►
From YouTube: Akiva Lipshitz Discussion (DevoWorm, OpenWorm)
Description
Akiva Lipshitz, leading a discussion during the DevoWorm weekly meeting on June 6, 2016 on "Modeling Gene Expression with Differential Equations" by Chen, He, and Church
A
I'll
just
try
to
give
an
overview
and
then
maybe
we
could
just
have
a
group
discussion
afterwards.
So
the
premise
of
the
paper
is
there's
a
lot
of
there's
a
lot.
We
have.
We
have
the
genome
of
many
different
organisms
mapped
today,
but
there's
there's
a
big
leap
in
understanding
between
the
genome,
but
between
just
the
static
genome
and
then
the
functional
or
the
function
of
that
genome.
So
yet
we
need
to
develop
some
framework
for
understanding
understanding
how
behavior,
both
at
the
macro
and
a
micro
level
emerges
from
the
genome.
A
So
so
two
things
here,
one
is
I'm
actually
talking
out
of
order
in
the
slides
I'll
just
talk
that
the
second
is
okay.
The
first
is
I'm
a
model
for
gene
expression,
and
the
second
is
an
algorithm
for
parameterizing
the
model
given
gene
expression,
giving
gene
expression,
data
which
can
be
a
which
would
be
obtained
with
some
sort
of
imaging
in
the
organism
of
interest.
So
there
there
has
been
some.
A
There
has
been
some
previous
work,
but
so
one
of
the
things
that
that
is
discussed
in
the
paper
is:
are
these
boolean
models
where,
where
genes
are
modeled,
where
genes
are
either
expressed
or
not
expressed?
And
that
was
that's
a
little
bit
simpler
here
in
this
paper,
they
they
try
to
try
to
present
a
gene
expression
model
based
on
the
actual
rate
of
expression
of
the
of
each
gene
and
I
am
I.
Being
clear
here.
A
C
C
A
Okay,
so
so
there
have
a
very
simple
thing
and
maybe
I
should
have,
and
I
didn't
should
have
looked
into
all
of
these
other
models
that
have
these
other
works,
that
are
that
I've
been
quoted
so
the
first.
The
first
is
a
boolean
model
for
gene
expression
where
genes
are
either
on
or
off
and
and
then
this
second
paper
Liang
two
tries
to
propose
a
framework
for
parameterizing
this,
the
boolean
network
model
by
computing,
the
mutual
information
between
the
different
gene
expression,
time
series.
A
A
You
know
what,
since
I
could
I
don't
I,
don't
have
anything
on
that
I
could
just
open
up
the
Wikipedia
page,
what
it
what
it,
what
it
tries
to
do
is
compute
the
entropy
for
the
Joint
Distribution
of
the
two
different
of
the
of
of
your
two
time
series
so
where's
the
time
series.
Okay!
So
we're
talking
about
time,
series
on
a
gene,
gene
concentration
over
time
or
or
for
any
number
of
genes
in
your
organism
and
yet.
A
A
C
B
A
The
mutual
information
between
the
two
different
time
series
and
I
don't
know
how
they
exactly
the
details.
There's
I
know
with
mutual
information.
There's
some
a
lot
of
nuance
depending
you
get
different
results,
depending
on
how
you
on
how
you've
been
your
data
for
the
probability
distribution,
so
I,
don't
know
how
they
did
that
and
I.
Don't
I,
don't
intuitive
intuitively
I
can
infer
how
they
might
have
tried
to
parameterize
their
the
boolean
model
with
the
mutual
information,
but
I.
A
Don't
I,
don't
know
exactly
how
that
was
done,
ready
so
that
that
was
that
to
know
that
that's
an
old
model
and
they
tried
they
tried
to
they
try
to
build
on.
They
try
to
build
something
more
more
exact
where
you,
where
you
can
model,
not
not
not
not
boolean
gene
expression,
but
you
can
actually
take
into
account
the
rate
of
expression
and
that's
what
gives
this
model
a
little
bit
more
smoothness
or
we
want
it
would
have.
