►
From YouTube: DevoWorm (2023, Meeting #29): DevoLearn, Comsol Modeling, Molecular Bio of Differentiation, G-P Maps
Description
DevoLearn updates. Challenges of modeling in COMSOL. Genomics of Differentiation: proteomics of development, bioinformatics, comparative biology, and paper planning. Genotype to Phenotype (G-P) maps: theory, computational models, and evolution. Attendees: Sushmanth Reddy Mereddy, Bradly Alicea, Morgan Hough, Susan Crawford-Young, Richard Gordon, and Lukas.
C
C
I
was
working
on
the
loss
function.
I
wrote
this
all
code,
which
are
looking
as
green
here
right
now.
This
all
is
a
part
of
last
function.
I
have
implemented
different
loss
functions
and
I
am
comparing
between
them,
which
loss
function
will
give
the
best
results
right
now.
I
have
the
dice
lost
PCA
loss
and
to
implement
this
I
need
to
read
a
paper
actually
all.
C
C
Yeah
I
have
read
this
paper
and
I
have
implemented
it
unified
focal
loss.
This
is
about
a
loss.
Function
mainly
use
it
for
segmentation.
This
paper
was
completely
dependent
on
that
only
mainly
focused
on
the
particular
loss
function
called
unified
focal
loss.
It
is
an
intersection
of
dice
loss
and
cross
entropy
laws.
I've
read
the
whole
paper
and
I
understood
and
I
have
implemented
their
methods
in
the
loss
functions
so
right
now,
whatever
I
could
have
was
that
dice
loss
was
a
common
function.
C
Everyone
will
use
I
added
the
dice
pce
loss,
IOU
loss,
IO,
PC
loss
I
mean
different
things.
Symmetrical
focal
loss
but
unified
focal
law
is
a
mix
of
two
loss
functions.
One
is
asymmetrical.
Sorry
one
was
dice
dice
loss
and
one
was
PC
loss.
This
complete
will
took
around
time
around
this.
Only
reading
the
paper
understanding
it
implementing
commenting
every
code,
how
it
is
working
and
extra
Mike
told
me
to
write
code
in
this
way
like
a
parameters.
What
will
it
return?
Etc?
C
So
it
took
a
couple
of
days
to
implement
all
this
stuff
learn
about
it,
and
these
loss
functions
are
not
available
over
internet.
Actually,
generally,
loss
functions
will
be
available
in
the
pytots
library
or
something
like
that,
but
this
unified
focal
loss
was
completely
different.
I
couldn't
find
it
online.
So
I
wrote
my
own
code
and
I
made
it
work.
It
is
working
right
now.
Actually
this
code
whole
path.
I
will
try
to.
This
will
be
my
this
week.
Update
would
be
this
only
and
yeah
I
haven't
seen.
C
Actually
you
told
Bradley,
that's
my
this
upgrade
will
be
extended
on
Monday
haven't
seen
checked
today,
but
yeah.
Hopefully
it
get
extended.
Yeah-
and
this
is
my
update
and
this
week
I
will
be
writing
the
doc
about
the
code.
What
you
have
mentioned,
and
all
stuff
and
Implement
all
these
things.
There
are
a
lot
of
things.
C
I
need
to
implement
it
yeah
that
was
my
pretty
update
all
about
and
these
loss
functions
are
mainly
used
in
the
devonet,
not
in
the
Sam,
so
these
loss
functions
will
be
compared,
which
loss
function
will
give
us
better
output,
so
I'll,
compare
which
one
will
use
better
output
and
I
will
Implement
them
I'm,
adding
the
project
structure.
In
this
way,
parameters
will
be
taken
from
one
file
and
those
will
be
implemented
code
directly.
No
one
needs
to
run
some
terminal
commands.
I
am
using
our
parts
Library.
C
Yeah,
this
is
a
python
Library
which
is
used
for
command
line
arguments.
We
can
keep
our
all
whatever
we
need
in
a
file
and
those
will
be
taken
as
the
command
line
arguments.
We
can
change
we
need
to.
If
someone
wants
to
use
deponent,
they
just
need
to
change
the
parameters
here
and
just
run
a
one
line
of
code,
so
you
can
train
it
completely
by
itself.
So
that
was
my
implementation
and
yeah.
Pretty
this
week,
I
have
worked
on
the
loss
functions
only.
C
C
Yeah
I
will
I
will
change
all
these,
which
you
have
mentioned
right
now,
because
there
are
a
lot
of
things.
I
need
to
change.
I
haven't
seen
the
dog
actually
after
you
mentioned
it,
because
due
to
this
paper,
I
was
reading
this
paper
and
trying
to
understand
so
it
took
some
time.
