►
From YouTube: DevoWorm (2022, Meeting 2): Life to the Stars, 1-D Ising models and Mollusk shells, Stochastic Bio.
Description
Life to the Stars (Tardigrades and Nematodes at 10% the speed of light). Final draft of our Google Summer of Code 2022 projects. 1-D Ising models, Wolfram patterns, and Mollusk shells. Neural control of the pattern forming secretome, Stochastic biology (Stochastic vs. Deterministic models), OpenWorm Browser in Oculus Quest (Github Issue #113). Attendees: Tom Portegys, Richard Gordon, Karan Lohaan, Valentina Perricone, and Bradly Alicea.
C
B
D
B
B
E
F
B
E
B
B
B
Yeah
yeah
I'd
be
happy
too
yeah,
okay,
so,
let's
so
coming
up,
we
have
a
couple
things.
First
of
all,
we
have
the
google
summer
of
code
program
and
last
week
I
talked
about
some
of
the
projects
that
we'd
be
doing
this
year,
and
so
let
me
share
my
screen.
This
is
the
thing
I
submitted
to
the
incf,
so
we
had
two.
B
I
think
I
mentioned
this
before
the
graph
neural
network,
says
developmental
networks
and
then
the
digital
microsphere,
which
is
the
stuff
that
susan's
working
on
with
axolotl
data.
So
the
first
one,
of
course,
is
the
graph
neural
networks
as
developmental
networks.
I
don't
know
if
I
had
written
the
whole
thing
up
by
last
week.
I
think
I
may
have
but
again
to
mention
this
to
people
who
are
here,
so
you
know
we're
looking
at
different
types
of
networks
and
biology
and
biological
development.
B
So
there
are
things
like
neural,
connectomes,
gene
regulatory
networks,
interactive
networks
and
anatomical
networks,
and
the
idea
is
we
have
these
cell
tracking
data
that
we've
been
working
with:
we've
also
created
cell
tracking
data
by
segmenting
cells,
out
of
images
and
then
creating
databases
with
that.
We've
done
that
with
deep
learning,
different
deep
learning
techniques.
The
last
several
summers:
we've
done
a
number
of
different,
deep
learning,
algorithms,
and
we
have
the
diva
learn
package
now,
which
is
now
there's
a
preprint
on
it,
which
is
here,
and
it
describes
the
platform
and
it's
on
pipey.
B
But
the
idea
here
is
to
extend
that
that
library
of
things
so
we
have
things
in
evil,
learn
and
we
have
like
these
different
deep
learning
techniques,
and
then
we
also
have
this
diva
worm
ai,
which
is
a
platform
where
we
have
different
tools
that
you
can
use
to
extract
data
out
of
a
out
of
an
embryo.
Basically,
we
haven't
benchmarked
that
on
anything
more
than
embryos,
but
we've
looked
at
mainly
c
elegans
embryos
and
there
are
a
couple
other
types
of
embryos.
D
B
All
sort
of
you
know
those
platforms
would
help
us
do
those
things.
The
graph
neural
network's
idea
is
to
extend
that
to
sort
of
creating
these
graph
embeddings,
which
actually
are
representations
of
some
of
these
networks,
so
we're
extracting
data
from
images
or
we're
using
cell
tracking
data,
which
is
where
they
put
a
die
in.
B
You
know
in
the
cell
nucleus,
and
then
they
track
it
with
computer
vision.
So
you
know
this
can
be
done
either
as
you're
collecting
the
data
or
in
the
form
of
second-hand
data.
But
in
any
case
we
can
use
these
data
to
create
these
graph
embeddings,
and
then
we
can
actually
use
those
as
sort
of
like
templates
for
different
graphs.
B
So
if
we
want
to
describe,
for
example,
the
c
elegans
connectome
as
it's
developing,
we
can,
you
know,
use
these
the
first
class
of
models,
the
pre-trained,
deep
learning
models,
which
you
know,
those
work
by
just
taking
some
data
and
developing
the
algorithm
so
that
they're
sort
of
well
adapted
to
that
data.
And
then
you,
you
know
you
use
that
you
don't
have
to
use
a
training
set
anymore.
You
have
this,
it's
it's!
B
The
training
data
is
embedded
in
the
structure
of
the
algorithm,
so
you
just
take
the
algorithm
and
you
use
it.
You
don't
have
to
train
it,
necessarily
that's
limiting
in
a
number
of
ways,
because,
as
you
can
imagine,
you
know
if
you
have
a
lot
of
auto
distribution
data
or
things
like
that,
it
makes
it
hard
to
adapt
the
model.
