
in
canals
and
branch
points,
and
so
this
is
something
that
people
do.
People
might
be
interested
in
doing
single
scale,
dynamic
single
cell
dynamical
landscape.
D
This
is
a
specific
type
of
data
that
looks
at
gene
expression
and
large
scale,
gene
expression-
and
you
can,
you
can
unpack
a
lot
of
the
others,
data
using
dynamical
systems
methods
and
then
again,
single
cell
dynamical
landscapes,
brain
attractor
landscapes,
which
then
goes
to
the
brain
and
some
other
things
going
on
there
beyond
the
evocative
picture.
So
this
is
actually
talking
about
this
mathematical
framework
and
visual
metaphor
connection.
D
So
you
know
they're,
now
we're
talking
about
doing
higher
order,
mathematics,
using
landscapes
and
actually
landscapes
can
capture
things.
So
this
picture
describes
a
dynamic
system
followed
by
homoresis
homeoresis,
which
represents
the
return
to
a
stable
trajectory,
a
set
line
after
perturbation,
and
this
is
unlike
homeostasis,
which
returns
to
a
set
point
or
a
particular
state.
So
this
is
something
that
you
know
now
we're
talking
about
regulation
and
some
of
the
theories
of
regulation
that
exist,
and
so
this
is
now
moving
towards
sort
of
theory
building.
D
D
Okay,
well
thanks
for
meeting
and
we
won't
be
meeting
next
week.
It's
a
holiday
in
the
u.s
we'll
be
meeting
the
following
week,
so
the
gsox
students
say
I
want
them
to
give
updates
in
the
slack.
I
think
we
can
do
that
and
then
the
following
week,
maybe
we'll
have
some
nice
demos
to
show
from
your
several
weeks
of
hard
work.
The
first
evaluations
for
gsox
students
is
at
the
end
of
july.
D
So
that's
not
that
far
off,
but
I
want
you
know.
I
don't
usually
make
it
hard
for
people.
I
I
try
to
make
it
easy
for
people.
So
it's
you
know.
I
just
want
to
make
sure
that
you're
making
progress
on
your
projects,
and
so,
if
you
have
any
questions,
let
me
know,
but
it's
usually
just
a
formality.
I
think
all
of
you
are
doing
well
so,
but
just
keep
that
in
mind,
because
we
I'll
you'll
actually
owe
them
an
evaluation
or
you
have
to
like.