There
are
a
lot
of
technical
terms.
I
need
to
understand
before
implementing
this
all
stuff.
So
yeah
I
will
read
these
comments
and
I
will
try
to
implement
that
yeah.
C
C
C
A
I
would
wait
that
would
be
kind
of
one
of
the
final
things,
because
once
you
get
through
the
paper,
you'll
see
kind
of
how
that
shapes
out
to
be
like.
If
you
have
you
describe
what
what
the
platform
is,
the
different
things
that
you
worked
on
describe
the
results
and
then
think
you
know
what's
what's
kind
of
The
Next
Step
here.
A
What
is
it
that
we
could
do
next
and
we've
talked
about
some
of
those
things,
but
some
of
those
things
are
not
like
obvious
until
you
actually
go
through
the
results,
because
sometimes
there
will
be
things
that
you'll
notice
in
the
results
there
like,
oh
yeah,
we
need
to
work
on
that,
then,
usually,
when
people
do,
you
know
like
future
directions
or
things
like
that.
They'll
usually
think
at
two
levels,
they'll
think
it's
sort
of
like
the
immediate
level
where
like.
A
If
you
look
through
the
results
and
you
notice
there's
something
that
isn't
really
resolved.
You
know
you
might
say:
well
we
want
to
do
this
to
improve
the
loss
function
or
we
want
to
do
this
to
improve
the
performance
of
segmentation
and
then
or
you
know,
and
then
that
those
sorts
of
things
are
the
first
kind
of
immediate
level
future
Direction.
Then
the
next
level
is
like.
How
do
you
want
to
integrate
this
into
some
other
system?
A
C
We
need
to
integrate
with
lineage
population
when
we
give
a
direct
image.
That
would
be
our
immediate
immediate
of
extension.
Of
this
we
are
trying
to
segment
the
cells
right.
According
to
the
cell
positions,
we
will
try
to
extract
the
teenage
population.
I
mean
I,
have
read
these
papers
actually
in
that
cell,
or
something
like
names
given
are
based
on
the
position
of
the
cell.
If
they
are
located
at
anterior
exterior
position,
it
is
called
as
a
something
like
that.
I
am
trying
to
understand
it.
Yeah
I
mean
next
project
would
be
that
only.
A
A
B
D
D
D
D
B
C
A
A
Yeah
so
yeah
we've
been
working
on
this
paper,
the
bioinformatics
part,
so
Lucas
has
been
working
on
the
protea
or
well
the
protein
blasting
and
then
I've
been
working
on
some
other
things,
and
we've
been
talking
about
how
to
make
comparisons
and
matches
so
Lucas.
Did
you
wanna
talk
a
little
bit
about
what
you've
been
doing?
Oh.
E
A
E
I'm
I've
been
facing
like
a
small
challenge
because
for
the
beta
catamine
protein
I
couldn't
find
any
protein
sequins
for
the
draws
a
few
year.
One
for
the
fruit
fly
yeah,
but
yeah
I've
been
discussing
some
interesting
proteins
with
Bradley
as
well,
so
we're
gonna,
I
I've,
no
I've
taken
I've
taken
note
of
those
ones
as
well,
so
the
interesting
protein.
E
B
A
All
right
so
put
something
in
the
chat
couple
things
in
the
chat,
so
the
first
thing
was,
of
course,
the
beta
kitten
and
so
a
lot
of
times
in
different
species.
It's
not
always
the
same
thing.
They
don't
call
it
the
same
protein
or
there's
a
homolog,
which
means
it's
a
it's
this
it's
a
sort
of
a
Divergent
form
of
that
thing,
but
it
has
a
different
name.
A
That's
you
know
that
this
is
nomenclature
a
lot
of
times
like
in
C,
elegans
and
drosophila,
the
genes
of
different
names,
but
they're,
basically
the
same
thing.
So
if
we
can
like
find
out
if
you
could
make
a
list
when
you're
doing
this
analysis,
if
you
see
something
it
doesn't
match,
you
know,
send
it
along
or
create
a
table
is
a
better
strategy,
and
you
know
you
update
the
table.
A
You
would
have
like
a
note
that
would
say
this
is
also
called
this
in
drosophila,
so
you'll,
you
can
see
that
the
similarity
in
the
sequence,
but
you
know
it's
not
going
to
be
the
same
name.
So
that's
that's
how
you
kind
of
like
clean
up
some
of
those
entries
when,
when
we
do
comparison,
sometimes
we're
not
able
to
compare
directly.
So
we
have
to
make
these
indirect
comparisons.