So
we
have
a
pre-trained
deep
learning
model
for
c
elegans.
B
With
graft
neural
networks,
you
basically
characterize
the
network
embedding
and
then
you
use
these
neural
networks
to
analyze
the
data
and
it's
just
another
way
to
like
extract
data
and
make
some
you
know
basically
say
something
about
the
biological
system.
So
this
is
a
newer
technique.
Craft
neural
networks.
We've
talked
about
this
in
meetings
before
and
I'm
just
looking
for
people
to
work
on
this
library.
B
B
You
know
like
nine
or
so
different
perspectives
of
the
embryo,
so
you
know
you're
looking
at
it
from
the
bottom
from
the
top
and
the
sides
from
diagonally,
and
then
you
have
all
a
series
of
images
from
each
of
those
perspectives.
So
then,
what
you
can
do
is
you
can
stitch
those
together
into
these
spherical
representations.
B
So
I
have
you
know
nine
images
from
different
perspectives
and
they
get
like
probably
most
of
the
surface
of
the
embryo,
and
so
then
what
you
can
do
is
you
can
take
each
of
those
perspective,
views
tile
them
onto
a
sphere
and
then
you
know
use
different
warping
algorithms
to
fit
them
onto
the
surface
of
the
sphere.
We
had
some
people
working
on
this
last
year,
but
no
one
got
selected
for
the
summer
code
project
and
I
mean
you
know
we
have
a
couple
people.
H
B
Of
warping
and
implementing
different
techniques
for
building
maps,
so
you
know
this
is
like
the
surface
of
the
earth.
It's
like
you
have.
You
know
we
have
map
flat
maps
that
are
basically
taken
from
things
that
are
on
the
surface
of
the
earth.
So
you
know,
a
map
of
the
world
is
basically
taking
the
surface
of
a
sphere
pulling
it
off,
and
then
it's
flattening
it
out
and
using
a
projection
to
approximate
some
of
the
gaps
that
you're
going
to
find,
and
so
this
is
the
reverse
of
that.
B
This
is
where
you're,
taking
a
flat
image,
that's
captured
as
flat
image,
I
should
say,
and
then
you're
titling
a
sphere,
and
so
that's
that's
most
of
the
technical
work
and
then
susan
will
provide
a
lot
of
the
data.
She's
got
more
data
coming,
we
have
probably
the
present.
I
think
we
have
several
gigabytes
of
data,
but
we
would
like
to
have
many
gigabytes
of
data,
probably
not
terabytes,
but
I
think
for
a
summer
code
project.
B
So
people
are
interested
in
applying
up
to
that.
That's
that's!
What's
going
to
be
up
on
the
incf
website
soon,
and
so
the
the
project
schedule
is
that
we'll
have
these
applications
up
for
a
while
people
will
be
able
to
apply
to
them
and
then
we'll
have
the
application
process.
Where
people
you
know
come
up
with
a
solution.
B
They
can
attend
the
meetings
here,
get
feedback
and
then
I
think
in
I
don't
know
when
when
they
make
when
well,
the
application
period
formally
opens
like
in
march
or
something
and
then
they
have
a
month
to
put
their
applications
in
and
we
evaluate
them,
then
sometime
in
may,
the
selections
are
made.
So
that's
always
fun
because
people,
people.
B
B
So
that
I
I
you
know,
I
think,
that's
probably
enough
projects
for
one
year
last
year
we
had
three
projects
and
we
only
got
one
project
selected
so
so
that
was,
you
know
that
so
that's
coming
up.
So
the
next
thing
I'd
like
to
talk
about
is,
let's
see
so
dick
sent
me
this
article,
and
I
thought
it
was
interesting.
B
This
is
earth
sky.
This
is
a
popular
magazine
and
the
question
is
here:
can
we
and
should
we
send
life
to
the
nearest
star?
This
is
a
nice
speculative
piece.
Dick
has
an
interest
in
astrobiology
and
I'm
interested
in
it
too.
You
know
I've
been
had
a
long-standing
interest.
I
never
really
pursued
the
science
of
it,
but
we
have
this
article
and
this
is
a
tardigrade.
So
this
is
an.
B
Right
yeah,
so
this
is
what
they
call
a
tardigrade
and
tardigrades
are
very
resilient
organisms.
They
can
survive
the
vacuum
of
deep
space
apparently,
and
they
can.
You
know,
they've,
taken
them
up
on
the
space
shuttle
and
and
various
things,
so
they
they
know
they're
pretty
resilient,
and
the
idea
is
to
send
these
tardigrades
into
outer
space
and
things
and
they're
they're.
