►
From YouTube: DevoWorm (2023, Meeting #32): DevoSAM, DevTOGL, Molecular Order to Differentiation, Genetic Mosaics
Description
DevoLearn updates. DevoSAM, zero-shot learning from embryo microscopy, DevTOGL, graph embeddings. Overview of the Molecular Biology of Differentiation paper. Genetic Mosaics in Drosophila paper (from 1972) and how to make genetic mosaics. Gene Regulatory Networks of gut and intestinal differentiation in C. elegans. Attendees: Sushmanth Reddy Mereddy, Himanshu Chougule, Bradly Alicea, Jesse Parent, Morgan Hough, Lukas, and Richard Gordon.
B
A
B
C
B
C
D
Great,
did
you
send
them
to
in
the
chat
here.
B
D
C
D
C
I
came
across
this
video
actually
two
days
back
and
Justin
bought
me
some
new
idea,
I'm
gonna
start
each
other
right
now
we
don't
have
any
models
for
for
technically
plotting
the
cell
lineage
analysis,
but
in
this
video.
C
Now,
I
in
this
video,
they
are
directly
segmenting,
the
they
are
making
custom
custom
data
set.
Let
me
show
it
to
you
yeah.
If
you
can
see
this
image,
they
are
making
custom
data
set
and
try
fine
tuning
on
the
data
prone
to
the
model
from
meta
AI
to
get
cells.
What
are
alpha
granules
Etc
in
CL
against
case
we
can
do
this
thing.
C
These
are
mentioned.
Actually
it
was
fine-tuned
on
this.
It's
more
I
mean
this
tutorial
can
be
useful
and
we
can
use
that
in
C
elegans
for
cell
image
analysis
right
now.
We
have
a
model
on
Cell
lineage
analysis,
but
it
doesn't
tell
itself
belongs
to
what
I
mean
which
cell
named
ASM.
It
just
tells
the
account
like
these.
C
Many
asms
are
there:
the
T
cells
are
there
if
you
want
to
create
our
own
data,
set,
naming
all
these
conventions
actually
in
this
video
they
have
mentioned
how
to
make
their
own
data
set,
how
to
find
unit.
Also,
okay,
actually
couple
of
days
before
G-Shocks
I.
Don't
remember,
you
asked
your
garden
Austin
me
mention
me
about
right
now,
I
mean
there
is
no
segmentation
model,
which
represents
the
same
sentence
like
it
is
a
ASL
on
top.
C
D
C
Right
now,
right
now,
this
week,
I
have
completely
worked
on
the
manuscript
I
have
already
here.
Any
comments
would
be
useful
by
next
meeting.
I
think
so
I
will
add
some
more
matter
and
I
will
add.
The
pictures
also
right
now
is
struggling
with
the
pictures
like
plotting
all
this
stuff.
I'll.
Try
to
add
those
pictures
before,
like
I,
have
seen.
Some
of
people
are
working
on
the
same
fine-tuning
on
cell
images,
so
I
was
thinking
to
host
it
as
a
preprint.
C
D
Yeah
yeah,
that
would
be
fine,
oh
yeah.
So
let's
look
at
the
third
paper.
You
said
so
you
sent
the
diva
and
the
microsam
and
then
the
third
one
is
the
even
yeah.
So
this
is
yeah.
This
is
evil
net.
What's
the
difference
between
the
devonet
and
the
micro
SIM,
the
micro
segment?
Isn't
the
general
case
or.
C
Microsoft
is
like
a
fine
tuning,
sound
model
yeah,
which
was
given
by
Beta,
but
when
it
is
like,
like
whatever
the
proposal
I
have
wrote
right,
it's
completely
depend
on
that.
I
was
not
only
working
on
one
model
right
now.
Parallel
working
on
two
models,
I
was
trying
to
work
it
out
in
the
same
manner.
In
one
model,
it
can
segment
perfectly
at
zero
shot
performance,
zero,
shot
performance
means
whatever
image.
If
you
give
it,
it
will
obviously
segment
it
segmented
and.
C
C
Area
and
timestamp
their
position,
Etc.
D
C
But
we
are
lacking
in
the
cell
lineage
analysis
I
think
so
like
extracting,
which
cell
is
named
as
ACP
selection?
I,
really
don't
know
the
nomenclature
of
it,
how
we
name
those
cells,
but
we
can
crack
this
with
this
data,
I
mean
I
have
kept.
The
video
also
feel
free
to
check
it
out
when
you
are
free
driving,
I
think
this
could
be
next
year,
gsock
project
or.
C
C
That's
the
reason
I
was
thinking
this
for
next
t-soft
project
next
year.
I,
maybe
I
could
help
it.
I
could
help
you
with
this,
and
these
modern
songs.
D
Yeah
I
think
that
would
be
great.
We
should
evaluation
for
a
hacktoberfest
for
for
to
get
people
working
on
it.
So
yeah
we
have
the
data
for
the
cell
types
too,
and
we
have
like
the
Cell
nomenclatures
It's,
just
in
a
tabular
data
form,
so
I'll
I'll
send
that
to
you.
D
D
Tabular
data,
though,
for
the
cell
lineage,
like
the
the
nomenclature
and
everything
so
I
can
send
that
to
you.
We
can
put
together
something
I,
don't
know
where
we
want
what
we
want
to
do
with
them.
I
mean
in
terms
of
like
hosting
them
or
extracting
information
out
of
them,
but.
C
C
C
See
right
now
in
this
image
it
was
segmenting
it,
and
it
was
also
mentioning
that
it
is
Sky
Etc
yeah.
We
are
able
to
fine
tune
this
model
on
serial
Gans
data.
We
could
mention
like
what
is
the
cell
name
according
to
the
cell
x
y
z,
that
you
can
get
the
cell
image
analysis
and
the
position
of
the
cell
also
you
can
fine
tune
and
we
can
predict
like
develense
model.
Only
I
have
checked,
though
everywhere,
like
this
model.
C
Wasn't
there-
and
this
was
this
model-
was
primed
on
I
mean
lots
of
data,
so
it
could
be.
It
could
give
a
zero
shot
performance
on
Clans.
If
you
have
data
of
around
700
images
or
games
a
different
number
of
Sims,
it
could
give
better
results
as
model
I
thought
of
working
on,
but
I
have
a
lot
in
a
lot
in
more
writing
three
manuscripts,
making
a
clear
documentation
so
maybe
after
completing
this
project,
I
will
contribute
to
it
or
I'll
start
with
another
project.
D
We
yeah
I
think
definitely
Oktoberfest
would
be
good.
So,
let's,
let's,
let's
shoot.
E
B
D
And
then
I
can
get
the
data
to
you
and
we
can
set
up
something
for
people
to
to
use
like
you
know,
we'll
we'll
have
it
all
set
up
on
GitHub
and
then
you
know
basically
hacktoberfest
works
by
people
going
to
the
GitHub
repository
contributing
and
then
we,
you
know,
give
them
some
prize
for
participating
a
certain
amount
and
then
that's
pretty
much
it.
But
then
they
they
submit
work
and
get
credit.
So
yeah
I
think
that
would
be
good
I.
Think
two
years
ago
we
did
hacktoberfest.
It
was
actually
pretty
good,
so
yeah.
C
Yeah
I
think
it's
my
time
you
did
this
I
really
don't
remember.
I
have
participated
in
it.
E
C
Okay,
those
are
my
updates.
I
have
completely
worked
on
the
manuscript
of
these
three
papers,
comparing
the
models
getting
the
results
Maybe
by
next
week.
Once
you
read
these
papers,
what
I
am
writing?
I
have
shared
links
with
you
personally
and
in
the
chat
also,
if
they
are
there.
If
you
can
guide
me
through,
this
I
will
complete
these
papers
within
two
weeks,
and
we
could
end
this
project.
I
will
clear
the
documentation,
work.
Also,
okay,.
B
D
A
B
C
C
D
C
D
C
C
D
Okay,
thank
you.
So
thank
you,
sushman
for
the
updates.
Let's
see
we
had
a
couple.
People
joining
us,
dick
and
Lucas
and
Jesse
was
here
when
I
came
in
and
hamachi
was
here
and
we'll
get
to
hamacho's
update
now
it
looks
like
Susan
had
a
meeting
with
her
advisor
this
morning,
so
she
won't
be
here.