A
A
Yeah
I
mean
a
lot
of
times
yeah,
it's
just
like
people
are
working
on
it
like
when
in
C
elegans,
there's
a
community
of
people
who
have
like
characterized
the
genes
and
everything
and
then
drosophil
has
another
community
and
they
don't
necessarily
talk
to
one
another
about.
Is
this
the
same
gene
because
remember
you
know
a
lot
of
these
genome
draft
genomes
didn't
exist
when
they
were
naming
the
genes,
so
they
were
just
sequencing
genes.
A
You
know
one
at
a
time
and
they're
trying
to
figure
out
what
they
were,
and
so
this
is
where
you
know,
there's
no
overarching
nomenclature
so,
but
sometimes
there
is
I
mean.
Sometimes
people
know
what
something
is
and
they
give
it
the
name,
but
sometimes
there
isn't
so
then
there's
balchin
protein
involvement
in
chromosome
cytoskeletal
organization,
which
might
give
some
Clues
yeah.
Please
make
that
note
of
that,
and
then
you
know
try
to
maybe
do
a
search
for
that.
B
A
A
A
I
sweated
every
thought:
it's
like
yeah
yeah,
so
drosophila
yeah,
so
drosophila
is
one
of
our
comparison
groups.
That
herself
was
actually
a
Texan,
so
it's
like
can
or
hebditus.
We
have
Kenner
abdidis
elegans
and
we
have
some
other
worms
that
are
related
and
actually
found
a
good
paper
on
that
this
week.
I'll
probably
talk
about
it
after
but
yeah
it's
a
drosophila.
A
The
one
we're
usually
interested
in
is
melanogaster,
which
is
just
you
know,
it's
the
model
organism
so,
like
you
know
it's
like
C
elegans,
and
then
there
are
other
drosophila
species
and
I
think
those
have
been
less
characterized
by
like
models
at
that
model.
Organisms,
sort
of
level
of
specificity,
but
but
we
there
have
been
a
lot
of
so
you
may
find
things
that
are
drosophila,
Spa
or
which
is
sp
period,
which
means
they
don't
know
what
the
species
is
or
some
other
drosophila
species.
A
If
you
can
find
resolfo
a
melanogaster,
then
you
know
that's
the
thing
you
go
with,
but
if
it's
drosophila
Spa,
then
that
would
okay
be
okay,
especially
in
this
context,
where
we're
doing
like
we're
comparing
the
C
elegans.
So
it's
like.
Obviously,
there's
a
difference
there,
quite
a
big
taxonomic
difference
so.
A
B
E
A
The
the
annotations
and
usually
they
do
like
they'll
just
cite
like
studies
and
that's
The
annotation.
So
it's
like
the.
If
you
notice
the
annotations
are
rather
broad
like
it'll,
say
phosphorylation
or
it'll,
say
cytoskeleton,
and
it
doesn't
get
much
more
specific
than
that
and
that's
okay
I
mean
that
gives
us
Clues
to
where
you
know
what's
going
on,
but
you
know
I
I,
think
they're
I,
don't
see
anything,
that's
really
not
involved.
A
I,
don't
you
know,
we
don't
really
know
like
there's
a
lot
of
genomic
data,
a
lot
of
molecular
data
being
generated
and
a
lot
of
it's
been
characterization
up
to
this
point.
So
when
we
say
something
about
function
with
a
lot
of
molecular
data
unless
we're
doing
like
a
specific
experiment,
it's
hard
to
really
kind
of
say,
make
definitive
statements
about
what
that
looks.
Like
I
mean,
there
are
very
few
studies
that
actually
show
like
a
broad
study
like
that,
where
they
say
you
know
these.
A
These
Gene
these
proteins
are
involved
in
this
process.
These
proteins
are
involved
in
this
process
and,
let's
make
a
comparison,
I
mean
it
does
occur.
But
again
it's
like.
Usually
it's
pretty
focused
on
specific
questions.
So
that's
something
we'll
have
to
sort
of
work
around
too,
but
that's
that'll
be
for
later.
A
So
there
is
also
a
lot
of
structural
stuff
to
be
expected
cell,
so
adhesion
microtubules
Etc
at
high
counts.
Yes,
yes,
there
should
be
a
lot
of
structural
stuff,
because
when
you
do
differentiation
of
all
cell
division-
and
you
know
that's
going
to
happen-
and
you
have
to
in
fact
you
have
to
have
a
cell
division
a
lot
of
times
have
differentiation
I,
don't
know
if
that's
entire
I
mean
well,
you
do
have
to
have
it
to
get
the
you
know.