A
It
would
have
been
cool
if
I
had
actually
tried
to
simulate
something
with
this
model,
but
I
didn't
okay,
so
that's
the
boolean
network
and
I'll
go
back
to
the
slideshow
so
that
they're,
just
at
the
beginning
of
the
paper,
they
just
talked
about
some
other
things
that
have
been
done.
It
doesn't
didn't
really
seem
to
me
to
affect
the
ideas
presented
in
the
paper.
So
much
I
think
it
was
just
an
overview
there's
there
are
some
other
if
they
talk
about
some
other
methods
for
optimizing.
B
A
Yeah
yeah,
something
like
a
theater
thing.
Interestingly,
a
few
a
few
weeks
ago,
I
saw
this
paper
which
had
they
developed
software,
where
you
could
specify
a
logical
model
for
for
for
gene
expression
and
then
it
would.
It
would
automatically
generate
the
DNA
sequences
promoter
sequences
so
that
you
can
implement
that
logical
model
in
in
real
DNA.
So
I
thought
that
was
cool
yeah.
B
A
Yeah,
okay,
well
and
then,
and
then
this
last
paper
tried,
tries
to
build
upon
this
the
Michael's
methods
for
optimizing,
this
logical
model
there.
Apparently
there
are
some.
There
are
some
errors
in
the
there's,
some
just
noise
and
the
data,
and
that's
what
that's
what
this
last
paper
deals
with,
so
the
objective
so
that
that's
that's
all
the
prior
work
that
the
paper
cites
and
now
I'll
just
go
into.
This
is
all
in
the
paper
the
objectives
that
they
state
for
their
model.
They
want.
The
idea
is
you
have
they
have
some?
A
You
have
some
prior
mathematical
model
for
gene
expression
for
any
number
of
genes
in
the
organism,
and
you
should,
from
this
data,
derive
drive
the
relationships
between
how
certain
genes
affect
the
expression
of
other
genes.
This
model
should
stuck
to
the
entire
genome
level,
I,
enabling
you
to
simulate
do-do
gene
expression,
modeling
for
for
an
entire
cell,
and
then
the
last
thing
is
in
the
last
thing
is
important.
Then
they
only
get
to
that
in
the
end.
Take
the
take
the
time
delay
into
account
because
it
takes.
A
It
takes
a
certain
amount
of
time
for
mRNA
to
be
transcribed
and
translated
into
protein.
So
that's
they
don't
take
that
into
account
in
their
initial
model,
but
in
the
very
end
they
they
they
add
on
to
their
initial
model,
to
deal
with
which
to
deal
with
that
little
little
little
nuance
of
time
delay
so
I'm
I
I,
don't
want
to
go
into
I.
Don't
want
to
go
into
all
the
math
because
I
understand
it's
somewhat,
but
to
present
it
would
be,
is
a
little
bit
too
much
for
me.
A
But
if
you
want,
we
could
go
through
the
we
could.
You
could
pull
up
the
paper
and
and
maybe
discuss
it
in
as
a
group,
exactly
the
math
behind
all
of
this,
were
you
expecting
me
to
present
the
math
behind
all
of
this
okay,
because
most
of
the
paper
is
actually
just
so.
Here's
here's
the
model
and
I
can
explain
how
this
how
this,
how
these
two
differential
equations
work
and
then
a
lot
of
the
rest
of
the
paper
just
goes
into
how
you
solve
these.
A
A
So
there
are
two
equations
here:
two
differential
equations,
one
of
them
just
one
of
them
models,
mRNA
concentration
or
one
of
the
models,
the
the
derivative
of
mRNA
concentration
and
the
other,
the
other
models,
the
the
protein
concentration.
So
these
all
of
these,
this
is
I,
don't
know
what
the
technical
term
for
it,
but
it's
a
vector,
vectorized
differential
equation.
So
all
of
these
variables
are
vectors,
so
mRNA,
a
the
change
in
mRNA
is
going
to
be
a
is
going
to
be
a
function
of
the
concentration
of
Trent
I.