You
know
about
a
millimeter
long,
so
they're,
maybe
at
about
the
scale
of
c
elegans.
B
B
I
B
B
B
B
You
know
there
what
they
basically
do
to
to
revive
a
colony
of
c
elegans
is
they
have
the
eggs
they're
frozen
and
then
they
just
put
them
in
a
plate
with
some
food
and
they
end
up
hatching
and
then
starting
up.
You
can
freeze
c
elegans
pretty
easily.
You
just
have
to
basically
extract
the
eggs
from
a
population
of
worms,
and
then
you
can
freeze
the
eggs
in
a
medium
and
the
eggs
will
be.
You
know
remain
intact
so
that
when
you
unfreeze
it,
you
put
it
on
medium.
They
just
start
to.
B
Going
through
their
life
cycle,
so
c
elegans
is
another
candidate.
It's
I
think
it's
kind
of
interesting
some
of
these
organisms.
You
know,
maybe
it's
a
nice
sort
of
thought,
experiment
to
think
of
all
the
organisms
that
you
could
say
put
in
that
class
of
organisms
that
you
want
to
send
to
space.
What
makes
them
sort
of
candidates
for
that,
and
so
I
would
imagine
like
one
of
the
one
of
the
criterion
would
be.
B
B
B
B
B
Know
if
we'll
still
be
around
as
a
species,
but
it's
not
likely
that
anyone
will
remember
it.
So
that's,
and
so
now
you
have
to
create
a
ship
that
get
there.
You
know
in
a
reasonable
amount
of
time,
like
within
a
century
maybe
and
so
they're
having
to
you
know
any
ship
that
would
do
that
would
have
to
travel
at
a
pretty
high
rate
of
speed
so
20
to
30
percent
the
speed
of
light,
maybe
and
then
this
kind
of
goes
through
some
of
the
things.
E
B
A
B
B
B
B
B
E
K
K
K
B
K
D
G
B
K
K
K
D
K
To
a
larger
neighborhood
with
more
rules-
and
the
question
is:
are
those
rules
redundant
or
do
they
hand,
write
new
patterns,
an
ssg
now?
The
reason
that's
important
is,
if
it's
closed
with
respect
to
the
original
wall.
Purples
notice,
the
middle
from
the
original
ovals
generate
all
the
patterns.
You
can
get
then
walter's
new
kind
of
science,
which
is
the
name
of
this
book
he's
doing
this
stuff
would
have
some
basis
possible
basis.
D
D
K
B
Okay,
so
this
is
an
example
of
a
wendy
automata.
I
drew
this
on
my
board
here.
So
let's
go
over
this,
so
this
is
a
one
dimensional
automata.
So
the
one
dimension
is
this
line
here
and,
as
you
can
see
that
you
have
several
cells
that
are
sort
of
next
to
one
another.
There
isn't
a
second
dimension
where
you
would
have
rules
in
the
up
or
down.
B
It's
just
right
to
left,
so
you
can
see
that
this
automata
has
two
states
on
and
off.
The
states
are
then
determined
by
these
neighbors,
so
this
cell
here
has
two
neighbors
this
one
and
this
one.
This
is
a
neighborhood
of
order
one.
So
there
are
three
cells
in
this
neighborhood
and
the
focal
cell
is
here.
The
neighbors
are
here
and
then
the
rules
basically
define
how
this
cell
is
turned
on
and
off.
B
So
if
this
cell
is
on
this
cell
is
on,
and
if
this
cell
is
off,
it's
still
on,
because
the
rule
here
is
that
if
either
cell
is
on,
then
this
cell
is
on,
and
so
each
cell
can
have
different
rules,
and
so
we'll
be
talking
in
this
meeting
about
how
you
know.
Does
the
automata
do
things
beyond
the
rules,
if
you
model
say
like
a
biological
system
or
does
a
biological
system,
if
you
model
it
as
a
1d,
automata
behave
well
within
the
scope
of
all
possible
rules
that
are
applied
to
it.
K
K
K
C
K
Yes,
you'll
find
patterns,
all
sorts
of
patterns
and
snail
shells
and
the
question
is:
can
they
all
be
subsumed
under
welfare
rules
or
are
vulture
rules
not
closed?
Okay?
So
I
think
we
can
ask
a
very
fundamental
question
about
the
wolfram
pattern.
Wolfram
would
like
to
claim
the
basis
of
all
physics.