D
Hamacho
had
a
couple
of
things
that
he
mentioned.
He
mentioned
that
custom
annotations
would
be
required
for
the
tabular
data
is
that
in
reference
to
the
yeah,
the
stuff
that
I
said,
that
was
in
the
tables,
I
I
guess
well,
sushma
was
talking
about
a
certain
type
of
model
that
can
incorporate
those
data
or
be
specific
to
those
data.
So
segment
with
respect
to
the
locations
of
those
specific
cells
so
did.
Is
that
a
custom
edit?
Would
you
consider
it
to
be
a
custom
annotation
such
month
or.
C
I
guess
we
have,
according
to
the
cell
position,
if
we
have
the
name
of
the
cell
also,
it
would
be
useful.
C
D
And
then
Munchies
there
is
a
c
underscore
raw
underscore
data,
CSV
I'm
working
on
which
I
use,
which
was
used
in
devocraft
2..
It
has
a
data
set
of
the
lineage,
treating
as
coordinates
for
time
stamp
and
size,
so
yeah
I
think
that
was
one
of
the
files
I
was
thinking
of.
So
that's
on
the
GitHub
repository
in
the
GitHub
repository
for
Devo
worm,
not
Devo,
learn
and.
D
Well,
I
have
some
separate
data,
that's
not
on
our
GitHub
reposit
in
integrator
repository
I
can
get
that
together.
I
have
it
somewhere,
but
there's
like
a
video
where
it
actually
has
and
I'm
not
really
I
can't
remember
the
quality.
I,
don't
think
it's
like
the
highest
quality
video,
because
it's
pretty
old,
but
it
has
like
the
identities
of
the
cells
Mark
in
this
in
the
image.
I,
don't
know
if
that'll
help,
but
I
can
provide
that
as
well.
D
E
Yes,
okay,
so
right
now,
I've
been
I'm
working
with
the
like
assistant
of
allergy,
part
of
the
pocl
graph,
so
like
I,
was
for
figuring
out,
like
what
kind
of
forget
and
what
kind
of
or
topological
analysis
would
be
good
on.
The
data
set
that
we
have
currently.
So.
E
The
brief
overview
of
that
was
to
get
like
node
attributes
and
create
a
node
map
to
take
all
the
views
of
the
graph
and
apply
some
kind
of
filtration
function,
which
will
give
us
the
persistent
diagram
and,
after
that,
it's
just
aggregation
which
helps
us
to
like
incorporate
it
into
a
graph
neural
lens.
So
the
main
part
over
here
is
like
the
what
kind
of
filtration
functions
required
to
use,
and
that
gives
us
the
corresponding
persistent
diagram.
E
So
last
week,
I
had
a
developed
like
last
night
tutorial
on
this
low
star
image,
filtration
function,
which
was
used
on
the
data
set,
which
was
for
our
image
segmentation,
and
it
gave
treated
as
a
good
result
for
like
the
lifetime
Act
of
the
cell
and
and
after,
like
smoothing
the
data,
it
gave
a
diagram
similar
to
this
one,
which
was
a
Pino
years.
But,
and
also
after
he
plotted
like
the
points
or
this
was
the
output
which
gave
us
like
a
low
like
the
critical
points
in
the
image.
E
Coordination
and
also
the
size
and
I
was
working
on
the
like
I,
like
plotted.
The
entire
thing
and
oh
I
was
just
working
on
the
filtration
function,
which
was
which
would
be
helpful
for
the
this
kind
of
data
set
and
right
now,
I've
Googled
a
lot
of
things
but
I
couldn't
find
the
example,
which
was
exactly
for
a
use
case.
E
Working
on
it,
I
would
continue
doing
this
and,
after
that,
I'll
figure
out
a
way
to
operate.
This
cogn
repository,
which
was
like
the
main
people
which
are
referring
to
for
the
graph
new
Redlands
and
like
look
into
this
repository
and
find
out
how
to
use
its
components,
which
are
important
and
we
can
be
faster
and
use
it,
especially
like
the
persistent
homology
part,
like
this
persistent
analysis
and
another
thing
or
like
odd
analyze
for
sustainability
statistics,
part.
E
Also,
like
I,
couldn't
work
much
last
week
because
I
have
an
eye
infection
and
it's
it's
kind
of
like
a
sky,
but
it's
so
much
worse
and
it's
in
the
inner
Island.
So
I
can't
like
look
into
screens
that
much
which
was
really
hard.
Oh
yeah,.
D
Okay,
yeah,
that
sounds
good.
It
looks
like
you're
getting
there
so
yeah,
let's
yeah,
let's
keep
going.
D
That's
that's
good
I
guess
you
know
I
by
the
end
I
guess
we
should
have
at
least
a
sort
of
a
an
analysis.
Just
kind
of
sort
of
you
know
not
not
like
a
full
formal
analysis
with
just
a
few
and
that
you
know
a
few
analyzes
to
show
kind
of
what
the
graphs
should
look
like
what
the
output
should
look
like.
So
if
they're
tables
you
know
that
sort
of
thing
because
I
don't
know
yeah
I,
don't
know
if
we
know
like
or
I
I,
don't
know
necessarily
what
it
should
look.
D
E
D
A
all
right,
so
there's
some
manchu's
update.
It's
like
he's
making
progress
on
the
this
is
like
the
topological
data
analysis
stuff
that
he's
been
working
on,
and
this
is
basically
the
C
elegans
embryo
so
or
we
also
had
a
training
set
with
larval
images.
E
Yeah
I
use
the
different,
like
the
data
set
of
which
we
use
for
a
memory
segmentation.
This
was
the
one
yeah.
E
So
that
is
what
I
use
for
the
you
know,
a
filtration
function
for
the
image,
but
for
the
graph
I
am
for
the
graph
data
set.
I
am
working
on.
A
E
So
for
the
graph
data
set
it
there
will
be
a
similar
filtration
function,
but
it
would
be
dependent
on
the
distances
corresponding
to
it.
So,
like
these
points
are
similar
together
and,
for
example,
this
point
is
much
farther,
and
so
there
will
be
like
a
distance
function
which
would
give
like
a
spa.
E
A
matrix
with
respect
to
its
neighboring
points,
and
so
like
figure
out
a
way
to
such
that
the
for
each
time
stamp
and
for
each
point
or
like
data
in
the
diagram
of
each
cell,
basically
like
the
ab
cell,
the
ABS
cell
and
the
Abal
cell,
etcetera,
or
how
they
are
related
to
each
other.
D
Yeah,
so
thank
you.
Hamachu.
D
All
right,
so
that's
our
gsoc
updates.
Now
we
turn
to
dick
and
Lucas
and
myself
we're
working
on
this
paper.
D
D
Okay,
well
yeah
I,
I,
think
yeah
that
I
Incorporated
the
changes
now
Lucas
made
and
I'll
send
you
what
I
have
tonight
I've
corrected
all
the
references
as
well
when
it's
so
good
and
I'll
send
that
tonight.
B
D
D
D
Actually
I
can
go
over
it
a
little
bit
in
the
meeting.
Let
me
share
my
screen.
A
D
So
this
is
the
paper
abstract
introduction
it.
Basically,
it
goes
through
go
through
this
idea
of
differentiation
codes,
which
is
where
you.
B
D
So
this
is
figure
one
which
is
restylized
from
the
original
diagram.
So
it's
basically
where
you
have
in
this
case
we're
talking
about
a
binary
code
that
is
based
off
of
the
lineage
tree
or
of
differentiation
events
in
the
embryo.
So
you
have
these.
D
B
D
And
so
the
idea
is
that
you
have
it
reor.
You
have
the
different
lineage
tree,
basically
reorganized,
so
that
you
have
contract
things
that
get
smaller
or
contraction
waves
going
to
the
left
and
things
that
get
larger
expansion
waves
going
to
the
right.
So
that's,
basically,
this
whole
idea
of
a
differentiation
tree
or
a
differentiation
code,
and
so
in
in
animals
that
are
not
C.
Elegans,
like
the
regulatory
embryos,
regular,
regular
embryos
of
say,
like
Axolotl
or
Mouse.
D
These
circles
are
tissues,
whereas
in
C
elegans,
and
we
did
a
paper
on
this
back
in
2016.
These
circles
are
cells
of
different
size,
and
so
we
can
actually
reorganize
the
lineage
Tree
in
this
way
and
then
put
a
code
attached
to
the
different
constituents.
