►
Description
Attendees: Bradly Alicea, Mayukh Deb, Ujjwal Singh, and Richard Gordon. GSoC Updates and discussion, discussion of open project issues, papers from the reading queue.
A
B
A
A
A
A
B
Oh,
we
went
camping
in
a
place
north
of
dolphin
manitoba.
D
B
D
A
D
E
D
A
B
E
A
Yeah,
so
that's
good
and
and
you've
been
working
hard,
the
last
couple
years,
so
that
should
hopefully
that's
what
paid.
C
A
Do
you,
what
do
you
have
an
update.
C
D
Image
and
then
a
discriminator
tries
to
discriminate
whether
or
not
it's
a
real
image
or
not
okay.
So
there
are
basically
these,
like
it's
trained
on
images
like
these,
like
these
are
the
kinds
of
images
that
it's
trained
on
suppose
the
generator
like
this
part.
First,
this
part
will
try
to
generate
a
model
like
generate
an
image
of
an
environment,
and
then
this
part
would
try
to
classify
whether
or
not
that
image
is
fake
or
real,
and
as
the
pin
goes
on,
both
of
them
get
stronger.
D
As
they
train
so
at
the
end,
the
generator
becomes
better
and
better
and
better
with
further
training.
So
that's
what
happened
and
like
the
discriminator
returns
a
probability
between
zero
to
one
on
what
it
thinks
is
the
probability
of
the
generated
image
being
as
close
to
the
real
image,
so
the
ideal
image,
so
the
ideal
value
for
the
discriminator
at
the
end
would
be
0.5,
because
it's
perfectly
in
the
middle,
it
can't
decide
whether
or
not
it's
like
like
whether
or
not
it's
real
or
fake.
D
D
That
I
just
generated
a
few
samples
and
like
that
is
it
for
this
notebook
and
other
than
that.
What
I
did
was
that
I
extended
this
like
this
is
a
trained
model
right.
I
saved
this
train
model
and
then
I
used
this
model
and
I
extended
it
into
that.
I
extended
and
I
kept
this
model
as
another
utility
within
like
this
library
that
I'm
working
on,
so
the
user
basically
generate.
B
D
D
D
B
D
Like
in
bulk
and
then
like
what
I
did
was
that
I
just
go
back
to
the
ready.
D
A
D
D
D
D
D
That
has
those
populations
exactly
like,
or
approximately
those
problems
like
if
I
enter
like
one
one:
zero:
zero,
zero,
zero
so
like
it
means
really
low
population
right
for
all.
The
lineages
people
generate
a
very
sparse
kind
of
an
image
with
about
like
very
few
cells,
but
if
I
give
it
like
255
50
32,
something
like
that
there,
you
generate
a
really
like
an
embryo
which
is
almost
in
its
l1
stage,
because
it's
like
almost
there
right
now,
I'm
working
on
this,
it's
pretty
much
the
same
initially.
D
B
D
B
D
D
On
it
has
two
arguments
like:
can
you
guys
see
here?
It's
got.
Output
and
label,
like
the
generator
looks
at
both
of
these
values.
Label
means
the
population
label
means
these
values
like
the
population
values
just
below
these.
Like
you
can
see
it
takes
both
of
these
arguments
and
tries
to
generate
some
kind
of
an
image
with
that
corresponding
level.
But
it's
not
working
right
now.
The
code
is
basically
it's
just
broken.
Just
don't
look
very
sunny
stuff
yeah.
So
that's
pretty
much
my
opinion.
I'll
start
with
this.
After
my
semester,
exam
center.
A
Oh
yeah,
that's
pretty
impressive
yeah.
I
saw
something
about
that
in
the
slack
as
well.
So
this
is
so,
let's
go
back.
Let
me
think
about,
I
think,
through
this,
through
a
little
bit,
I'm
going
to
bring
the
share
your
screen
again.
A
So
you
have
this
yeah,
so
you
have
this
gan,
this
generative
adversarial
network
that
you're
using
to
build
these
embryos
and
you're
using
the
data
that
you
got
from
the
data
center
generate
populations
yeah.
I
think
that's
actually
pretty
interesting,
that
you
have
like
these
data,
that
you're
kind
of
plugging
in
and
then
you're
generating
these
fake
embryos
and
then
you're,
seeing
which
ones
you're
trying
to
classify
them
as
real
and
fake
and
then
no.