K
K
K
K
L
K
K
E
B
B
B
K
K
J
E
K
K
E
K
E
E
D
B
B
K
B
B
B
K
B
B
So
I
took
out
some
figures,
so
what
they
argue
in
this
paper
is
that
this
these
patterns
are
based
on
a
neural
circuit,
and
so
they
show
these
like
stripe,
stripings
and
there's.
They
actually
do
talk
about
how
there's
a
secretion
mechanism
for
creating
some
of
these
patterns
during
development.
So
there's
this
the
anatomy
of
the
mantle
region,
you
have
mantle.
B
Yeah
and
then
then
they
have
in
the
paper
here.
They
have
some
assumptions
that
under
why
their
model
so
they're
actually
building.
I
don't
know
if
it's
a
computational
model,
but
it's
a
a
model
that
tries
to
describe
this
process
so
that
you
know
they
talk
about
the
mantle
and
how
there's
secretions
that
are
deposited
on
the
shell
surface
and
then
they
kind
of
go
through
the
this
model,
and
then
they
also
talk
about
linking
this
to
the
the
nervous
system
of
the
mollusk,
and
so
this
is
figure
nine.
B
So
I
think
it's
this
figure.
This
is
a
diagram
of
the
model
where
they
have
a
mantle
pigment
cell,
which
is
mpc,
which
are
these
here.
These
squares,
these
yeah
rectangular
cells,
pcn,
which
is
a
pigment
cell
neuron,
which
is
this
triangle
down
here,
and
so
the
pcn
synapse
is
on
the
mc
mpc
and
then
you
get
the
mnn,
which
is
the
mantle
neural
net.
So
that's
this
underneath
so
there's
this
neural
network
underneath
it's
just
in
the
case
of
a
biological
neural
network.
B
It's
you
know,
connectivity
between
these
cells
and
it's
coming
down
to
this
cg
or
central
ganglion.
So
a
ganglion
is
just
this
cell
that
has
a
lot
a
lot
of
reticulating
connections
and
or
or
a
center
of
cells.
It
has
a
reticulating
set
of
reticulating
connections
and
they're
usually
used
as
like,
sometimes
central
pattern,
generators
or
other
things
in
a
a
lot
of
invertebrates.
B
K
B
Is
the
their
basic
model
when
they're,
just
basically
saying
that
you
know
it's?
It's
the
nervous
system
is
made
for
this
sort
of,
I
don't
know,
call
it
secretion
management,
but
you
know
when
you
had
in
the
visual
field,
you
have
a
lot
of
cells
in
the
retina
that
are
responding
to
some
stimulus
in
the
environment
and
it
has
to
detect
patterns.
B
H
B
K
Any
pattern
at
all:
if
we
build
the
device
for
rotating
the
shell
and
just
accept,
we
take
a
picture
for
each
angle,
but
then
we
only
need
to
cut
the
pixels
that
are
in
front.
We
can
then
unroll
the
shell
okay
turn
it
into
a
flat
pattern,
it's
very
difficult
to
deal
with
a
real
shell,
because
it's
very
dimensional.
A
K
A
K
C
K
One
of
the
things
I
did
in
the
1970s,
while
I
was
still
in
a
man
running
around
the
world,
I
go
to
style
shops
and
buy
shelves
that
were
defective,
and
these
were
usually
shells
that
had
some
trauma
and
they
had
to
regrow
their
shell,
okay
and
then
the
pattern
would
resume
after
a
little
while.
So
this
gives
us
a
fresh
start
to
the
pattern
each
time
reaching
for
each
of
these
shelves.
B
B
So
here
they
have
this,
where
they're
reconstructing
a
pattern
of
a
snail
shell
but
they're
doing
it
with
characters
on
this
array.
So
this
is
a
very.
This
is
1986,
so
they
didn't
do
like
a
modern
analysis
of
this
or
even
a
modern
rendering
of
it.
But
this
was
basically
they
were
reproducing
the
pattern
using.
I
guess
ascii
2
characters,
so
you
can
see
the
pattern
here.
B
K
B
E
E
K
C
C
K
K
K
K
C
E
No,
I
guess
in
world
war
ii
the
you
know
some
south
pacific
islands,
the
air
force
landed
and
created
bases
and
stuff
like
that,
and
they
passed
out
all
these
goodies
to
the
natives
right
and
then
when
they
left,
they
figured
like.
Well,
let's,
let's
build
something
that
looks
like
an
airplane
and
then
they'll
bring
us
some
good
stuff.
B
B
So
we've
talked
a
little
bit
about
stochasticity
and
it's
basically,
where
you
have
a
model,
that's
probabilistic
and
that
is
is
is
quasi-random
and
that
is
it
takes.