So
there's
a
zero
or
one
zero,
zero,
zero
one,
one
zero
one,
one
one,
and
then
from
that
we
can
do
different.
You
know
that
that
provides
a
code
throughout
the
embryo
and
then
that,
but
that's
only
based
on
cells
and
tissues-
that's
not
based
on
the
genome.
D
So
that's
a
different
thing
that
we're
trying
to
investigate
in
this
paper,
and
we
do
some
things
with
Gene
regulatory
networks
we're
looking
at
how
genes
are
expressed
and
the
sorts
of
things
that
they
produce
when,
when
they're
expressed
to
further
produce
proteins
which
should
show
up
in
amounts.
You
know
basically,
that
the
proteins
that
we
might
expect
to
be
involved
in
development.
D
So
we
go
through
some
of
these
Gene
regulatory
networks
and
their
products.
Here
we
talked
about
Gene,
duplication
events
and
other
things
that
can
add
to
this
network
back
to
its
complexity,
and
then
we
basically
link
that
back
to
some
of
the
differentiation
codes.
So
there's
this
binary
code
that
you
get
from
the
cells
and
you
can
link
that
back
to
these
genetic
regulatory
Networks.
D
Then
we
kind
of
move
into
an
analysis.
So
there's
some
theoretical
background.
We
move
into
an
analysis
of
epigenomic
mechanisms
and
proteomic
mechanisms.
D
The
epigenomic
mechanisms
are
these
presence
of
these
CG
motifs
within
candidate
genes
and
then
the
proteins
are
protein
lists
that
was
provided
in
dick
and
Natalie's
book
on
what
we
should
expect
to
find
in
terms
of
proteins
important
in
differentiation.
D
D
Okay,
so
this
was
actually
put
together
by
Lucas,
and
this
is
our
so
we
we
did
was
Lucas
went
into
genbank
and
he
did
a
blast
P,
which
is
a
protein
search
for
different
proteins
associated
with
a
certain
species
in
a
certain
identity.
So
we
have
this
list
of
proteins.
We
have
four
species
that
we're
looking
for
and
we
get
a
count
of
proteins
or
matches
that
return
and
those
are
basically
proteins
that
have
an
identity.
I.
Think
there's
like
you
actually
did
a
count.
D
I
can't
remember:
I
have
the
methodology
written
in
here,
but
basically
a
way
to
get
account
for
each
protein,
and
you
know
there's
certain
things
you
have
to
do
to
make
sure
that
you
get
you
don't
get
false
positives
and
things
like
that,
but
basically
we
get
a
difference
in
the
count
across
different
species
and
across
different
types
of
development.
So,
in
this
table
here
we
have
our
proteins.
D
The
range
from
you
know
of
there
are
a
number
of
Pathways.
There
are
things
like
junk:
the
junk
pathway,
the
map
and
map,
K
pathway,
PLC
pkc
row
and
rock
and
wind,
and
so
these
are
represent
different
Pathways
in
development
and
and
so
we
should
see,
depending
on
the
type
of
development
differences
in
these
counts.
And
then
of
course,
we
have
C
Sarah
VC,
which
is
yeast
and
yeast,
doesn't
exhibit
any
sort
of
development
at
all.
It's
a
single
cell
organism
that
lives
in
colonies,
they're
metazones,
which
means
they're.
D
D
We
have
drosophila
melanogaster
and
C
elegans,
which
is
are
different
forms
of
Mosaic
development
and
so
their
their
development
is
different
in
in
certain
ways,
but
they
have
the
same
type
of
Mosaic
development
that
we
you
know.
We've
talked
about
with
respect
to
the
differentiation
code
and
differentiation
tree.
D
So
you
can
see
that
there's
some
similarities
and
differences
in
the
cup
and
the
protein
counts.
It's
it's
kind
of
hard.
There
isn't
a
clear
signal,
but
you
do
get
cases
where
the
same
type
of
development
shares
like
some
protein
counts
that
are
very
high.
D
With
respect
to
the
other
two
examples:
you
get
cases
like
cdc42,
where
everything
has
a
lot
of
that
protein,
but
you
get
cases
where,
indeed
like
dvl,
drosophila,
melanogastrous
two
and
then
the
other
taxa
have
zero,
and
this
is
in
terms
of
account
that
we're
making
and
then
ecad
has
14
NC
elegans,
nothing
in
the
rest,
so
you
actually
do
get
species
specific
protein
expression
here
in
ecad,
and
there
may
be
another
case
where
that's
the
oh
yeah
rack
is
also
like
that
and
went
for
that
matter.
D
Of
course,
there
are
some.
We
actually
did
check
for
some
of
the
we
did
check
for
some
of
the
homologues
for
different
species
so
like,
for
example,
the
wind
pathway
is
important
in
in
drosophila
melanogaster,
but
it's
not
the
same
protein
as
Winton
C
elegans,
so
we
didn't
count
it
as
being
there,
but
you
can
see
that
Mouse
and
CCR,
VC
or
Yeast
have
a
count
of
zero.
So
these
are
different.
You
know
these
are
going
to
be
a
little
bit.
You
know
when
you
make
account
and
bioinformatics.
D
A
Are
we
allowed
to
who
the
snap
of
the
figure
8.13
from
Richard's
book
so
that
they
could
see
what's
the
23
proteins
in
the
figure.
D
I
mean
we
cite
it,
so
it's
kind
of
people
can
go
to
the
book
like
we
cite
the
book,
but
also
the
figure.
D
So
that's
I
mean
that's
usually
enough,
but
we
could
put
the
figure
in
as
well.
If
there's
confusion
in
the
reviewer,
the
reviewers
have
some
confusion
about
it.
We
might
do
it,
then
these
are
the
protein
normalized
counts.
So
these
are
graphs.
That
kind
of
compare
this
one,
a
is
where
we
compare
C,
elegans
and
drosophila
melanogaster
with
let's
see
with
yeast
in
the
top,
so
you
can
see
there's
some
that
are
yeast,
and
this
is
just
like
it
compared
to
so.
D
D
Sometimes
that
happens,
and
sometimes
you
get
cases
where
drosophila
resembles
East
or
C
elegans
resembles
East
for
different
proteins.
Then
you
have
this
case
here
which
is
not
East,
but
for
Mouse.
So
again
you
have
the
C
elegans
example.
D
In
blue
the
drosophila
example
in
red
and
then
the
mouse
example
in
Gray,
and
so
you
can
see
that
the
same
holds
true
a
lot
of
times
you
get
drosophila
and
mouse
sharing,
proteins
or
having
a
similar
like
Trend
in
the
count
and
then
C
elegans
being
different,
and
so
you
know
that's
that's
again
a
case
of
sort
of
an
intersection
of
you
know
what
we
can
capture
with
these
kind
of
counts,
different
proteins
being
used
for
different
things,
and
you
know
which
may
be
that
it's
not
really
the
effect
of
development
type
isn't
really
the
same
or
you
know
could
be
just
species
specific
things
like
C
elegans
is
a
very
specific
mechanism
for
something
or
drosophila
is
a
very
specific
mechanism
for
something.
D
So
that's
that's
the
protein
counts.
We
also
did
a
bunch
of
Canada
genes
for
the
proteins,
so
in
C
elegans
you
can
go
to
different.
There
are
different
resources
that
you
can
use
for
this
there's
worm
base,
there's
ncbi
and
there's
a
literature,
and
so
you
can
go
to
all
those
sources
and
find
analogs
to
different
proteins.
So
there
are
a
number
of
papers
where,
for
example,
you
could
look
up
a
protein
like
map
K
and
you
can
get
Canada
genes
for
that
protein.
D
So
in
C
elegans,
that's
actually
quite
common
and
there's
you
know
pretty
decent
set
of
papers
where
you
can
go
through
and
they
have
they've
identified
the
genes
for
the
protein
yeah.
It's
not
true
of
every
species,
but
in
C
elegans.
We
can
do
this,
and
so
having
done
that,
for
was
that.
D
A
D
Okay,
so
this
is,
this
is
basically
what
we
got
so
what
happened
here
was
we
have
these
genes?
We
can
go
back
when
we
know
what
the
Canada
genes
are
get
Gene
sequences
from
ncbi
and
then
analyze
those
Gene
sequences
for
the
CG
motifs.