A
You'll
have
to
at
least
replicate
DNA
to
have
the
things
expressed
that
you
need
to
have
so
it's
it's.
It's
a
it's
tied
to
differentiation
cell
division
and
some
of
the
other
in
things
like
cell
cell
adhesion.
You
know,
cells
are
moving
around
they're,
changing
their
shape.
They're.
You
know
that
that's
what
those
are
going
to
be
important
things.
D
If
you
want
a
picture
of
where
some
of
the
molecules
are
in
their
function,
you
can
look
at
this
the
paper
there,
because
it's
got
the
alpha
beta
continents
and
microtubules
and
I,
don't
know
if
it
has
microtubules.
But
anyway
it's
got
actin
filaments
and
where
they're
placed
in
in
the
cell
sort
of
It
kind
of
shows.
You
like
the
physical
layout
paper,
it's
a
an
older
paper,
but
that's
okay,
yeah.
A
A
All
right,
yeah
and
so
yeah
so
I
also
see
some
interesting
counts
for
things
related
to
phosphorylation
related
to
differentiation,
induction
and
human
stem
cells.
Yeah
I,
we
went
over
some
papers
on
this.
Phosphorylation
is
important
in
a
lot
of
things
in
the
cell,
especially
for
molecular
aspects
of
the
Sally
for
function.
So
you
have
this
whole,
these
phosphorylation
Cascades
that
help
in
function
in
some
functional
aspects.
So
there's
I
send
them
a
paper
on
human
stem
cells
where
they
have
this.
A
They
found
a
lot
of
phosphorylation
related
to
differentiation
induction
there.
So
this
this
is
like
I
guess,
a
naturally
differentiating
cell
that
is
going
to
from
a
stem
cell
to
a
precursor.
So
not
every
differentiation.
Related
thing
is
a
high
count,
however,
and
that
makes
sense
in
light
of
what
I
said
on
Monday
see
how
organs
exhibits,
Mosaic
control
differentiation,
yeah.
B
A
The
mouse
data-
because
you
would
have
you
know
so
you'll-
have
like
basically
four
categories,
so
there's
a
mouse
or
there's
C
elegans.
So
let
me
right
on
the
board
here.
If
you,
if
I'm
sharing
my
screen,
I'm,
not
sharing
my
screen,
let
me
share
my
screen
so
you're
going
to
have
four
things
that
are
four
categories
that
you're
comparing.
So
you
have
C
elegans,
which
is
Mosaic
development,
you'll
have
drosophila,
you'll,
have
I'm
running
out
of
space.
Mouse
and
you'll
have
East
so
you'll
have
the.
D
A
A
If
we
see
in
a
similar
pattern
in
drosophila,
but
maybe
not
the
same
proteins,
and
we
can
also
say
that's
another
example
of
Mosaic
development
and
mouse.
You
have
regulative
development,
and
so
that
means
that
you
should
see
like
a
more
consistent
pattern,
but
we
don't
know
and
then
yeast
would
be
no
development.
A
D
A
There's
this
one
paper
where
they've
dissociated
the
two
cell
stage
and
they
form
like
their
own
little
cell
culture
of
C
elegans
embryo
and
they
just
form
like
a
weird
pattern
on
each
side
and
that's
but
that's
they
I've,
never
seen
anyone
take
Wicca,
sell,
put
it
in
some
other
part
of
the
organism
and
then
it
takes.
You
know,
then,
of
course
the
flatworm
stuff.
You
can
take
a
single
cell
and
it
can
generate
the
entire
organism.
But
that's
that's
a
different
thing.
I
think
yeah.
B
A
Yeah
yeah,
so
yeah,
that's
and
then
so
that's
the
logic
there
and
then
so.
This
is
where
drosophila
would
come
in
handy
and
perhaps
even
Mouse.
If
we
had
Mouse
drosophila,
C,
elegans
and
yeast
for
all
candidate
proteins,
we
would
have
no
development,
yeast,
Mosaic
development,
a
mosaic
development
B,
which
is
drosophone,
C,
elegans
and
then
regular
development
Mouse.
We
could
then
evaluate
the
differentiation
and
non-differentiation
related
proteins
to
see
if
they
fit
that
model
yeah,
and
so
we
have
the
the
protein
list.
D
A
A
What
I'm
doing
is
I'm
going
back
and
I'm
trying
to
find
the
genes
for
different
proteins
and
then
I'm
looking
up
the
CG
motifs
in
that
Locus,
so
I,
don't
know
what
the
where
the
promoter
is
and
I
don't
know
where
the
coding
region
is
necessarily
but
I.