A
Don't
know
what
is
a
word
for
it:
transcription
factors
at
minus
the
degradation
rate
of
mRNA,
so
the
the
existing
mRNA
will
degrade,
and
so
you
need
to
take
that
into
account
when
you're.
Looking
when
you
want
to
know
what
the
what
the
change
in
mRNA
concentration
is
going
to
be,
and
similarly
for
the
evolution
of
for
the
protein
concentration,
Duras
is
going
to
be
a
function.
A
linear
function
of
the
current
concentration
of
mRNA.
A
Because
proteins
proteins
are
translated
with
the
ribosome
from
mRNA
the
degradation
rate
of
protein,
so
obviously
you
need
to
know
what
F
of
P
is
and
I
don't
so
one
of
the
questions
I
have
for
you
is
why
why
FFP
needed
needed
to
be
undefined
here?
Why
why
they
needed
to
solve
for
it
and
why
they
couldn't
have
just
built
that
into
this
initial
model,
do
you
have
any
ID?
Do
you
have
any
ideas.
B
B
A
A
A
Okay
s,
just
on
a
side
note
I
was
thinking
to
to
understand
like
how
proteins
compute
things
or
how
proteins
can
perform
their
function.
You
can
think
of
proteins
as
like.
A
mechanical
neural
network,
where
the
where
the
propagation
of
information
is
is
really
just
molecular.
Intermolecular
forces
or
forces
between
the
amino
acids
and
the
state
of
each
neuron
is
the
position
of
the
amino
acids
in
time
a
position
of
amino
acids
in
space.
So
you
can
think
about
proteins
as
as
neural
networks
and
then
using
the
same
techniques.
A
A
B
A
C
A
A
So
they
they
try
to
solve
for
F
of
figure
out
what
F
of
P
is.
They
view
it
as
a
black
box
and
then
try
to
describe
it
relative
to
other
things.
So
in
the
unit
paper
and
I
I
understand
the
nature
of
this,
but
I,
don't
underst
I
haven't
really
studied
Taylor
Taylor,
approximations
I
haven't
I've,
studied
calculus
in
bits
and
pieces
as
I've
needed
it.
So
I
haven't
really
done
a
formal
experience
with
all
aspects
of
calculus,
but
so
they
at
the
end
they
they
describe.
A
A
A
A
A
Okay,
this
is
going.
This
is,
is
all
of
the
the
detail
they
just
they
talk
about
how
they
how
you
solve
you
solve
this
differential
equation
with,
given
the
input
given
the
gene
expression,
time
series,
and
then
they,
then
they
talk
about
this
other
another
model
that
builds
upon
them.
This
model
that
I
was
just
talking
about
to
model
just.
A
A
A
So-
and
I
should
have
should
have
said
I
don't
I
don't
understand
what
c
is
they
don't
really
define?
What
c
is
it's
just
some
some
matrix
for
evolving
for
some
matrix
for
describing
the
relationship
for
describing
the
relationship
between
the
protein
concentration
and
when
you,
when
you're
starting
to
look
at
M,
RNA,
mRNA
trajectories?
That's
over
here,
that's
highlighted
I,
don't
I,
don't
understand
why
they
couldn't
have
just
put
C
up
in
this
model.
I
don't
know,
but
these
other
these
other
matrices
L
is
a
translation.
A
Is
a
trend,
the
translation,
the
the
protein
translation
rate,
given
the
mRNA
concentration
V
is
the
degradation
rate
of
mRNA
you're.
Given
the
current
concentration
of
mRNA
U
is
the
degradation
of
proteins,
so
you
can
given
the
current
concentration
of
protein.
What
is
the
degradation
rate
and
I?
Suppose
C
is
the
four
friends
late
transcription
rate,
given
the
current
concentration
of
proteins
and
yeah
like
the
I?
Didn't
the
eigenvalues
will
tell
you
the
most
the
axe,
the
the
axes
of
most
variation
among
the
genes
right.