D
Now
we
get
a
count
of
CG
motifs
across
these
different
genes,
but
across
also
the
protein
categories,
and
then
we
have
a
what
I
determine
information
count
which
is
basically
looking
at
the
variation
of
different
patterns
within
the
count.
D
So
we
have,
it
can
take
short
windows
of
12
bases
and
see
how
many
CG
repeats
we
have
within
those
little
Windows
of
of
DNA
and
the
reason
to
do
that
is
because
these
CG
repeats
tend
to
occur
in
islands,
and
those
Iowans
are
sort
of
tell
you
something
about
how
it's
regulating
and
so
or
how
the
gene
is
regulated
and
the
places
where
you
can
have
a
greater
regulation
of
the
gene.
So
this
is
based
on
the
literature.
It's.
B
D
Little
bit
different
mechanism
that
happens
in
mammals,
but
you
still
get
this
sort
of
effect.
So
this
is
a
summary
of
that
those
window
counts,
and
these
are
the
raw
counts
for
these.
So
you
get
these
motifs
in
different
genes
and
there's
actually,
if
you
look
across
different
genes
for
different
proteins,
so,
for
example,
rme6
and
rab5
both
involved
in
ck1,
they
have
different
counts
and
they
have
different.
Well,
not
quite
the
differ.
D
They
have
about
the
same
information,
but
they
have
different
counts,
and
some
of
that
is
an
artifact
of
different
lengths
of
the
sequence.
But
this
information
count
is
probably
the
most
informative
thing
in
that
analysis,
and
so
we
can
use
that
as
well
to
tell
us
something
about
how
the
gene
is
regulated
and
how
that
relates
to
the
protein,
and
so
there's
further
analysis
of
how
that
you
know
how
those
things
relate
to
one
another
in
different
proteins.
D
Then
we
build
these
Gene
regulatory
Networks,
which
are
constructs
that
you
know
where
you
sort
the
genes
and
they're,
sometimes
they're
in
different
chromosomes,
sometimes
they're
on
the
same
chromosome
and
you
sort
them
by
their
information
content.
The
things
that
have
a
higher
information
content
around
the
right
and
the
things
that
lower
information
content
are
on
the
left
and
the
the
ones
on
the
right
are
sort
of
attenuating
the
things
on
the
left.
D
And
then
you
have
a
protein
count
here,
and
so
this
this
organization
should
give
you
a
higher
or
lower
protein
count
and
so
give
I
give
a
couple
examples
here
for
different
genes
or
different
proteins
with
their
constituent
genes
in
this
network,
and
then
there's
a
discussion
of
that
and
then
that
maps
to
the
differentiation
Tree
by
being
sort
of
the
thing.
That's
in
this
this
cell,
you
know
the
thing
inside
the
cell:
that's
controlling,
potentially
this
state.
D
So
if
you
go
to
a
one
of
these
networks
is
in
cell
a
and
then
the
cell
a
divides
into
zero,
zero
and
zero
one.
So
it
serves
as
sort
of
a
switch
between
these
two
states,
and
so
the
question
is:
how
do
we
determine
the
threshold
for
switching
between
the
two
cells
and
that's
a
little
bit?
That's
kind
of
a
not
a
very
good
way
to
say
that.
D
But
it
there's
a
lot
to
consider
here
and
so
there's
a
little
bit
more
here
on
some
of
the
configurations
of
these
Gene
regulatory
networks
with
respect
to
the
differentiation
tree
and
what
you
might
see
inside
of
you
know
for
each
gene.
Is
it
evolves
in
it?
D
This
is
a
phylogeny
where
you
have
different
configurations
of
the
gene
regulatory
Network,
so
this
tells
us
that
there's
an
evolution
over
you
know
there's
change
over
Evolution,
but
there's
also
change
in
development
for
these
different
networks
and
they
translate
into
this
tree
and
then
ultimately
to
their
code.
D
That
I
showed
you,
and
so
that's
pretty
much
the
paper.
You
know
that
was
a
kind
of
a
whirlwind
tour,
but
I'll
send
out
a
copy
today
and
we'll
look
it
over
and
see
and
the
deadline
is
Thursday,
so
yeah
we'll
have
we'll
have
some
time
we
might
ask
for
an
extension,
perhaps
or
I,
don't
think
there
will
be
that
much
to
change
Maybe
so
we'll
see.
A
D
Right
so
yeah
there's
a
lot
of
a
lot
of
work
there,
thanks
to
Lucas
for
doing
such
a
good
job
on
the
protein
counts
and
helping
us
put
this
together
is
very
and
there's
a
lot
of
work,
that's
sort
of
that
he
produced
in
terms
of
like
what
kinds
of
things
are.
D
You
know
like
he
came
up
with
these
searches
and
he
found
a
lot
of
things
that
were
related
to
other
things,
so
that
that's
some
of
that
is
included
in
the
paper,
and
some
of
that
is
not
so
there's
a
lot
more
data
that
was
generated
from
this
analysis.
That
might
be
useful
for
other
things
and
I'm
not
sure
what
it's
useful.
For
necessarily
I
mean
it's
just
it's
a
lot
of
good
work.
I,
don't
know
there
might
be
some
other
opportunity
there
to
look
at
some
of
these
things
with
fresh.
A
Air,
you
might
have
to
ask
the
open
world
Foundation
I
could
talk
to
one
of
the
guys
from
there.
There
might
be
some
useful
data
specifically
because
they
need
like
they
make
Jupiter
notebooks
and
like
models
and
stuff
like
that.
I
can
do
models
with
that,
but
I
think
it's
useful
for
the
community.
I
have
to
ask
the
person
in
open
world
Foundation
or
maybe
a
seal
against
lab
I
only
noticed
that
so,
but.
D
Well,
we
can
do
a
lot
of
that
stuff.
We
can
put
together
like
Jupiter,
notebooks
and
things,
and
we
have
done
that
in
the
past,
where
we've
archive
stuff
like
that.
So
maybe
we
could
do
that
too,
and
they
could
do
it
through.
You
know
we
kind
of
we
kind
of
have
an
arm's
length
relationship
to
the
open
arm
foundation
in
terms
of
that
stuff,
because
they
do
a
lot
of
stuff
with
largely
with
like
neurophysiology
I
mean
that's
a
lot
of
their
data
sets
involved,
but
our
data
sets.
D
You
know
we
have
a
lot
of
development
specific
stuff,
so
we
might
be
able
to
put
together
like
a
public
data
set
where
we
kind
of
put
a
lot
of
synthesize
a
lot
of
the
stuff.
You
were
finding
because
you
know
there's
a
lot
of
interest
in
comparative
analyzes,
and
so
some
of
the
stuff
you
brought
up
I
think
was
very
useful
to
comparative
analysis,
and
that
might
be
something
worth
looking
at
more,
but
I'll
look
through
it
later
right
now,
I
want
to
get
this
paper
submitted
and
through
the
system.
D
Yeah
Jesse
Yeah,
you
can
take
a
look
at
it,
so
yeah,
we'll
okay,
so
that's
good!
D
So
now
I
want
to
turn
to
some
other.
Let's
see
so
we
talked
about
differentiation
code
paper.
I
have
this
interesting
paper
here.
Let
me
share
my
screen.
It's
a
classic
paper.
It's
on
what
we'll
see
in
a
minute.
D
So
this
is
a
classic
paper
under
esophual
mosaics
and
this
is
from
1972.
So
this
is
going
to
have
a
lot
of
hand-drawn
pictures
in
it
because
that's
what
people
used
to
do
in
biology,
they
used
to
draw
pictures
of
hand-drawn
things,
and
some
of
it
was
really
good.
It's
kind
of
a
lost
art,
but
so
there
are
no
like
images
in
here.
Microscopy
images,
it's
just
drawings.
Apparently
this
is
actually
about
talking
about.
So
the
the
headline
here
is
by
making
genetic
mosaics.
D
So
genetic
mosaics
are
basically
a
term
for
different
types
of
mutate,
mute
genotypes.
D
By
making
genetic
mosaics
and
constructing
embryonic
fate
Maps,
it
is
possible
to
locate
the
anatomical
site
of
abnormalities
affecting
Behavior,
so
can
see
elegans.
You
have
this
as
well,
where
you
can
create
different
mutants.