It
basically
generates
states
at
random
probabilities
at
random
instead
of
a
deterministic
system,
where
you
know
what
the
outcome
is
going
to
be
in
advance.
B
So
these
are
just
you
know,
an
exam
one
small
example,
but
in
biological
systems,
sometimes
cell
differentiation
works.
This
way
in
in
c
elegans,
for
example,
we
have
deterministic
cell
differentiation,
a
cell
from
a
certain
lineage
will
become
a
certain
type
of
cell.
There
isn't
much
ambiguity
about
that,
maybe
except
in
mutant
genotypes,
but
still
a
stochastic
differentiation
scheme
is
where
a
cell
will
be
a
stem
cell
and
it
can
differentiate
into
a
number
of
fates
and
it
depends
on
the
environmental
factors.
B
B
So
in
this
paper
they
talk
about
this
in
the
context
of
development
during
development
cells
need
to
make
decisions
about
their
fate
in
order
to
ensure
that
the
correct
numbers
and
types
of
cells
are
established
at
the
correct
time
and
place
in
the
embryo.
Such
selfie
decisions
are
often
classified
as
deterministic
or
stochastic.
B
So,
if
you
think
about
this
statement,
ensuring
that
cells
are
established
at
the
correct
time
and
place
in
the
embryo
that
you
know
that
might
suggest
that
this
is
a
deterministic
process.
It's
like
the
clock
maker
model
that
people
often
adhere
to
in
the
enlightenment,
and
maybe
a
little
bit
later
in
science,
the
history
of
science,
but
actually
a
lot
of
cells
are
stochastic
and
a
lot
of
differentiation
is
stochastic
and
surprisingly,
or
not,
they
meet
the
it
tends
to
meet
this
criterion.
B
You
know,
even
when
you
have
a
lot
of
stochastic
differentiation,
embryos
are
not
chaotic
places.
Necessarily
there
is
quite
a
bit
of
order,
and
so
this
is
an
interesting
kind
of
you
know.
Maybe
paradox,
maybe
not
because
if
you
know
anything
about
chaotic
systems,
you
know
that
chaotic
things
can
exist
on
the
edge
of
chaos
where
you
have
this
a
lot
of
stochasticity,
but
you
also
have
a
lot
of
order
as
well.
B
So
we
discussed
these
frameworks
within
which
such
clear
definitions
make
sense
and
highlight
when
certain
ambiguity
prevails.
As
an
example,
we
examine
how
these
terms
are
used
in
studies
of
neuronal,
selfie
decisions
and
point
out
areas
in
which
definitions
and
interpretations
have
changed
to
matured
over
time.
B
We
hope
that
this
review
will
provide
some
clarification
and
inspired
discussion
on
the
use
of
terminology,
so
this
is
largely
trying
to
clean
up
the
terminology
in
the
biological
literature
and
link
it
more
to
the
mathematics,
and
so
we
know
that,
like
these,
two
different
types
of
developmental
processes
exist,
deterministic
and
stochastic,
and
so
we
can
look
at
this
experimentally
empirically
they
address
these
issues
in
this
review.
They
give
a
mathematical
definition
of
stochasticity
and
then
discuss
the
potential
origins
of
stochasticity
and
cellular
developmental
systems.
B
B
Is
this
mathematical
model
where
you
have
a
hypothetical
experiment,
e
or
observation
e,
with
an
outcome
x,
and
so
the
outcome
x
is
maybe
the
number
of
cells
in
a
certain
type
that
could
be
quantified
as
a
particular
developmental
stage.
So
in
other
words,
if
you
have
an
experiment
and
you
have
a
manipulation,
there's
an
outcome
x,
and
so
in
some
cases
you
might
have
some
observation,
where
the
number
of
cells
of
a
certain
type
in
a
certain
developmental
stage
varies
between
observation.
B
Although
the
underlying
process
is
basically
the
same,
so
you
know
we
think
of
experiments
if
you
control
them
enough,
they're
they're
reproducible,
but
in
a
stochastic
system,
sometimes
they're,
not
so
reproducible.
Sometimes
you
get
fluctuations
and
of
course
we
know
this
from
the
standard
deviation
of
an
experimental
measurement,
but
sometimes
this
is
more
fundamental.
This
is
actually
the
way
the
system
works,
and
so
you
can
see
where
this
might
be
a
problem
if
you're
a
hardcore,
experimentalist
and
you're
trying
to
get
a
handle
on
the
mechanism
that
this
fluctuation
is
your
enemy.