What
I
can
do
is
count
up
the
number
of
CG
motifs
they're,
just
two
based
motifs
and
then
look
inside
that
Locus
for
these
just
the
count
and
then
maybe
patterns.
B
A
Yeah
I
think
that
yeah,
so
that's
like
the
third
part,
would
be
to
map
those
findings
to
like
the
differentiation
code,
or
you
know
something
like
that,
but
that's
actually
something
that
can
be
kind
of
proposed.
I
mean
we
have.
The
like.
I
showed
the
2016
paper.
Where
we
have
this
worked
out
for
cell
size
and
cell
or
lineage
tree.
We
don't
have
the
single
cell
Pro
or
genomics
and
proteomics.
So
we
have
to
go
with
the
whole
organism
and
in
genes,
but
we
can
actually
propose.
A
D
Non-Motile
cell
to
a
moving
cell-
and
it
happens
as
soon
as
there's
no
pressure
or
flow
on
on
the
endothelial
cell.
It's
it
just
switches
cell
type
instantly,
so
there's
a
number
of
cells
that
do
that.
But
that's
the
one
I'm
aware
of
okay
and
it's
I,
don't
know:
I
have
another
picture
of
one
him
in
this
PowerPoint
that
I've
got
of
interesting
cell
things.
I'll
find
it
okay,
yeah
the
PowerPoint
is
98
slides
long.
So
it's
a
bit
long,
yeah.
B
A
A
A
So
you
know,
like
the
annotations,
tell
you
something
about
the
function
like
I
said
it's
imperfect,
but
we
can
actually
maybe
score
them
by.
You
know
core
versus
peripheral
function
and
then
give
a
rationale
for
that.
Why
did
they
do
that
and
then
show
the
differences
out
there
so
yeah
and
then,
like
maybe
some
figures
as
well?
Maybe
we
put
together
a
figure
on
what
that
looks
like
in
terms
of
a
differentiation
code
or
a
differentiation
tree.
A
You
know
what
what
kind
of
Clues
can
we
draw?
So
if
we
had
the
ideal
data
set,
which
would
be
every
cell
in
C
elegans
with
a
you
know,
a
gene
genome
sequence,
a
proteome
analysis.
We
have
some,
you
know
annotations
that
are
excellent.
That's
not
what
annotations
are
currently,
then
we
could
say
this
is
the
way
we
look.
So
we
can.
Actually,
you
know,
map
that
out
a
little
bit.
A
You
know
it's
like
what
you
use
with
you.
There
are
a
lot
of
different
options
for
this,
but
I
just
use
a
Google
jamboard
I'm,
using
a
tablet
on
my
end
and
I'm,
drawing
on
it.
So
yeah
I
have
a
a
tablet
here.
So
that's
so
yeah!
That's
that's
what
I'm
thinking
and
then,
of
course
we
have
the
version
10
draft,
which
has
a
lot
of
the
theory
in
it
from
deck.
A
Yeah
so
yeah
I,
think
and
then
yeah
Lucas
said
I
could
help
you
with
Gene
counting
as
well
yeah.
It's
we'll
we'll
talk
about
that
closer
to
when
we
get
to
that
point
so
yeah
so
for
right
now,
Lucas,
you
know,
let's
get
the
all
the
different
species
down.
Let's
get
you
know
those
together,
I,
don't
know
if
you
want
to
put
them
in
a
table.
E
Like,
let's
say,
APC
KPC
all
about
only
all
these
proteins
and
then
I'll
do
the
I'll
do
the
account
for
table
one
and
then
the
the
protein?
Maybe
we
could
continue
and
then
all
put
like
all
update
the
bioinformatics
detail.
We
like
this
V2
and
then
list
all
the
proteins
on
my
thoughts.
So
we
could
continue
from
that
and
then
we
could
use
both
of
those
data
for
the
table.
Two
that
you're
working
on
the
Gene
Kelly
thing.
B
B
A
I
mean
I
can
help
you
with
that.
You
know
it's
a
matter
of
getting
the
right
sort
of
version
of
it
or
this
the
sort
of
the
visualization
yeah,
which
is
hard
to
do
because
you
have
to
like
kind
of
say
how
do
I
want
to
present
this
like.
If
it's
like
a
you
know,
sometimes
you
know
different
ways
of
plotting.
It
look
better
and
they're
more
informative.
So
that's
another
thing!
Okay,
so
you
understand
what
I'm
saying
yeah
yeah,
okay.
B
A
So
this
is
another
paper:
Susan
posted
a
biomechanical
biophysical
and
biochemical
modulators
of
cytoskeletal
remodeling,
and
this
is
so.
This
is
for
stem
cell
commitment
or
a
cell
lineage
commitment
in
stem
cells.