D
D
D
E
D
D
D
D
A
Yeah,
so
I
pointed
out
that
a
couple
years
ago
we
did
the
morphozoic
paper.
We
did
the
morphozoic
model,
which
is
a
cellular
automata
that
it
sort
of
like
generates
patterns
and
then
classifies
them.
So
it's
not
really
like
this,
but
it's
kind
of
a
similar
type
of
thing
where
you
have
kind
of
generated,
I
guess
like
they're,
fake
embryos,
it's
generating
patterns
and
then
it's
classifying
them
into
different
things,
and
it
has.
B
A
So
you
have
the
the
cellular
automata
and
then
you
have
this
neural
network,
it's
a
standard,
neural
network
and
then
it's
classifying
different
patterns
that
are
being
generated
as
like
an
embryo
or
not
an
embryo,
or
something
else
that
I
like
this.
The
way
this
works.
This
is
the
first
again
so
we
talked
about
that.
I
think
last
fall
in
more
detail,
but
the.
E
Work
so
like
what.
E
C
A
Yeah,
thanks
for
the
update,
we'll
keep
yeah
we'll
revisit
this
in
coming
weeks.
I
don't
know
well
we'll
see
how
you
know
it
goes
and
then
eventually
maybe
we'll
put
this
together.
D
A
Well,
yeah,
I
guess
we
could
think
yeah
I
mean
we
have
the
names
that
were
sort
of
proposed
at
the
beginning
of
the.
When
you
applied
projects
we
don't
have
to
use
those
like.
Is
it
possible
to
make
polls
on
slack,
oh
yeah,
we
can.
I
think
you
can
do.
D
E
A
A
A
D
D
D
A
Yeah
I
like
that
idea,
so
yeah
we
can
send,
you
know,
send
some
candidate
names
out
to
see
who
people
like
and
then
yeah.
A
I
think
the
usability
is
a
good
issue
to
focus
on
too
that
you
know
being
able
to
make
it
easy
to
use,
instead
of
even
having
that
problem
in
one
of
my
in
one
of
my
other
groups,
where
they
don't
like
the
you
know,
you
everyone
thinks
that
it's
easy
to
use
like
some
notebook
that
we're
trying
to
use,
and
it's
not
as
easy
as
we're
running
into
some
problems
with
it.
So
it's
never
a
bad
idea
to
make
it
as
simple
as
possible.
It's
simply
because
it's
you
know
sometimes
it'll
work.
A
Sometimes
it
won't
supposed
to
be
universal,
but
I'm
not
sure
about
that
anyways.
So
thank
you
for
the
updates.
I
wanted
to
mention
that
the
second
round
of
reviews
are
coming
up
this
week
for
g-sox,
so
I'll
be
filling
those
out
yeah.
A
So
I
mean
this
is
again.
You
know
I'll
just
fill
them
out
and
it's
you're
both
past
I
mean,
I
don't
think
that's
that
comes
as
a
great
shock,
but
you
know
I
will
be
suggesting
some
things
you
know
like
they
always
ask
you:
what
could
they
improve
on
so
I'll?
Think
of
something,
and
then
you
know
we'll
we'll
keep
going
on
this
in
the
coming
weeks,
so
yeah
so,
okay,
let's
see
what
do
we
have
that
I
wanted
to
go
through?
Why
don't
we
move
on
to
maybe
going
over?
A
A
A
Some
of
these
are
ongoing,
so
the
outline
of
action
items
for
the
periodicity
paper
that
was
done
last
week,
but
we're
still
working
on
that.
So
that's
this
pic.
This
is
the
repo
here
and
then
the
study
plan
is
here.
So
this
idea
of
creating
this
paper
about
periodicity
in
the
embryo,
so
we've
got
a
division
of
labor.
Here
people
are
interested
in
a
specific
role.
A
They
can
contact
me.
I
put
downs,
I
didn't
put
down
susan,
so
you
know
I
mean
she
tells
me
what
kind
of
specific
role
she
wants
to
play
I'll
put
her
in
the
list.
A
A
We
have
some
data
sets
that
we
would
use
for
this
paper.
We
have
zebrafish
a
simulated
embryo
and
c
elegans,
and
then
we
have
some
ideas
of
how
to
compare
them.
So
it's
still
ongoing.
I
haven't
made
any.
We
haven't
made
any
advances
on
that,
but
that's
in
progress,
the
axolotl
embryo,
animations
and
segmentation
is
still
in
progress.