D
So
if
you
have
something
that
is
like
you
know,
a
certain
Gene
you
want
to
know
the
function
of
you
can
create
a
defined
mutant
for
that
mutation
in
that
Gene,
and
it
can
have
all
sorts
of
strange
behavioral
effects
it
can.
It
can
be
like
a
slow
crawler,
it
can
be
like
it
can
crawl
backwards.
You
know
these
sorts
of
things
and
they're
very
you
know
the
you
see
this.
D
Even
you
can
do
this
in
Mouse,
even
in
some
cases,
but
this
is
mainly
restricted
to
organisms
where
Gene
is
really
tightly
related
to
behavior,
so
it's
like
C,
elegans
and
drosophila
as
well.
So
you
have
this.
This
is
their
paper
where
they
kind
of
introduced
this
idea.
So
this
is
mapping
a
behavior
and
drosophila
mosaics.
D
Let
me
zoom
into
the
paper
a
little
bit,
so
they
talk
about
the
circuit
components
of
behavior,
from
sensory
receptors
to
central
nervous
system
to
affect
their
muscles
are
constructed
under
direction
of
the
genes.
So
we
have
now.
In
the
last
couple
years,
we've
gotten
a
number
of
different
subconnect
ohms
of
drosophila
we've
gotten
a
couple
of
the
different
circuits
for
behavior
and,
for
you
know,
motor
behavior
and
sensory
Behavior,
so
we're
kind
of
figuring
out
what
the
connectome
map
looks
like
now.
D
The
role
of
genes
in
that
is
a
little
less
clear,
but
we
know
at
least
what
that
connection
map
looks
like
a
mutation
affecting
the
structure
function
of
any
of
these
components,
May
alter
Behavior,
so
you
can
get
these
defined
mutants
that
sort
of
control.
You
know
they
can
change
the
behavior
of
this
connectome.
You
know
whether
it's
by
knocking
out
some
neurons
or
altering
the
way
that
the
connections
are
made
or
whatever
he's
in
classical
genetic
recombination
mapping.
D
The
locus
of
the
mutant
Gene
with
the
one-dimensional
chromosomal
array
can
be
determined,
so
they
do
this.
They
do
this
genetic
recombination,
mapping
technique
and
they're
able
to
get
sort
of
localize
the
location
of
this
mutant.
This
is
before
they
have
widespread
Gene
sequencing
by
the
way.
So
this
is
a
technique
they
had
to
use
to
get
it
like
identifying
these
mutants.
They
couldn't
just
sequence.
The
Gene
and
look
at
the
mutation.
D
A
D
Where
you
have
this,
you
know
where
to
how
does
the
gene
act?
It's
always
been
a
question
of
you
know:
where
does
the
gene
act?
What
does
it
do?
What's
the
function
so
like
in
the
paper
that
I
just
showed
you,
you
know
we
have
candidate
genes
for
specific
proteins
and
even
that's
not
a
tissue.
That's
a
protein,
and
but
some
of
those
genes
are,
you
know,
associated
with
other
proteins
as
well.
So
it's
not
a
clean
cut
Association,
but
nevertheless
that's
the
data
that
we
have.
D
D
So
now
we
have
this
a
bunch
of
interactions
working
under
the
hood
to
you
know
we
that's
the
great
mystery
of
genes
to
Pro
or
genes
to
phenotype
or
genes
to
behavior,
and
so
they
they
talk
about
this
Mosaic
technique
that
they
use
mosaics,
are
composite
individuals
in
which
some
tissues
are
mutant
and
the
rest
have
a
normal
genotype,
making
it
possible
to
identify,
which
part
must
be
mutant
to
produce
the
altered
Behavior.
So
this
is
an
example
of
the
drosophila
mosaics.
D
So
what
they
do
is
they
take
I
guess
they
could
do
this
in
development,
but
they
basically
take
different
cells
and
put
them
in
I.
Guess,
I,
don't
know
how
they
make
these,
but
well
I,
guess
you
can
do
it
through,
like
the
germline
like
putting
in
mutations
where
they're
only
expressed
in
certain
cells,
but
in
this
case
what
we
have
are
we
have
these
adult
drosophila
and
they
have
this
Mosaic
phenotype.
D
So
one
part
of
this
is
expressing
a
certain
is
derived
from
one
cell
line
and
the
other
part
is
expressed
from
another
cell
line.
So
in
this
case,
body
parts
derive
from
the
XX
cell
line,
which
is
shaded
or
female,
and
then
Parts
derived
from
the
XO
are
unshaded
or
male
and
also
Express
recessive
mutant
characters
uncovered
by
the
loss
of
the
other
X.
So
basically,
they
have
two
different
genotypes
in
them.
D
The
cells
on
this
side
of
the
organism
have
one
genotype
cells
on
this
side
of
the
organism
and
up
another
genotype,
and
so
that
means
that
you're
going
to
have
the
mutant
expressed
sort
of
locally.
So
you
can
have
it
like
down
the
midline,
which
you
can
also
have
it
in
different
organs.
You
can
have
it
in
the
head
and
in
the
wings
you
can
have
it
in
the
wings
and
maybe
parts
of
the
body
etc,
etc.
D
So
you
can
find
you
know
you
can
refine
this
technique
to
express
one
genotype
in
one
part
of
the
body.
I
think
this
is
maybe
a
lost
art
as
well,
because
now
we
have
things
like
crispr
or
we
have
other
types
of
retroviral
constructs
that
you
can
just
put
into
certain
tissues,
and
you
can
actually
get
expression
of
genes
locally,
but
this
is
something
that
is
a
sort
of
a
tradition
in
genetics,
creating
these
Mosaic
flies,
and
so
that
means
that
they
can
actually
create
a
map
of
the
body
parts.
D
So
this
is
a
really
fascinating
diagram
here.
This
is
where
you
have
like.
You
know
the
head,
the
thorax
and
the
abdomen,
and
you
have
these
body
parts
that
are
like
taken
apart.
It's
almost
like
a
if
you
ever
built
a
model
car
and
you
have
like
the
parts
lay
down
on
the
table
before
you
build
it
and
they're
just
kind
of
sitting
there
all
in
pieces,
and
you
have
to
identify
them
before
you
can
start
because
you
have
to
say
how
many
wheels
do
I
have
how
many
doors
and
so
forth.
D
So
this
is
what
it
looks
like
kind
of
you
have
these
parts
that
are
labeled,
they're,
all
kind
of
splayed
out
legs
and
the
art
in
the
the
forearms
and
the
head
and
everything
it's
just
kind
of
felt
like
this,
and
so
this
these
are
the
external
body
parts.
D
Each
part
is
formed
independently
from
an
imaginal
disk
present
in
the
larva,
so
in
during
metamorphosis,
the
parts
develop
into
the
adult
form
and
are
sutured
together
to
form
the
exoskeleton.
So
this
is
just
the
outer
shell
of
the
organism,
but
it's
exploded
like
this
to
show
the
parts
and
so
then
they
they
are
able
to
make
a
map
of
surface
landmarks
from
this
diagram
and
they're,
basically
trying
to
analyze
these
ipsilateral
ipsilaterally
and
contralaterally
across
the
body
and
they're
trying
to
find
some
sort
of
distance
between
these
landmarks.
D
So
you
do
this
in
the
normal
phenotype
and
you
do
this
in
the
meat
and
phenotype,
and
now
we
have
these
mosaics,
so
the
normal
phenotype
and
the
mutant
phenotype
are
in
the
same
organism.
It's
just
that
when
you
go
to
a
certain
part
of
the
organism,
it's
normal
or
it's
immune
and
so
you're
measuring
these
landmarks
across
these
different
parts.
D
It's
kind
of
a
hard
thing
to
wrap
your
head
around,
but
basically
you're.
Looking
at
the
effect
of
the
mutation
on
the
phenotype
on
the
shape
of
the
body.
Since
you
have
it
in
the
same
individual,
you
can
look
within
the
individual.
You
can
look
at
the
other
side
of
the
organism.
If
that's,
where
the
normal
versus
mutant
part
is-
and
you
can
compare
those
two-
you
shouldn't
see
a
lot
of
difference
because
there
isn't
a
lot
of
asymmetry,
fluctuating
asymmetry,
but
also,
if
you
look
across
the
organisms
to
another
normal
versus
a
mutant.
D
So
there
are
different
ways
you
can
measure
this.