So
again,
this
is
where
stem
cells
are
going
from
a
stem
cell
to
something
like
a
precursor
cell
to
differentiate
itself
and
each
at
each
step.
It's
going
through
a
cell
division.
It's
remodeling
itself!
It's
doing
all
this!
These
things.
A
D
A
Oh
yeah
yeah,
it's
people
want
to
figure
out
how
that
works,
so
yeah
we
actually.
Last
week
we
talked
about
a
paper
in
I
think
it
was
in
mammals
where
they
took
a
cell.
This
is
the
the
the
lineage
commitment
idea:
they
they
looked
at
different
cells
in
a
tissue
and
they
found
that
they
had
this
heterogeneity
of
you
know
differentiation.
So
there
were
different
ideas.
They
were
either
indifferent
States
or
they
were
like
in
expressing
different
things.
So
even
in
the
so-called
regulative
species,
you
have
a
lot
of
variation.
A
A
A
That's
great
well,
thank
you
for
the
update.
There
are
actually
some
papers
that
I,
so
I
was
showing
Lucas
some
papers
on
some
of
these
things.
So
we
have.
There
are
a
lot
of
papers
on
like
phosphorylation
there's
this
one
protein
kinases
and
phosphatases,
and
the
drosophila
genome.
So
this
kind
of
goes.
A
So
these
are
you
know
these
are
kinases
that
function
in
many
intracellular
signaling
Pathways,
and
they
were
classified
based
on
their
tendency
to
phosphorylate
sites
surrounded
by
basic
amino
acids,
so
they're,
making
some
sort
of
chemical
modification
to
you
know
different
things
in
the
in
the
during
transcription
and
during
protein
formation
that
are
important
for
changing
the
function
a
lot
of
times.
So
you
have
all
these
different
families,
and
this
is
kind
of
how
they've
broken
it
down.
So
they've
actually
made
a
comparison
here
in
table.
A
One
between
fly,
which
is
drosophila
worm,
would
just
see
elegans
and
then
humans,
which
is
the
the
mammalian
counterpart
or
the
mammalian
comparison,
and
so
you
can
see
that
they
have
for
these
different
families
different
numbers
here,
and
so
you
know,
we
don't
know
what
these
these
numbers
mean.
They
don't
really
Trend
in
the
same
direction
all
the
time,
but
they
you
know
it's
it's
a
comparison,
so
we
can
say
well,
there's
a
difference
here
in
this
species
versus
maybe
this
type
of
development
versus
just
maybe
random.
D
A
Can
see
that
you
know
they're
not
all
the
same
and
that
you
have
variation
in
some
of
these
things.
This
is
a
paper
on
phosphorylation
Dynamics
in
human
embryonic
stem
cells.
This
is
the
paper
that
would
Lucas
referenced
when
we
were
talking
about
stem
cells
in
the
rural
phosphorylation.
So
the
reason
I
I
gave
him
all
these
papers
is
because
you
know
a
lot
of
these
proteins
are
involved
in
phosphorylation
and
that
function
is
distinct
from
say,
like
the
cytoskeleton,
but
it's
also
very
important.
A
We
know
what
those
look
like
I
wouldn't
bring
that
data
here
directly
to
this
problem,
but
it's
important
to
remember
that
pluripotent
stem
cells,
this
thing
called
self-renewal,
so
anything,
that's
Cody,
Faulkner
pluripotent
self
renew,
so
they
can
divide
infinitely
and
not
die
off,
whereas
cells
that
are
differentiated
tend
to
have
a
limit
where
they
we
have
a
certain
number
of
Divisions
and
then
they
get
they
go,
maybe
towards
apoptosis.
So
this
is
something
where
you
get
this
software
renewal
aspect
and
then
this
this
is
characteristic.
A
So
we
have
to
go
from
the
self-renewal
phase,
where
the
cell
is
dividing
and
just
self-renewing
versus
a
phase
where
the
cell,
you
know,
goes
through
a
division
or
goes
through
a
cycle
of
DNA
synth,
RNA
synthesis
and
then
you
know,
actually
changes
its
state
or
moves
towards
changing
its
state.
So
we
don't
really
understand
how
this
works
in
terms
of
all
the
mechanisms
we've
people
have
looked
at
different
mechanisms
which
we've
talked
about.
A
A
1399
of
those
were
phosphorylated
and
3067
residues,
which
means
that
they
were
able
to
these
are
out
of
the
ones
they
they
identified.
These
are
the
ones
that
are
phosphorylated.
Approximately
50
percent
of
these
phosphatase
phosphatases
were
regulated
within
one
hour
of
differentiation
induction.