A
Susan
said
she
might
send
me
some
more
data
on
that,
but
other
than
that
that's
kind
of
been
on
hold,
but
I'm
kind
of
having
it
in
progress,
because
I
you
know
still
kind
of
working
on
getting
that
together
and
getting
it
going
the
friday
coffee
hour,
hackathon
that
is,
that's
been
put
on
hold
for
a
couple
weeks.
A
A
Complexity
measures
we
haven't
talked
about
that
in
a
while,
but
that's
something
that's
involving
jesse
was
interested
and
dick
might
be
interested
and,
of
course,
george
mickelovsky.
I
haven't
really
contacted
him
about
that.
Yet,
but
a
model
of
motel
sells
an
embryo,
not
really
sure
what
that
was.
A
That
might
be
the
thing
related
to
g-suck
create
embryo
model
in
blender,
so
this
was
something
that
bourgeois
was
working
on.
Of
course,
this
isn't
like
the
priority
during
gsoc,
if
he's
doing
g-sock
stuff,
so
I
didn't
want
to.
A
A
You
have
to
revisit
that
at
some
time
at
some
point
in
the
meetings,
but
that's
still
in
progress.
Krishna
said
he
was
going
to
present
a
paper.
He
hasn't
done
that
yet
he's
that's
why
I
put
it
on
hold,
but
he
has
done
some
stuff
with
tutorials
and
he's
created
some
tutorials
in
for
the
neuromatch
academy
sort
of
our.
My
other
group
is
working
on
neuromatch
academy
stuff,
so
he
worked
on
some
tutorials
for
that.
A
So
like
some
coding,
tutorials
and
things
like
that,
so
those
are
actually.
I
didn't
pull
them
up
for
the
meeting
today,
but
they're
pretty
interesting
that
one
create
narrow,
match
kinds
of
diva
worm.
We
didn't
really
do
that.
Nero
matches
this
summer
school
that
I've
been
working
with
the
neuromatch
academy
on
and
it's
about
it's
week
three
of
three,
so
we've
been
busy
doing
different
things.
A
We
never
really
created
a
lot
of
tie-ins
with
it,
but
we
might
do
that
after
the
after
the
academy.
Just
kind
of
you
know,
with
the
contacts
that
we
made
there
so
discuss
ai
innateness,
that's
on
hold
reorganized,
evo
zoo.
This
is
something
I
think,
probably
that
we
can
put
in
finish
since
we're
kind
of
that's
kind
of
a
gsoc
thing.
Oh,
it's
not
signed
in
okay,
well
I'll,
put
number
seven
and
finished.
A
When
I'm
logged
in
here,
okay,
axolotl
montaging,
that's
something
that's
something
that
is
on
hold
because
you
know
when
we
start
moving
on
this
one
issue
on
the
axolotl
stuff,
then
we
move
on
this
other
one,
because
they're
kind
of
linked
the
models
for
biological
living
networks,
I'm
not
sure
what
that
is,
but
that's
on
hold.
A
Let's
see,
there's
some
other
papers
here
that
are
or
ideas
that
are
in
the
to-do
list.
I
don't
think
any
of
these
are
really
something
that
we
are
immediately
able
to
do.
We
still
have
this
docker
container
for
open
worm,
the
embryo
visualization.
A
That
actually
might
be
something
number
six,
that
we
can
take
some
of
the
stuff
we've
been
doing
in
devo
or
in
summer
our
code
and
contribute
to
so
I
got
a
email
from
from
one
of
the
people
at
openworm
they're
having
their
annual
meeting
in
september,
and
he
wants
a
five-minute
presentation
from
diva
worm.
A
So
I'm
gonna
go
over
some
of
the
stuff
we've
been
doing
this
year
and
I
we're
going
to
do.
A
Of
course,
I'm
going
to
do
some
of
the
deep
learning
stuff
and
then
maybe
put
in
some
other
things,
and
you
know
I
don't
know
five
minutes
is
a
lot
of
time
for,
but
it's
you
know
we'll
probably
do
that,
and
then
we
also
have
this
docker
container,
where
all
the
other
projects
have
contributed
some
sort
of
something
you
can
run
in
the
browser
or
something
you
can
run
in
the
run
on
your
hard
drive.
A
That
is
like
an
illustration
of
the
you
know
that
project,
so
some
of
the
other
projects,
because
they're
based
around
actually
just
designing
a
software
package,
it's
pretty
easy
for
them.