So
this
is
the
formation
of
a
mosaic
fly
by
loss
of
One
X
chromosome.
So
this
is
where
this
happens
in
the
egg,
where
you
knock
out
an
X
chromosome
in
one
cell,
would
you
keep
it
in
the
other
cell
and
then,
as
the
cells
proliferate?
You
get
this
set
of
not
Knockouts
and
you
get
this
set
of
whale
type
cells,
but
you
don't
have
to
knock
out.
D
You
start
with
one
cell.
You
can
start
with
one
cell
and
modify
it
and
that
can
persist
throughout
development
and
that's
what
it
looks
like
at
the
end.
So
it's
very
interesting
and
then
they
build
this
fate
map.
So
this
is
what
a
fate
map
looks
like
in
drosophila
blastoderm.
So
you
have
this.
This
is
constructed
by
Mosaic
mapping.
D
D
So
this
is
a
principle
of
Mosaic
mapping,
just
showing
how
you
have
to
you
know,
find
the
distance
between
these
different
landmarks
and
what
they
should
look
like
in
the
mutant
versus
the
wild
type
cells,
and
so
you
can
actually
yeah
so
Mosaic
mapping.
D
So
this
is
called
drop
dead
and
this
is
I,
guess,
a
mutation
that
is
involved
with
the
worm,
dropping
dead,
I
guess
the
drop
dead,
mutant
flies
initially
develop
and
behave
normally,
but
at
some
time
an
individual
begins
to
walk
in
an
uncoordinating
or
uncoordinated
manner
within
a
few
hours
dies.
So
this
is
a
syndrome
that
they
express
over.
You
know
at
some
point
just
at
some
point:
it
just
there's
an
onset
and
it
goes
pretty
quickly.
D
After
an
initial
lag,
the
number
of
survival
survivors
drops
exponentially
with
a
half-life
of
about
two
days.
So
this
is
what
this
looks.
This
looks
like,
so
this
is
something
that
we
can
perhaps
look
at
markers
on
the
phenotype
to
see
if
we
can
identify
this
mutant
before
we
actually
observe
this
dropped
in
behavior
and
so
they're
looking
at
different
places
on
the
phenotype,
where
you
might
be
able
to
identify
the
mutant.
D
So
if
you
have
a
drop
dead
Mutant,
you
should
be
able
to
identify
it
from
the
actual
phenotype,
and
so
it
turns
out
that
a
lot
of
there's
a
lot
of
variation
and
it's
kind
of
hard
to
do
but
yeah,
and
then
they
talk
about
mapping
interacting
Foci.
D
So
again,
there
are
a
lot
of
interactions
that
are
underlying
this,
so
you
have
to
consider
mapping
these
interactions
between
genes,
which
is
kind
of
you,
know,
a
challenge
even
today,
with
a
lot
of
the
data
that
we
have,
which
is
much
better
than
this
data
they're
different,
you
know
so
the
various
configurations
of
the
Mosaic
boundary
we
have
to
know
kind
of
where
the
boundary
is
so
we
can
do
this
mapping
properly
and
genetic
interactions
make
that
even
harder.
So
that's
the
point
here:
yeah.
A
D
This
is,
then,
this
is
another
mute
Wings
up,
so
you
can
see
that
there
are
a
lot
of
mutants
in
drosophilus
as
well
as
C
elegans.
We
have
these
very
clear
effects
and
their
single
point
mutations,
so
it
makes
it
even
easier
to
identify
them.
The
problem
is
identifying
them
in
the
phenotype.
You
can
identify
the
behavior,
but
are
there
other
markers
in
the
phenotype
and.
D
To
get
at,
and
so
these
are
some
interesting
pictures
of
the
orientations
of
the
boundaries
of
these
mosaics,
so
you
have
differences
in
these
boundaries,
they're,
not
very
clear
in
the
diagrams
that
they
were
showing
it's
just
like
a
line
down
the
midline,
where
there's
specific
organs
that
have
one
genotype
versus
another
in
reality,
and
especially
in
the
in
the
embryo,
you
have
it's
it's
kind
of
more
of
a
contour
map,
I
guess,
and
it's
not
as
clear-cut.
D
D
It
involves
a
lot
of
drawings,
so
yeah
so
I,
don't
know.
If
you
have
any
questions
comments.
D
Okay,
so
that's
all
I
have
for
today
we'll
get
that
paper
out
to
people
this
evening,
all
right,
it's
for
attending
and
have
a
good
week.
E
D
Turns
out,
there
are
a
lot
of
ways
you
could
build
Gene
regulatory
networks
and
one
of
them
is
to
build
a
network
of
genes
to
see
you
know
what
regulates
what
what
another
thing
you
can
do
is
do
some
sort
of
dynamical
simulation
of
the
activity,
and
so
that's
what
this
paper
concerns
more.
So,
let's
go
over
this
a
little
bit
in
diagrammatic
form,
so
Gene
regulatory
network
is
basically
a
statement
of
Regulation.
D
D
A
lot
of
ways
this
can
be
done
in
the
literature.
There
are
a
number
of
examples
for
one
you
can
have
genes
such
as
they
are
in
a
relationship,
so
you
can
say.
First
of
all,
you
can
just
have
plain
Gene
networks,
and
these
are
not
regulatory
things
at
all,
but
they're,
just
kind
of
like
genes
that
are
related
to
one
another.
D
So
this
might
have
an
output.
This
might
have
an
output,
but
these
this
is
like
measuring,
say,
gng
interactions,
and
so
you
can
make
a
statement,
the
gene,
a
Gene,
B
and
Gene
crl
related
to
one
another.
They
might
be
associated
with
a
certain
protein.
They
might
be
associated
with
certain
phenotype
Gene
a
is
related
to
Gene
B,
so
it
has
an
effect
on
GB.
D
D
So
we
might
hypothesize,
then,
that
the
regulation
of
Gene
a
leads
to
the
down
regulation
of
Gene
B
or
the
up
regulation
of
geneal
of
the
down
regulation
and
Gene
B.
But
you
can
make
this
statement
about
gng
interactions
and
it's
just
basically
a
static
snapshot.
So
this
is
a
static
snapshot
of
a
gene
regulatory
Network.
D
But
of
course,
that's
not
all
genes
do
they
regulate
they
produce
products.
They
don't
just
sit
there
with
information
in
them
or
with
a
lot
of
people,
don't
like
the
information
metaphor,
but
I
think
it's
apt,
because
basically,
in
this
context
it's
telling
other
components
of
the
network
what
to
do
so
if
we
have
a
b
and
c
they're
regulated.
So
that
means
it's
something
is
making
copies
of
this
Gene
in
terms
of
transcripts
and
those
transcripts
are
going
off
and
doing
things
they
may
be
making
a
protein,
they
might
be
regulating
other
genes.
D
D
D
D
It
can
exist
in
different
ways,
and
so
this
is
kind
of
what
they're
getting
at
in
this
paper
in
this
paper,
though,
they're
talking
also
about
feed
forward
regulatory
logic,
so
they're
talking
about
a
feed
forward
system,
which
is
where
you're
going
from
A
to
B
or
from
B
to
C,
there's
no
feedback
C
day
would
be
feedback
A
to
B
and
B
to
C
would
be
feed
forward
and
this
regulatory
logic,
which
means
that
there's
a
certain
logic
where
you
know
certain
elements
have
a
certain
effect
and
you
can
actually
model
this
as
a
set
of
logic
gates
or
a
set
of
logical
statements,
controls
the
specification
of
differentiation
transition
and
terminal
sulfate
during
C
elegans,
underderm
development,
so
they're,
looking
at
the
endodermal
cells
and
c
elegans
and
as
we
talked
about
the
C
elegans
lineage
tree
is
deterministic,
and
then
it
always
produces
the
same
types
of
cells.
D
So
this
the
cells
hold.
They
have
this
sort
of
terminal
fate
where
they
produce
a
certain
cell
type,
but
that
cell
type
is
primed
early
on,
and
it's
always
going
to
that
type.
That
subline
edge
is
always
going
to
become
a
certain
type
of
cell.
D
So
the
abstract
reads:
the
architecture
of
Gene
regulatory
Networks
determines
the
specificity
infidelity
of
Developmental
outcomes.
We
report
that
the
core
regulatory
circuitry
for
endoderm
development
in
C
elegans
operates
through
a
transcriptional
Cascade,
consisting
of
six
sequentially
expressed
data
type
factors
that
act
in
a
recursive
set
of
interlock
feed
forward
modules.