A
A
Pause
and
then,
if,
if
any
of
the
eigenvalues,
if,
if
all
the
real
parts,
the
eigen
values
are
non
positive,
if
they're
negative,
then
you
end
up,
you
end
up
gaining
a
semi,
stable
system,
it's
just
oscillate,
but
the
oscillations
get
the
amplitude
of
the
oscillations
get
larger
and
larger
and
larger.
Oh
okay,.
B
A
So
I
just
thought:
my
interest
was
in
I've,
been
interested
in
neuroscience.
That's
what
I'm
doing
here,
Princeton
and
I
was
thinking
you
could
apply.
The
same
techniques
to
understanding
genetic
networks
and
I
was
I
was
hoping
me
that
gene
expression,
gene
expression
plays
or
definitely
plays,
a
role
in
the
development
of
C
elegans.
So
I'm
wondering
how
you
think
this
could
integrate
with
devil
worm.
Where
does
where
would
it
specifically
come
in.
A
B
A
A
A
B
C
A
B
B
B
A
Right
so
I
was
in
tenth
grade
and
filthy.
Until
a
month
ago,
I
was
a
little
bit.
I
was
very
miserable
in
10th
grade,
I
wasn't
really
able
to
learn.
I
felt
I
felt
a
little
bit
stifled
very
settled,
so
I've
been
very
I've,
been
involved
with
open
worm
for
the
past
few
months
and
wanted
I
wanted
to
pursue
neuroscience
the
greatest
extent.
A
So
I
for
the
last
month,
I've
been
at
a
neuroscience
lab
at
Rockefeller
University
and
then
now
I'm
at
the
in
the
lab
at
Princeton
and
Murphy
Murphy
lab
they're
doing
a
fruit
fly,
courtship
courtship
behavior
with
fruit
flies
and
Drosophila.
So
that's
where
I
am
right
now,
yeah,
that's
nice.
A
C
A
C
A
C
And
yes,
I
started
college
at
age,
sixteen
Wow,
okay
and
it
was
worth
doing
it
was
really
you
know:
I
didn't
miss
much
by
skipping
the
last
year
of
high
school.
C
C
C
B
A
B
C
B
C
A
C
A
B
C
C
C
C
A
C
A
Neuroscience
well
now,
I
meant
so
in
the
past
few
weeks,
as
I
have
talked
to
some
of
the
biologists
at
Rockefeller
and
I've
realized
that
I
region
aliy
thought
that
neuroscience
was
so
was
so
amazing
because
you
want
you're
trying
to
understand
computation
in
these
complex
systems
when
you
have
many
interacting
components
and
there's
emergent
complexity,
and
that
just
fascinated
me
but
now
I'm,
seeing
that
it.
You
see
this
in
all
of
biology
and
in
fact,.
A
C
To
do
a
paper
which
I
wrote
recently
with
Rob
stone,
called
cybernetic
embryo
and
it's
it's
the
puzzle
of
how
an
embryo
builds
itself
which
is
kind
of
what
the
Aviva
word
project
is
about.
Okay,
okay,
so
I'll
send
you
that
it's
it's
impressed
right
now.
Thank
you,
but
it's
the
thing
that
we're
finding
is
standard
approaches,
emergent,
behavior
physics,
approaches.
Reductionism
said
these
are
not
working
in
terms
of
solving
these
problems,
so
the
this
isn't
a
stab
of
trying
to
find
a
middle
ground.
That's
productive!
C
B
Know
I
was
kinda
involved
in
that
I.
Well,
I
still
have
a
little
bit
artificial
life
and
about
like
10
or
so
years
ago.
There
was
sort
of
a
consensus
that
oh,
it's
it's
it's
kind
of
plateaued,
because
we
don't
have
the
right
chemistry.
You
know,
that's
not
rich
enough,
so
it's
hard
to
build
compensation
models
of
something.
That's
so
yeah.
You
know
so
amazingly
complex.
You
know.