They
say:
okay,
we'll
just
do
the
software
package
and
you
know
you
can
run
it
in
there
and
that's
it.
We
have.
We
don't
really
have
a
single
software
package
associated
here
with
the
group,
but
I
think
maybe
something
with
like
that's
what
I
was.
You
know
we're
talking
about
the
animation
earlier,
but
it
might
be
appropriate.
A
Maybe
to
do
something
like
one
of
the
examples
of
the
pre-trained
models
or
something
I'm
not
really
sure.
What's
appropriate,
I
don't
want
it.
It
can't
be
something
that's
that
takes
a
lot
of
processing,
that's
the
problem,
so
that
might
actually
not
be
appropriate
for
it.
They've
had
some
trouble
with
some
of
the
simulations
like
spitting
out
a
bunch
of
files,
output
files
and
only
in
people's
hard
drives.
A
A
All
right,
let
me
go
back
and
re-share
my
screen.
There's
one
more
thing:
I
wanted
to
talk
about
on
the
board,
so
yeah
there's
this
lecture
on
pca,
umap
and
t-sne.
A
So
I've
noticed
that
a
lot
of
the
papers
that
people
that
you
read
on,
like
not
just
in
developmental
biology
but
in
a
lot
of
areas,
they're
doing
anything
with
big
data,
they're
doing
something
with
these
different
methods,
and
so
these
are
all
basically
dimensionality
reduction
methods
and
in
the
case
of
umap
and
t-sne,
they
build
these
really
beautiful
visualizations
of
the
data.
A
The
problem
is,
it's
very
hard
to
interpret
them,
especially
with
umap.
They
create
these
visualizations
that
are
very
hard
to
interpret.
So
I
was
thinking
of
doing
a
lecture
on
these
different
methods
and
kind
of
demystifying
them.
I've
been
doing
lectures
from
my
other
group
in
conjunction
with
neuromatch
academy.
So
then
I'm
getting
good
at
the
art
of
putting
together
a
half
hour
presentation
on
something
really
complex.
So
problem
is,
of
course
I
don't
really
have
any
spare
time
to
do
that,
but
I'm
planning
on
doing
that
soon.
A
And
then
that's
so
that's
all.
I
want
to
talk
about
about
the
board
again.
If
you
have
ideas
for
tasks,
I
think
you
can
add
to
this
board
or
if
you
want
to
email
me
something
that
we
need
to
put
on
the
board.
Let
me
know
again:
these
tasks
are
kind
of
at
this
level
of
granularity,
where
you
know,
if
we
think
about
it,
if
we
make
some
progress,
it'll
split
into
new
tasks,
but
it's
always
good
to
have
these
sort
of
interface
on
a
regular
basis,
so
that
we
remember
good.
A
A
A
Interested
the
pdfs
all
right,
so
the
first
one
is
this
paper,
art
and
science
at
the
sale,
holder,
miso
scale,
and
so
this
is,
I
think,
a
fairly
new
paper
in
trends
in
biochemical
sciences
and
the
thing
about
this.
These
trends
papers
is
that
they
have
these
glossaries
on
the
side,
so
they
define
different
things
in
the
paper
that
they
mentioned.
A
So
that's
always
good
if
you're
looking
for
likes,
if
you're
getting
introduced
to
an
area-
and
you
want
to
know
more
about
it
so
so
this
paper
is
a
review.
The
abstract
is
experimental.
Information
from
microscopy
structural
biology
and
bioinformatics
may
be
integrated
to
build
structural
models
of
entire
cells
in
molecular
detail.
A
A
Last
week
we
talked
about
open
worm
being
this
hyper
realistic
type
of
model,
and
that's
kind
of
the
same
thing
here
is
what
they're
focusing
on.
So
you
know
you
don't
have
to
build
a
model,
that's
hyper
realistic,
but
a
lot
of
the
cellular
models
are
especially
for
like
biochemistry
so
they
have
these
closely
packed
and
heterogeneous
components.
A
The
second
issue
is
that
the
wealth
of
available
experimental
data
is
scattered
among
multiple
resources
and
must
be
gathered.
Reconciling
and
curated
so
they're
talking
about
different
data
sets
that
you're,
integrating
and
bringing
together
to
do
these
experiments.
I
think
you've
seen
with
the
the
data
set
that
mayor
is
working
with
that
there
is
this.