So
this
is
a
quick
process,
and
so
this
is
reviewing
a
complex
interplay,
a
phosphorylation
that
works
spanning
different
signaling,
Pathways
and
kinase
activities.
So
this
is
involved
in
a
lot
of
different
signaling
Pathways.
A
There
are
other
things
cdk1
slash
two.
This
is
something
that
controls
self-renewal.
So
you
know
a
lot
of
these
papers.
They'll
identify
key
genes
or
proteins
that
are
involved
in
some
process.
Sometimes
they'll
also
identify
the
network,
but
usually
you
know
it's
based
on
genes,
because
there's
this
tradition
where
you
would
say:
okay,
this
is
the
thing
that's
responsible
for
this
thing.
Maybe
the
protein
you
go
back
home
the
Gene
and
do
an
experiment,
and
nowadays
that,
of
course
we
have
this
systems
approach
where
we
can
look
at
all
these
things
in
simultaneity.
A
So
you
know
that,
but
but
people
really
haven't
embraced,
like
I'm,
going
to
replicate
this
network,
experimentally
I
mean
I,
don't
even
think
it's
a
lot
of
cases,
it's
really
possible
to
do,
but
so
that's
that
paper
and
then
there's
the
human
phosphoprodium.
So
this
is
a
functional
landscape
of
what
we
see
in
humans.
This
is
where
we
have
all
these.
This
phosphorylation
across
the
human
proteome
phosphorylation
is
a
keep
post-translational
modification
regulating
protein
function,
so
this
is,
after
the
DNA
has
been
turned
into.
A
It's
been
transcribed
and
or
translated
and
then
transcribed.
So
those
are
two
separate
processes
during
translation,
you're,
just
creating
mRNA
and
then
after
transcription
you're
creating
amino
acids
that
then
serve
as
the
building
block
of
the
proteins
and
so
post-translational
modification
is
where
you're
taking
those
chains
of
amino
acids
and
you're
modifying
them,
and
this
is
what
phosphorylation
does
so.
It
modifies
the
function
of
the
protein,
and
so
this
is,
you
know,
they're
involved
in
a
lot
of
things,
but
you
know
you
have
this.
A
The
specific
role
in
differentiation-
and
so
this
is
kind
of
gets-
gives
a
lot
of
information
at
this
very
broad
level.
So
you
know
I,
don't
know
if
there's
a
I,
don't
think
there's
a
paper
on
that
specific
mechanism
with
respect
to
like
see
all
against
differentiation,
but
that's
that's
kind
of
where
we're
headed
actually
in
some
ways,
because
we're
talking
about
the
proteins
and
seeing
which
ones
are
important
final
thing
I
want
to
talk
about
today.
A
You
know
what
is
the
relationship
between
their
lineage
trees
and
I
had
said
that
their
differences
in
the
lineage
trees,
which
is
true
but
there's
a
paper
that
came
out
I,
think
in
2019.
So
it's
not
that
old,
but
this
is
Ralph.
Schnabel
who's
done
a
lot
of
stuff
in
C
elegans
and
some
other
authors,
and
this
is
on
comparing
C,
elegans
and
I
think
three
other
species,
so
this
is
called
20
million
years
of
evolution,
the
embryogenesis
of
four
counter
hepatitis
species
that
are
indistinguishable
despite
extensive
genome
Divergence.
A
So
what
they're?
Looking
at
in
this
paper
as
I've
identified
for
cancer
hebditis
species,
you
have
C
elegans,
C,
Brig,
z,
a
C,
Romani
and
c
brenery,
and
these
are
just
like
different
species
of
that
counterharditis
Quade
or
that
that
genus,
and
so
the
related
evolutionarily,
but
pretty
closely.
But
they
have
different.
You
know
differences
in
their
in
their
lineage
tree.
They
show
more
Divergence
at
the
genomic
level
than
humans
compared
to
mice.
A
A
No
no
yeah
they're
so
like
in
some
drosophila
species,
and
you
know
they're
they
can
breed,
they
actually
can
produce
viable
Offspring,
but
I.
Think
C
elegans
is
pretty
much
C
elegans,
there's
no
I
I,
don't
think
there's
any
evidence
of
them.
Inner
interbreeding
with
other
species
of
cannot
haveditis.
A
However,
the
behavior
and
anatomy
of
the
nematodes
are
very
similar.
So
this
is
where
you
look
at
under
a
microscope.
They
look
very
similar,
but
you
know
this
is
something
that
we
know
that
there's
a
lot
of
genomic
Divergence.