D
So
these
are
six
data
type
factors.
These
are
just
genes.
They're
sequentially
expressed
they
acne
what
they
call
a
recursive
series
of
interlock
feed
forward
modules.
So
that
means
that
the
modules
are
nested
or
something
like
that
they're
self-referential
and
that
in
that
sense,
they're
intra
watts
to
their
behavior.
It's
kind
of
like
this
ABC
Motif
that
I
showed
you
here.
There's
a
feedback
element
here,
but
depending
on
I,
can
organize
this
so
that
it's
just
simply
nested
in
a
way
that's
functionally
relevant.
D
So
there
are
different
ways
you
can
organize
the
topology
of
these
networks
different
ways
you
can
turn
them
on
and
turn
them
off.
And
then
you
have
a
logic,
so
this
a
b
and
c
would
be
like
Gene
names
they
would
have.
Each
gene
would
have
some
sort
of
role
in
that
Network
and
we
could
model
basically
this
type
of
phenomenon.
B
D
Is
interesting
because
if
you
go
back
to
our
Network
here,
this
is
only
three
elements,
but
let's
say
we
had
something
like
a
heat
element,
Network
and
I'm,
going
to
turn
this
into
a
larger
sort
of
diagram.
Here,
let's
suppose,
I
had
eight
factors
here
for
and
I'm
just
getting
different
inputs
from
different
places.
Here.
D
D
Is
eight
factors
we
have?
Let's
see
I'm
going
to
do
this,
just
to
be
kind
of
a
feed
forward
kind
of
guy,
so
this
is
our
a.
This
is
our
B?
Our
B
goes
to
C.
It
also
goes
to
this
one
here
when
our
we
have
a
name
to
yet
and
see
you're
getting
inputs
from
the
environment.
D
D
is
here
e.
Is
here
yeah?
We
can
do
it.
That
way.
F
is
here,
G
is
here
and
H
is
here,
so
you
can
see
that
b
affects
both
C
and
H
or
C
and
G.
Then
C
affects
both
D
and
E,
and
C.
Both
affect
d
has
an
effect
on
F
this.
This
I
just
put
this
B
in
here.
This
is
not.
This
is
from
the
environment.
D
D
D
Well,
then
I
recover
the
effect
of
F
and
the
effect
V,
which
is
minimal,
and
so
there
may
be
some
modulatory
effect
in
d.
That
e
is
having,
but
it's
it's
mitigated
by
the
knocking
out
of
f,
so
that
has
sort
of
the
same
effect
as
f,
but
let's
knock
out,
f
and
g.
D
D
You
don't
have
an
output,
don't
have
an
outlet.
So
basically
it's
this
part
of
the
network
kind
of
operating
on
its
own,
not
doing
not
finding
a
way
to
express
itself
and
then
h
being
the
thing
that
doesn't
have
anything
it's
regulating
it
and
the
whole
network
falls
apart.
So
this
is
how
you
get
these
differential
effects.
You
have
to
find
the
right
location
to
make
these
mutations,
but
the
way
this
specific
network
is
set
up,
one
one
mutation
does
or
one
knocking
out.
D
One
factor
doesn't
have
a
much
of
an
effect,
but
knocking
out
two
factors
can,
and
so
this
is
what
where
I
was
trying
to
get
at
here,
which
is
that
these
Cascades,
which
are
these
sequential
activation
or
factors
can
have
you
know,
depending
on
the
topology
of
the
network,
can
have
wildly
different
effects.
D
D
The
phenotypic
strength
is
successfully
predicted
with
a
computational
model
based
on
the
timing
and
levels
of
transcriptional
states.
So
if
we
go
back
to
our
first
diagram,
we
see
that
there's
this
promoter
repressor
it's
producing
an
output
of
this
Gene,
so
the
products
of
this
Gene
have
an
output.
The
promoter
tends
to
make
more
copies.
D
The
repressor
brings
a
number
copy
number
down,
and
so
this
is
what
they're
getting
at
here
is
that
you
can
actually
the
timing
and
levels
of
transcriptional
states
can
serve
to,
because
some
genes
are
dependent
on
genes
that
are
Upstream.
They
can
actually
have
these
kinds
of
effects,
so
you
can
have
a
repressive
effect
without
actually
having
like
a
repression,
regular
repressor
regulatory
element.
D
So
this
is
where
this
Gene
regulatory
Network
acts
to
establish
sort
of
these
State
domains,
which
are
you
know
where
you
have
different
areas
that
are
basically
upregulated
or
non-regulated,
and
this
is
a
different
state
than
what
would
be
say
inside
the
gut.
So
if
you're
at
the
boundary
of
the
gut,
your
Gene
regulatory
Network
would
have
a
different
state
than
inside
the
gut,
so
you
can
Define
boundaries
in
this
way.
D
They
talk
about
the
common
genetic
toolkit
for
the
generation
of
diverse
animal
forms,
data
transcription
factors
play
a
conserved
role
in
the
development
of
diverse
cell
types,
including
those
of
the
endoderm,
the
first,
the
three
germ
layers
to
have
evolved
during
the
late
Precambrian
era
in
in
the
dipoblastic
phylonidarians
and
coniferans
peripheral
peripherals
data
factors,
a
specifically
expressed
in
the
endoderm
or
an
endoderm-related
cells,
suggesting
that
these
transcription
Regulators
may
have
driven
the
development
of
the
endoderm
germ
layer
and
gastrulation
at
the
dawn
of
metazone
evolution.
D
So
these
data
pathways
are
really
kind
of
important
for
regulating
cell
State
and
so
that
there's
an
interaction
with
some
of
the
proteins
we
were
talking
about
in
the
differentiation
code
paper
with
some
of
these
other
networks.
And
so
we
don't
consider
those
networks
specifically,
but
we
do
we're
sort
of
proposing.
A
very
general
mechanism
says
it
turns
out.
There
are
a
number
of
mechanisms
that
people
are
identifying
in
C
elegans,
for
example,
for
specific
cells
that
lead
to
the
differentiation
of
different
tissues
or
cell
types
and
see
elegans.
D
So,
as
I
said,
you
know,
different
parts
of
c
elegans
and
different
tissues
have
a
very
deep
origin
in
the
in
the
lineage
tree,
so
the
endoderm
arises
from
a
single
blastomer
formed
at
the
eight
cell
stage,
known
as
the
eso.
D
So
there's
a
sub
lineage
called
the
E
sub
lineage
and
the
E
sub
lineage
is
actually
going
to
go
on
to
form
a
lot
of
the
what
they
call
the
endoderm,
which
is
the
gut
and
some
other
parts
of
the
in
the
intestine
and
and
that
sort
of
area
of
the
animal
endoderm
development
is
driven
by
three
pairs
of
duplicated
genes,
encoding
data
like
transcription
factors,
and
they
talked
about
Med
one
and
med2
n13
and
elt27,
and
so
these
are
canonical
factors
and
these
are
Divergent
factors.
D
D
So
they
talk
about
sequential
redundancy
and
this
suggests
feed
forward
regulatory
circuitry.
So
this
you
know
this
is
kind
of
what
I
was
showing
before
I
showed
kind
of
a
loose
example
of
it's
not
based
on
this
paper.
But
it's
based
on
like
this
idea
of
having
different
feed
forward
configurations
of
the
factors
so
that
you
know
there
is
this
redundancy.
D
The
redundancy
that
you
see
in
those
kind
of
networks
is
evident
of
the
phenotypes
of
single
mutants.
Most
of
the
genes
expressed
in
the
C
elegans
endoderm
grn
would
show
either
no
overt
phenotype
or
weekly
penetrant
phenotype,
meaning
that
there
is
the
effect
of
that
mutant
isn't
great,
and
so
you
know
you
don't
see
it
in
like
penetrance
means
how
many
What
proportion
of
population
are
expressing
this
factor,
and
then
you
know:
if
it's
not
a
parent
in
the
phenotype,
then
it
doesn't
necessarily
have
a
strong
effect.
D
So
this
is
where
they
kind
of
talk
about
some
of
these
things.
D
You
know
you
could
remove
genes
and
you
can
see
what
their
effect
is
like.
We
saw
in
the
paper
with
Mosaic
mosaicism
in
drosophila.