A
A
We
know
this
from
like
the
deep
learning
venue
where
we're
doing
this
work
and
it's
taking
a
lot
of
computational
power
and
time,
but
you
get
these
models
that
are
that
tell
you
you
know,
they're
quite
informative,
so
we
present
recent
efforts
to
address
these
challenges,
both
with
artistic
approaches
to
depicting
the
cellular,
mesoscale
and
development
and
application
of
methods
to
build
quantitative
models
so
they've
they
had
an
article,
I
think
in
1991.
A
They
were
able
to
they
had
these
drawings
of
you
know
what
the
cellular
mesoscale
looked
like,
so
they
picked
this
objective
measure
this
this
level
of
measurement
and
they
kind
of
they
visualized
it
back
then.
Nowadays,
of
course,
you
can
do
this
computationally,
and
so
this
is
what
it
looks
like
today.
A
So
they've
gone
from
this
picture
to
this
model
and
they
define
the
cellular
lumiso
scale
as
scale
level.
Bridging
the
nanometer
scale
of
atomic
structure
in
the
micrometer
scale
of
cellular
ultrastructure,
so
they're
talking
about
like
basically
molecular
components
like
genes.
You
know
dna
rna
things
like
that
and
then
moving
up
to
this
micrometer
scale
of
the
cell,
so
features
on
the
surface
of
the
cell.
A
You
know
like
receptors
and
other
components
and
as
you
can
see,
the
cell
is
not
just
this
round
sphere.
It's
this
very
complex
thing,
and
this
is
a
cross
section
of
the
cell.
A
A
Image
a
lot
of
these
samples
that
they're
getting
they're
also
using
game
engines,
so
game
engines
are
like
software
and
hardware
that
allow
you
to
do
so,
a
lot
of
the
simulations
that
they
do
in
3d.
A
lot
of
these
types
of
simulations
are
done
on
game
engines
and,
of
course,
we
know
game
engines
are
used
by
video
games,
but
they're
optimized
for
graphical
processing.
A
So
that's
why
they
use
them
and
of
course
that's
where
gpus
come
from.
If
you
know
gpus
and
videos,
you
know
creating
hardware
for
video
games,
and
so
this
is
kind
of
where
this
comes
from.
A
A
So
this
is
where
the
different
data
sets
are
coming
from.
Quinary
structure
interaction,
often
transient
molecules
into
higher
order
assemblies
added
to
the
traditional
hierarchy
of
primary
structure:
sequence
to
quaternary
structure,
oligomerization
of
subunits.
So
this
is
where
you're
building
from
like
just
sequences
of
molecules
to
structures
of
proteins.
So
protein
structures
are
actually
pretty
high
dimensional
structures.
D
They
use
plugins
in
unity
to
simulate
like
spiders
and
other
crawlers,
and
try
to
make
them
walk
and
stuff
like
there
are
these
famous
deep
mind
right.
A
Oh
yeah
deep
mind
yeah
yeah,
they
do
a
lot
of
yeah.
They
use
a
lot
of
that
sort
of
stuff
too,
but
like
protein,
folding
actually
occurs
on
a
time
scale,
that's
pretty
short,
but
it
requires
an
immense
amount
of
computational
power
because,
what's
going
on
during
that
short
time,
scale
is
a
lot
of
like
conformational,
folding
and
stuff.
So
you
have
this:
just
an
incredible
amount
of
detail
needs
to
be
simulated.
A
Finally,
there
they
mentioned
virtual
and
augmented
reality,
and
so
why
do
they
mention
this?
This
is
because
you
can
use
these
technologies
to
visualize
a
lot
of
these
models
and
do
things
like
walk
through
the
models
and
view
them
at
different
angles
and
pick
them
up
and
rotate
them.
This
has
been
something
that
people
have
been
doing
for,
like
maybe
20
years,
and
you
know
I
think
in
like
some
like
gene
sequence
and
protein
folding
simulations.
A
Actually,
I've
used
vr
to
a
a
good
extent,
to
sort
of
visualize
a
lot
of
the
higher
order
structure.
You
know
because,
as
I
said
in
proteins,
you
get
a
lot
of
folding
and
that
helps
you
to
sort
of
you
know.
Sort
of
visualize
what
what
the
problems
are,
what
needs
to
be
solved
so
they
talk
about
the
challenges
of
the
meso
scale.
A
A
Mesoscale
environments
are
highly
crowded,
often
with
20
to
30
percent
of
the
space
filled
with
macular
macromolecular
components,
they're
highly
heterogeneous
and
have
flexible
functional
modes.