B
B
You
have
to
make
assumptions
about
the
probabilities
in
advance
you
can
make
you
know
you
can
turn
this
into
a
stochastic
or
a
randomly
generated
variable,
but
you
have
to
know
that
this
is
the
way
a
system
behaves
in
advance,
so
formally
x
can
be
described
as
a
random
variable
which
assigns
a
number
to
each
element.
In
the
event,
space,
a
random
or
stochastic
process
is
a
sequence
of
random
variables
that
can
be
used
to
describe
time-dependent
stochastic
phenomena
in
mathematics.
The
terms,
random
and
stochastic
are
often
used
interchangeably.
B
D
B
B
A
random
distribution,
and
so
this
random
distribution
is
where
you
assign
a
probability
px
each
possible
value
of
x.
So
your
probability
distribution,
you
know,
has
tails
and
it
has
a
center
of
mass
and
each
each
point
on
x,
which
is
deviation
from
the
mean
which
is
the
center
of
the
mass
of
the
distribution,
and
this
is
for
a
a
random
distribution.
B
B
For
random
distributions,
which
are
normal,
but
sometimes
you
have
skewed
distributions,
sometimes
you
have
other
types
of
distributions,
exponential
distributions
and
other
things
like
that,
and
so
it's
harder
to
make
those
predictions.
You
can
use
the
different
distributions
as
models,
but
you
know
that
you
have
to
make
that
assumption
in
advance
and
it
changes
the
result
in
the
underlying
mechanism.
B
B
So
this
is,
you
know.
This
is
one
of
the
things
you
have
to
think
about
when
you
model
things
in
a
stochastic
sense.
So
so
they
make
the
point.
Your
casinos
heavily
rely
on
the
fact
that
although
individual
games
are
decided
by
chance,
the
odds
are
such
that
gamblers
almost
certainly
lose
in
the
long
term.
B
So
the
degree
of
randomness
means
that
you
can't
really
predict
things
in
advance.
You
can
predict
things
occasionally,
but
you
know
it's
not
something
that
we
when
we
do
a
biological
experiment,
for
example,
it
reaches
havoc
on
our
on
our
model,
because
we
can't
make
the
kinds
of
predictions
we're
used
to
making
which
are.
You
know
you
manipulate
something.
You
hit
a
certain
outcome
and
then
it's
repeatable
in
a
system
like
this.
It
isn't
necessarily
repeatable
and
you
have
to
account
for
that
in
your
model
system.
B
So
there
are
things
like
hidden
variables
as
well,
so
hidden
variables
are
variables
that
you
don't
identify
a
priori
that,
but
nevertheless
have
an
effect
on
the
system
on
the
outcome
and
so
they're
models
for
uncovering
hidden
variables,
hidden
markov
models,
but
these
are
things
that
have
to
be
again
put
into
place
in
advance,
so
in
sulfate
choices,
for
example,
these
types
of
activities
integrate
a
multitude
of
chemical
and
mechanical
factors
which
often
vary
with
time
and
space.
So
a
lot
of
these
factors
and
we've
talked
about
them
in
previous
meetings.
B
You
know
we
have
to
know
what
they
all
are
and,
of
course,
if
we
don't
know
what
they
all
are
or
don't
know
or
don't
know
their
full
effect
on
the
system,
then
we
can't
really
say
with
any
certainty
what
their
you
know,
what
their
effect
is.
So
you
know
we
know
that,
for
example,
there's
certain
chemical
factors
or
certain
mechanical
factors
which
predominate
in
shaping
an
outcome.
B
So
you
know
we
have
to
work
with
what
we
have
and
if
we
think
of
the
history
of
science,
the
history
of
science
has
been
where
you
start
working
on
a
problem,
and
you
only
know
maybe
one
or
two
causal
factors
or
potential
causal
factors
and
then,
as
as
time
goes
on,
you
get
other
things
that
come
into
play
that
explain
some
of
the
variance
in
your
result.
So
you
know
you
might
have
something
that
maybe
explains
one
percent
of
the
variance
of
what
you're
observing
of
the
outcome.
B
But
say
you
have
things
you
know
when
you
start
working
out,
probably
have
things
that
account
for
10
or
20
or
30
of
the
outcome.
So
you
know
this
this
process
of
sort
of
finding
all
the
causal
factor.
Potential
causal
factors
is,
you
know
it's
a
it's
an
interesting
one
to
think
about.
In
light
of
like
reproducibility,
and
indeed
like
our
understanding
of
differentiation,
isn't
it
as
a
phenomenon.
So
this
is
an
example
here,
where
you
have
a
probability
distribution
based
on
certain
things.