D
D
This
observation
is
consistent
with
the
alternate
possibility
that,
rather
than
the
pairs
of
factors
acting
together
at
specific
tiers
in
the
Cascade,
each
factor
is
a
redundant
with
its
immediate
upstream
or
Downstream
factor
in
a
continuously
sequential
Cascade
or
what
they
call
sequential
redundancy.
D
D
So
you
see
what
they're
doing
is
they're
taking
what
I
did,
which
was
a
thrf
they're,
just
kind
of
making
up
like
boxes
and
and
they're,
saying
that
they
have
this
additive
effect
tier
and
then
they
kind
of
go
through
this
with
mapping
genes
to
those
boxes.
So
they
say
Med
one
and
two
are
a
plus
b
c
plus
d
is
n,
one
and
three,
and
then
E
and
F
is
elt2
and
elt7.
D
Furthermore,
you
have
this
feed
forward.
Network
a
b
and
c
a
can
also
affect
C.
A
can
affect
b
b
can
affect
C,
and
then
you
have
this.
This
mapping
to
the
genes,
so
med2
has
an
effect
on
Med
one
that
one
as
an
effect
on
N3,
Mid
2
has
an
effect
and
then
three
n
three
and
one
as
an
effect
on
N1.
D
So
it's
really
kind
of
like
you,
have
this
linear
pathway
from
med2
to
Mid
one
and
three
to
end
one
elt70
lt2,
but
you
also
have
these
shortcuts
that
are
feed
forward
from
E
and
D1
to
elt2,
for
example,
or
from
Med
to
end
three,
and
so
that's
that's
the
topology
of
their
regulatory
Network
and
they
show
the
viability-
and
you
know
some
of
these
other
factors
that
they're
measuring
in
these
different
mutants.
So
these
are
defined
mutants
on
this
graph
here
and
see
so
they
have
these
different
effects
of
genes.
D
So
they
have
the
single
genes,
single
mutants,
Med,
II,
Med,
one
and
three.
This
is
just
where
you
have
that
mutant
in
the
your
mutant
phenotype.
Then
you
have
these
sequentials,
so
you
have
mutants
in
double
mutants,
basically,
where
you
have
mutations
and
two
genes,
but
they're
in
the
same
regulatory
Network,
so
they're
sequential.
So
you
have
med2
and
Med
One.
So
there's
a
sequential
effect
there
and
you
have
a
double
mutant
and
that
shows
that
there's
it
has
a
different
viability
than
the
others.
D
Then
you
have
alternate
so
you
have
Med
tune
and
one.
So
this
is
where
they're
not
directly
affecting
one
another
they're
sort
of
indirectly
affecting
one
another
which
you
have
double
mutants
in
this
case,
where
you
have
two
of
these
genes
that
are
mutated
but
they're
not
directly
affecting
one
another,
and
this
just
shows
the
viability
here.
So
you
can
see
that
there's
differences
in
viability.
D
You
have
these
terminal
differentiation
genes
down
here
and
they're
affected
by
this
pathway.
So
you
have
two
different
topologies
you
have
in
J.
You
have
this
feed
forward
pathway
with
the
feedback
in
elt27,
and
then
this
one
in
K,
where
you
have
more
feedbacks
and
more
shortcuts
across
the
feed
forward
components
of
the
network,
and
you
see
that
there's
a
fraction
of
viable
animals
is
higher
while
they're
both
pretty
high,
but
there's
this
other
Factor
early
expression
of
elt2
relative
to
Wild
type.
D
This
is
a
predicted
amount,
so
you
can
see
that
in
these
these
organisms
there's
much
less
early
expression
of
elt2,
in
this
case
it's
more
diffuse
in
the
sequential
case.
So
this
is
a
yeah.
This
is
an
interesting
set
of
graphs.
I,
don't
know
we
could
barely
differentiates
the
two
topologies
that
much,
but
that's
basically
the
relationship
we
would
expect
to
see
on
the
terminal
we
differentiated
genes.
D
D
They
perturb
the
two
models
in
different
ways,
so
they
can
simulate
these
mutations,
they're
Computing,
the
expression
levels
of
different
genes
and
the
computed
elt2
expression
levels
in
the
turret
model,
only
weakly
correlated
with
the
phenotypes
of
the
mutant
combinations.
D
D
D
A
D
Model
both
of
these
in
a
computer,
so
in
contrast
when
the
feed
forward
model
was
tested,
our
corrupted
results
predicted
that
elt2
expression
would
occur
with
the
weight
onset
in
all
single
mutants
and
it
means
liking
alternate
pairs
of
get
effectors.
However,
elt2
expression
was
predicted
to
be
completely
abrogated
in
double
mutants,
in
which
sequential
members
of
the
endoderm
Cascade
are
removed.
So
this
is
in
the
sequential
network,
if
you
remove
those
there's
an
effect,
and
this
is
consistent
with
their
pronounced
developmental
defects.
D
So
you
can
do
this
where
you
can
take
expected
results
from
phenotypes
That
You
observe
you
can
model
different
types
of
network,
topologies
different
types
of
relationships,
and
you
can
get
it.
You
could
work
out
what
you
should
expect
from
those
different
topologies
and
then
you
can
make
sort
of
an
educated
guess
as
to
what
the
actual
genetic
regulatory
Network
looks
like
then
there's
synergistic
requirements
and
cross-regulatory
interactions
of
N1,
elt7
and
elt2.
D
Logistically,
which
means
they
both
need
to
be
there
all
three
need
to
be
there
to
have
the
effect,
but
it
has
a
greater
effect
than
just
the
additive
effect
that
we
might
normally
expect,
and
so
they
they
work
together
and
they
have
a
an
effect
on
phenotype,
as
described
above.
Both
elt7
and
elt2.
Single
mutants
contain
a
fully
differentiated
gut,
which
shows
a
contiguous
Lumen
from
the
pharynx
to
the
rectum,
which
is
surrounded
by
cells
of
normal
differentiating
morphology.
So
this
means
that
this
these
mutations
did
not
have
an
effect
independently.
D
These
double
means
invariably
lacked
both
a
defined
gut
Lumen
and
at
least
some
overtly
differentiated
gut
cells
I.E
an
apparent
sporadic
all
or
none
to
block
that
differentiation
along
the
length
of
the
animal.
All
the
differentiations,
highly
defective
in
the
absence
of
elt2
and
elt7
patches
of
all
differentiated
gut
were
nonetheless
evident
or
over
many
terminal.
Differentiation
genes
continue
to
express
in
the
absence
of
elt2
and
elt7.
D
So
this
is
an
important
point,
because
it's
not
only
that
these
genes
have
to
act
synergistically
to
have
an
effect
where
they
both
need
to
be
mutants
and
to
have
an
effect,
but
we
don't
know
the
full
extent
of
this
network.
It's
an
incomplete
Network,
and
so
there
are
a
lot
of
things
that
are
coming
from
the
outside
that
are
regulating
it.
This
is
a
Beyond
like
the
maybe
the
environmental
effect
on
the
network.
D
So
there's,
like
you,
know,
transduction
of
signals
from
the
environment
of
the
cell,
or
you
have
other
chemical
signals
that
that
activate
this.
You
can
also
have
other
genes
that
are
involved,
and
so
this
is
something
that
you
know
is
a
matter
of
inference.
It's
not
just
that.
You
know
these
genes
are
self-contained
and
that's
that
these
are
just
a
Canada
genes,
and
so
we
have
to
work
with
what
we
have
in
terms
of
the
candidates,
sometimes
they're,
predictive
and
sometimes
they're,
less
predictive.
D
D
So
there
are
a
lot
of
things
it
does
in
establishing
the
boundaries
of
the
system,
different
cells
of
the
system,
different
components-
and
this
is
of
course
another
set
of
examples
where
you're
knocking
out
different
things
and
the
actions
maybe
of
outside
factors
like
pal
one
and
wind
map
can
SRC.
So
you
can
see
that
there
are
different
things
that
this
Gene
Network
just
doesn't
operate
in
isolation,
it's
affecting
other
things,
it's
being
affected
by
other
developmental
factors
and
it's
affecting
other
developmental
factors.
So
it
can
shut
down
other
developmental
factors.
D
So
that's
all
I
had
to
talk
about
with
that
paper.
Hopefully
that's
food
for
thought
and
thank
you
for
listening.