A
C
The
problem
with
most
neuroscience
is
very
simple,
and
that
is
a
bit
usually
deals
with
the
adult
organism,
and
so
you
miss
how
on
earth
thing
got
constructed
and
then
started
functioning
or
functioned
as
it
got
constructed,
and
we
don't
really.
You
know,
I
mean
if,
for
instance,
you
take.
If
you
take
vision,
we
we
now
know
that
if
you
take
kittens
that
the
the
early
visual
experience
involves
a
lot
of
cell
death
in
the
visual
cortex,
you
familiar
with
this
yeah.
C
Okay,
and
so
my
attitude
has
been
that
we
have
to
go
back
to
the
embryo
and
see
how
the
how
the
brain
gets
built
in
the
first
place,
and
basically,
we've
been
working
at
the
levels
so
far
up
to
the
trying
to
span
the
gap
from
these
single
egg
to
the
formation
of
the
neural
plate,
from
which
the
brain
then
develops
and
most
neurophysiologist
start
at
the
adult
level
and
sometimes
reached
a
little
bit
back
to.
Is
this
huge
gap
between
neural
plate
level
and
the
functioning
adult
brain,
which
has
almost
a
very
little
investigation?
C
There's
a
bit,
but
not
so,
and
usually
it's
very
different
people
who
do
two
different
things.
So
by
hooking
up
with
the
Deebo,
with
the
open
worm
project,
we're
hoping
that
drag
the
neurophysiologist
and
the
computational
scientists
who
were
working
on
open
worm
back
a
little
bit
to
watch
how
the
worm
get
built
in
first
place,
and
maybe
there'll,
be
some
insights
and
how
it
functions
as
results
of
understanding.
How
it's
put
together
and
and
since
an
embryo
has
to
function
all
the
time
anyway,
there
may
be
a
development
of
the
functioning.
C
C
C
C
A
C
So
we
got
that
and
then
it
goes
on
and
divides
and
it
also
changes
kind
whenever
a
new
kind
of
cells.
So
if
we
could
tag
each
cell
with
what
we
know
about
gene
expression
in
each
cell,
we
might
start
to
see
patterns
in
this
and
start
to
make
some
sense
of,
and
that
might
be
better
than
what
you
the
kind
of
paper
you've
presented,
which
is
an
a
priori
model.
Look
how
it
occurs.
I
mean
you.
A
C
And
yeah
we
know,
gene
expression
is
not
lived.
So
it's
it's!
It's
it's
sort
of
just
smooth
over
all
of
that
stuff,
but
non-linearity
might
be
more
important
than
having
linear
expressions
because
often
nonlinearities
are
what
needs
to
trigger
and
to
flipping,
to
changing
the
state
to
bifurcations
things
like
that:
okay,
okay,
so
so
one!
So
what
I'm
saying
is
if,
if
you're,
really
interested
in
gene
expression
and
I,
don't
let
me
I
should
Bradley
how
many?
What's
how
many
cells
does
the
nematode
have
when
it
gets
its
first
nurse?
Oh
yeah,
yeah,
yeah.
C
B
B
A
A
Well,
there
are
only
a
few
neurons
in
the
in
the
embryo,
because
one
one
of
the
things
that
has
is
being
done
is
people
are
trying
to
our
probing
every
neuron
in
C,
elegans
I,
know
someone
at
in
the
barkman
lab
at
Rockefeller
is
doing
this,
but
if
you
knew
when
you
have
like
a
minimal
network
at
some
early
stage
in
development,
what
what
functions
these
neurons
play?
What
roles
do
these
norms
play.
B
Yeah
so
like,
like
even
let's
say
like
at
a
very
young
stage,
perhaps
then
they're
all
structures
are
there
for
locomotion.
You
know,
but
not
not
some
other
things.
You
know
you
know
so
like
it's
like
its
life
stages
as
it
as
they
evolve
and
they
get
more
complex.
It's
and
they're
all
system
would
come
into
play
to
to
to
enable
that.