You
know
certain
potential
causal
factors
that
exist.
B
So
if
we
just
look
at
this
outcome
x,
we
see
that
there's
this
normal
distribution
of
results,
the
value
of
x,
is
normally
distributed.
So
that
means
that
in
the
center
here,
at
the
mean,
that's
when
most
you
know
most
of
your
results
when
you
independent
observations
will
be
in
this
category
of
the
mean,
but
some
of
the
observations
will
be
on
the
tails
and
so
they're
lower
probability
events,
but
they
still
occur
now.
B
B
Distribution-
it's
it's
maybe
skew
normal
where
you
have
this
tail,
these
tails,
but
the
tails
are
uneven
and
the
probability
is
maybe
skewed
towards
the
left
here.
So
excuse
the
you
know
once
we
know
that
f1
is
involved
and
we
manipulate
f1.
We
revise
our
distribution
and
we
say
it's
a
skewed
normal
and
now
you
know
this
is
a
this.
Is
a
bayesian
calculation
here
x
given
f1?
B
That's
the
probability
that
we're
looking
at
so
it's
a
conditional
probability
of
x,
but
given
that
f1
is
having
an
effect
on
it,
but
then
we
have
you
know,
so
we
don't
know
about
f2
and
f3,
and
although
we
can
see
in
this
hidden
model
that
f2
and
f3
have
effects-
and
sometimes
these
are
synergistic
effects
on
each
other.
So
now
you
know
we
don't
know
about
these
in
advance.
We
know
that
f1
is
having
an
effect,
but
f2
is
still
having
an
effect
on
x,
but
we're
not
measuring
that.
B
So
this
is
why
our
distribution
is
somewhat
skewed.
Now,
if
we
go
to
where
we
know
the
full
model,
where
we
have
f1,
f2
and
f3-
and
we
know
the
relationships
and
they're
all
observable,
then
we
get
a
different
outcome.
We
get
this
outcome
where
we
measure
the
relationships
here
and
then
now
x
is
a
different
outcome,
which
is
that
most
things
occur
on
the
mean
and
very
few
things
occur
on
the
tail,
and
so
this
is
an
idealized
model,
because
my
guess
is.
D
B
You
know
there's
a
lot
of
slop
in
these
measurements
here,
because
you
don't
know
these,
you
don't
know
what
these
factors
are
upstream
and
so
it's
very
hard
to
really
kind
of
get
a
good
measure
of
x.
But
this
is
an
idealized
model
and
it
just
kind
of
shows
that,
as
you
uncover
more
of
the
mechanism
you're-
and
you
manipulate
this-
the
effect
on
it-
the
effect
shown
in
embodied
by
x
changes
over
these
different
scenarios.
B
But
stochasticity
is
also
so
yeah.
It's
the
experimental
outcomes,
can
appear
more
or
less
random,
depending
on
how
much
is
known
about
the
system.
Mathematical
terms,
this
can
be
illustrated
using
conditional
probabilities,
so
they
use
conditional
probabilities
here
and
it's
just
a
matter
of
basically
weighting
x
by
the
things
that
are
affecting
it
and.
D
B
This
is
all
related
to
stochasticity,
because
for
one
thing
these
variables
may
be
stochastic.
For
another
thing,
you
know
you
can
see
how
you
you
know.
Sometimes
you
can
get
a
better
handle
on
the
system.
It
becomes
maybe
more
deterministic
in
the
hands
of
the
experimenter
or,
conversely,
it
because
could
become
less
probable
or
it
could
become
more
probabilistic
and
less
deterministic,
because
now
you
have
these
variables
which,
before
you
just
assumed,
were
not.
F
B
B
Well,
as
it
turns
out
that
developmental
cells
or
cells,
that
exhibit
meiosis
have
a
cell
cycle
and
that
cell
cycle
goes
in
sort
of
has
a
period
a
time
period
and
then,
at
the
end
of
that
time
period
the
cell
has
to
make
a
decision
as
to
whether
to
self-renew
and
go
through
another
cycle,
meiosis
or
differentiate
into
a
differentiated
cell,
and
so
this
process
of
going
through
the
cell
cycle
and
then
making
this
decision
and
then
going
through
self-renewal.
This
is
all
can
be
modeled
as
a
stochastic
model.
B
More
specifically,
it
could
be
modeled
as
the
roll
of
the
dice
or
the
roll
or
the
spinning
of
a
roulette
wheel.
You
know,
there's
some
or
you
know
maybe
like
a
some
sort
of
stochastic
switch
where
you
know
one
state
or
another
pops
up
depending
on
just
kind
of
chance,
and
so
this
is
all
like.