So
they
talk
about
some
data.
You
know
some
sources
of
data
and
they
talk
about
the
state
of
modeling,
which
is
pretty
exploratory,
so
they
don't
really
offer
a
lot
of
solutions.
A
They
just
kind
of
talk
about
it
as
a
sort
of
a
frontier,
and
then
they
talk
about.
How
do
we
expect
progress
to
unfold?
A
We
also
expect
that
progress
will
be
driven
by
new
ways
of
seeing
both
in
the
experiments
that
probe
the
cellular
meso
scale
and
in
the
visualization
software
we
build
to
explore
integrative
models
in
the
early
days
of
macular
molecular
structure,
creative
innovations
and
visualizations
such
as
linus
pauling's
base
filling
representations,
engine,
richardson's
urban
diagrams
revolutionize.
The
way
we
think
about
biomolecules
crystallizing,
new
modes
of
understanding.
A
So
they
expect
there
to
be.
You
know
this,
this
type
of
modeling
to
be
driven
by
similar
types
of
innovations,
and,
of
course
these
are
not
really
tech
like
hardware
innovations.
A
These
are
representational
and
sort
of
visual,
so
you
have
diagram
systems,
for
example
those
those
are
types
of
things
that
they
envision
to
be
driving
progress,
and
so
I
think,
that's,
I
think,
that's
true
for
a
lot
of
like
types
of
modeling,
where
you
know
we
have
a
lot
of
data,
but
the
harder
part
is
like
integrating
it
into
a
framework,
and
that's,
I
think,
where
you're
getting
there's
a
lot
of
opportunity
there.
A
So
this
is
eso
skill
illustrations
as
thinking
tools.
So
this
is
an
illustration
of
the
cytoplasm
to
vacuole
targeting
process
autophagy
on
the
left.
A
A
So
there's
you
know
a
lot
of
opportunities
for
like
artists
and
scientists
to
get
together
and
do
things
that
are
you
know
to
sort
of
build
these
models,
these
sort
of
visual
models
that
aren't
necessarily
scientific,
highly
scientifically
accurate.
I
mean
they're
accurate
to
sort
of
the
visual
sense,
but
you
know
they
don't
have
like
they're,
not
quantitative
models,
but
they
give
you
idea.
You
know
ways
to
think
about
the
system,
so
you
know
to
be
able
to
see
it
like
this.
A
You
can
think
about
it
in
a
different
way
than
if
you
were
just
like
reading
numbers
or
if
you
were
just
talking
in
words
and
so
yeah.
So
this
is
just
kind
of
goes
on
about
this.
If
you
want
to
look
at
some
good
nice
images
both
hand
drawn
and
microscopy
produced,
then
this
is
a
nice
paper
to
look
at.
A
Yeah
they
give
some
outstanding
questions.
So
what
are
the
salient
features
and
properties
of
the
cellular
mesoscale
that
impact
the
process
of
life?
How
can
we
target
these
properties
and
pharmaceutical
interventions?
A
A
Okay,
so
then
the
next
paper
is,
I
think,
I'll
only
talk
about
two
today,
I
don't
know
if
I
want
to
talk
about
that
one.
Oh,
maybe
this
one
okay,
so
this
one
is
a
developmental
landscape
of
3d
cultural,
human
pre-gastrulation
embryos,
and
so
this
one
focuses
on
our
understanding
of
how
human
embryos
develop
before
gastrulation,
including
spatial
self-organization
cell
cell
type,
ontogeny,
remains
limited
by
available
two-dimensional
technological
platforms.
A
Here
we
report
a
three-dimensional,
blastocyst
culture
system,
so
what
they're
talking
about
is
like
cell
culture,
where
cell
culture
is
traditionally
two-dimensional,
you
put
cells
on
a
plate
and
you
plate
them
and
they
grow
to
you
know
in
populations,
and
then
you
can
assess
them
with
microscopy
and
it's
but
it's
two-dimensional.
A
But
of
course
they
have
three
dimensional
culture
systems
which
require
some
depth,
so
sometimes
they'll
use
like
scaffolds
to
grow.
The
cells
on
sometimes
they'll
just
grow
them
in
a
three-dimensional
chamber,
and
that
gives
you
a
little
bit
different
perspective
because
the
cells
are
allowed
to
move
around
in
a
three-dimensional
way,
instead
of
just
adhering
to
a
surface.