You
can
put
this
in
terms
of
casino
games.
B
If
you
wish,
I
wrote
a
paper
on
cellular
reprogramming
and
this
sort
of
idea
of
a
stochastic
cycle
of
where
you
have
self-renewal
or
differentiation,
and
this
this
has
been
addressed
in
literature
in
a
number
of
contexts
as
well.
So
this
is
definitely
something
to
you
know
if
you're
interested
there's
a
fair
amount
of
literature
out
on
this.
B
D
B
This
idea
of
self-renewal
is
sort
of
like
a
roulette
wheel
where
you
have
this
decision.
That's
made
based
on
just
kind
of
chance,
or
you
know
some
chance
outcome,
and
so
as
it
turns
out,
though,
those
chance
outcomes
are
modified
quite
a
bit
by
other
variables.
So,
like
I
said,
cells
communicate
with
one
another
cells.
Will
you
know
one
cell
might
differentiate
and
then
tell
the
neighboring
cells
to
differentiate
instead
of
renewing.
B
B
B
That
the
cells
are
completely
of
a
pattern.
It's
just
that
there's
a
certain
amount
of
noise,
that's
generated
as
genes
get
expressed,
so
there's
noise
in
the
amount
of
transcript.
That's
produced,
there's
noise
in
the
system
that
results
in
the
expression
of
transcripts
and
rna,
and
so
this
noise
is
a
fluctuation
of
amount
of
this.
B
The
the
amount
of
the
molecule
in
the
cell-
or
it
could
be
just
like
fluctuations
in
time,
so
more
gene
product
is
produced
at
time,
eight
and
time
b
and
then
even
more
time
c.
So
there's
this
fluctuation
and
so
all
of
the.
But
this
some
people
think
that
this
noise
actually
does
play
a
role
in
tuning
some
of
these
stochastic
processes.
B
So
you
have
intrinsic
sources
of
noise
and
extrinsic
sources
of
noise,
so
intrinsic
noise
is
sort
of
variation
in
that
machinery.
So
when
you
know
transcripts
and
rna
are
produced,
they
get
produced
at
different
rates
over
time
and
there's
fluctuations
just
based
on
like
the
different
physiological
conditions.
There
are
also.
B
They
go
back
to
drosophila,
so
they
connect
it
back
to
drosophila.
So
then
they
go
into
vertebrate
neural
retina,
which
is
another
good
example
of
this.
The
retina
is
full
of
different
types
of
cells.
How
do
they
differentiate
and
then
the
idea
that
stochastic
factors
are
also
involved
in
retinal
neurogenesis
gathered
traction
on
a
study
using
dissociated
retinal
progenitor
cells
revealed
that
clones
can
vary
highly
in
sizing,
composition
and
again,
we've
talked
about
this
in
in
c
elegans
development,
where
you
know
cells
have
a
certain
size
and
composition
in
different
tissues.
B
So
they
talk
about
dice
here
they
talk
about
throwing
dice
and
then
the
dice
can
be
loaded
so
that
certain
outcomes
more
likely
than
others,
and
so
this
is
selfie
decisions.
When
they're
you
know,
when
you're
looking
at
a
stochastic
model,
you
know
there
are
ways
that
you
can
look
at
it.
That
say
you
know
there
are
influences
on.
The
stochastic
can
be
like
a
biased,
stochastic
model,
and
so
then
they
actually
can
do
this
in
vivo.
B
D
B
B
B
You
have
a
model
worm
that
has
different
layers.
You
have
the
epithelial
layer,
which
is
the
outer
skin.
You
have
the
nervous
system,
you
have
the
nerves
that
are
connected
neurons.
You
have
the
neurons
themselves
identified
and
you
also
have
some
of
the
other
organ
systems.
So
you
can
see
on
the
left.
There's
this
control
panel,
where
you
can
switch
between
those
views
you
can
see
in
this
example.
These
views
are,
we
can
show
these
views.
B
And
you
can
also
manipulate
the
position
of
the
worm,
so
this
is
a
3d
model
and
blender.
If
you
look
at
this
example,
you
can
see
that
it's
a
standard,
blender
model
and
it
just
needs
to
be
brought
into
this
virtual
environment.
It
might
actually
end
up
building,
maybe
a
small
app
where
we
can
render
the
worm
and
maybe
fly
through
it.
I
don't
really
know
if
it's
how
this
model
this
blender
model
is
about
10
years
old
and
we
might
be
updating
it
soon.
So
stay
tuned
for
more
work
on
that.