A
So
they
report
on
this
blessed
culture
system
that
enables
human
blastocyst
development
up
to
the
primitive
street
stage.
These
3d
embryos
mimic
developmental
landmarks
and
3d
architectures
in
vivo,
including
the
embryonic
disc
amnion
basement
membrane,
primarium,
primate,
unique,
secondary
yolk,
sac
formation
of
anterior,
posterior,
polarity
and
primitive
streak
using
cellular
transcriptome
profiling.
A
We
delineate
ontology
and
regulatory
networks,
so
I've
been
underlying
the
segregation
of
epiblast,
primitive
endoderm
and
propheblast.
So
they're,
looking
at
like
the
molecular
mechanisms
for
those
embryos
divide
into
these
three
germ
layers
and
they're.
Looking
at
the
molecular
markers
of
this
as
it
emerges,
so
you
can
do
this
in
three
dimensions.
A
You
can't
necessarily
do
this
in
two
dimensions,
because
you
don't
get
the
signals
that
you
need
from
the
three-dimensional
shape.
If
I
do
like,
if
I
put
a
bunch
of
cells
in
a
two-dimensional
culture,
I
won't
get
that
sort
of
it
won't
necessarily
get
that
sort
of
organization.
A
So
then
they
do
some
implantation,
specific
pathways
and
transcription
factors
trigger
the
differentiation
of
say,
cytotrophoblasts,
extra,
villus,
cytotrophoblasts
and
centitiotrophoblasts,
and
then
they
talked
about
this.
That
epi
blast
undergo
a
transition
of
pluripotency
upon
implantation,
and
then
they
look
at
the
transcriptomes
of
those
cells
and
together.
Findings
from
our
3d
culture
approach.
A
Help
us
determine
the
molecular
and
morphogenetic
developmental
landscape
that
occurs
during
human
embryogenesis,
so
they're
looking
at
human
embryos
here
and
they're
building
these
three-dimensional
cultures
and
so
they're
allowing
them
to
form
these
massive
cells
that
are
three-dimensional
and
they're.
Actually
looking
at
gene
expression
in
those
three-dimensional
cultures,
and
so
then
they
have.
Let's
see.
A
So
this
is
the
transcriptomic
data,
so
these
are
the
gene
expression
data
that
they
use
for
these
different
genes,
so
they've
assayed
these
different
genes
for
these
different
conditions
and
they
have
a
heat.
This
is
the
general
way
they
assess
this
using
heat
maps.
A
So
you
get
like
you
get
different,
they
get
fold
expression
differences
is
how
you're
measuring
it,
so
they
usually
use
some
standard
like
gap
ph
and
you
measure
the
activity
of
these
different
genes
based
on
the
housekeeping
gene
is
what
they
call
the
standard,
and
then
you
have
upregulation
or
downregulation,
and
so
this
is.
This
is
actually
our
scrnac.
A
Having
trouble
moving
this
down
there,
we
go
so
they
show
some
of
the
structure
that
you
get
in
the
three-dimensional
cultures.
You
get
this
anterior
posterior
polarity
in
formation
of
psa.
A
You
get
these
different,
germ
layers
that
form
in
these
sculptures.
A
A
It
says
there
is
some
literature
and
forcing
embryos
to
develop
in
2d
I
mean
yeah.
There's
do
you
have
a
good
sense
of
what
that
literature
looks
like,
or
is
that
just
like
couple
of
papers
here
and
there.
A
So
yeah
it's
yeah
it's
well!
I
think
people
are,
you
know
trying
to
explore
3d
cultures
and
now
it's
I
know,
there's
been
this
work
with
embryoid
bodies
that
people
are
really
excited
about,
where
they're
actually
trying
to
get
like
grow
like
small,
neural
networks,
out
of
a
bunch
of
stem
cells,
so
they're
just
growing
these
embryoid
bodies
and
they
get
them
to
differentiate
into
neuronal
cells.
A
And
you
know
those
have
been
having
they've
had
various
success
with
that.
You
know
they
get
these
little
balls
of
neurons,
basically
that
have
electrical
activity,
but
beyond
that
they
haven't
really
gotten
any
sort
of
like
brain
out
of
it.
But
I
think
that's
that's
interesting,
see
what
people
are
doing
with
cell
culture.
A
Okay,
everyone!
Alright,
thanks
for
meeting,
see
you
later
dick
everyone
else
have
a
good
week.
Talk
to
you
on
slack.