So
we
present
a
detailed
analysis
of
the
embryonic
development
of
these
species
using
4D
microscopy
analysis
of
embryos,
including
lineage
analysis
terminal
differentiation
patterns
and
bioinformatics
quantifications
of
cell
Behavior.
A
Based
on
our
results,
the
embryonic
development
of
all
four
species
is
nearly
identical,
so
this
is
where
they
basically
have
the
same
pattern
of
Divisions
and
binary
divisions
and
different
tissue
formations,
and
things
like
that
suggesting
that
an
apparently
optimal
program
to
construct
the
body
plan
and
nematodes
has
been
conserved
for
at
least
20
million
years.
So
these
four
species
are
related
at
the
level
of
like
20
million
years
of
evolution,
so
human
and
mouse
are
probably
around
that.
A
It's
you
know
it
depends
on
who
is
doing
the
you
know
the
the
phylogeny,
but
it's
it's
comparable
to
humans
and
mice
in
in
mammals.
So
you
know
this
is
a
pretty
deep
for
for
related
species.
You
know
in
the
same
genus
it's
pretty
deep
history,
so
this
contrasts
to
the
levels
of
Divergence
between
the
genomes
and
the
protein
orthologs
of
the
counter
hepatitis
species.
So
we
have
these
protein
orthologs
that
we
can
look
at
so
they're,
not
all
the
same
name,
they're,
not
all
the
same.
A
You
know
some
of
them
are,
you
know,
duplicates
and
then
they've
diverged,
so
they're
different
I've.
Never
I
I
have
had
to
get
back
into
sort
of
some
of
this
like
genomics
and
proteomics
language,
so
they're,
orthologs
and
they're
paralogs
and
then
they're
homologs,
and
so
those
three
terms
are
not
the
same
and
they're
just
different
ways
of
looking
at
these
relationships.
I
can't
remember
what
orthologs
means,
as
opposed
to
like
homologs
and
paralogs,
but
so.
A
A
So
you
have
like
the
screen
and
this
purple
and
those
are
going
to
those
that
purple
cell
is
going
to
give
rise
to
a
bunch
of
cells
that
have
like
all
have
like
a
fate.
That's
either
like
in
that
tissue
type
or
a
series
of
tissue
types
that
are
related.
So
they
look
at
the
founder
cells
and
this
place
here
and
then
256
cell
stage
here.
So
you
see
a
much
denser
cloud
and
you
see
these
arrows
where
they
show
kind
of
the
different
colors.
A
A
A
A
The
a
actually
is
what
they
call
the
ox
distance
map,
which
I
don't
know
exactly.
Oh,
this
is
only
displays
distances,
differences
along
the
anterior
posterior
axis,
where
the
x-axis
of
the
embryo,
so
we've
done
stuff
like
this,
where
you
take
like
two
Dimensions
versus
three
dimensions
and
you
look
at
their
distance
and
this
is
not
the
one
lineage
tree,
but
this
is
like
a
distance
map,
so
we
do
this.
We've
done
this
for
different
things
and
then
they
just
show
this
across
species.
A
It
reflects
all
64
combinations
of
the
founder
cells,
the
eight
or
the
eight,
a
B
cell
aeb
founder
cells,
so
it's
just
the
a
b
sub
lineage
and
then
their
distance
here.
So
they
have
this
distance
Matrix,
and
this
is
kind
of
like
you
know,
just
kind
of
shows
you
that
they're
in
different
positions
in
the
embryo.
A
So
this
is
the
the
lineage
tree
in
this
case
they're
working
from
the
the
a
b
sub
linear,
which
is
the
anterior
and
and
they're
just
kind
of
looking
at
that.
So
I
don't
know
if
they
have
any
other
figures
in
here
worth
talking
about.
Yeah
I,
don't
think
they
have
any
lineage
trees
in
here
mapped
out.
A
This
is
where
the
cell
is
differentiating
into
the
V6
or
hypodermal
cell.
So
this
is
basically
looking
at
cell
cycle
length
and
time.
So
the
idea
is
that
you
know
it
takes
a
certain
amount
of
time
for
a
cell
to
divide
into
its
daughter
cells,
and
so
an
ABA
cell
divides
into
an
AVL
and
an
Abal
and
an
abap
or
something
like
that,
and
it
takes
a
certain
amount
of
time.
A
And
so
the
idea
is
that
between
species
that
time
can
shrink
or
expand,
and
so
that
has
an
effect
on
when
that
cell
is
produced.
And
if
that's
the
timing
is
off
it
can.
You
know,
maybe
be
like
in
a
mutant
where
there's
a
mutation
that
affects
the
timing
of
that
division.
It
can
become
non-functional
or
it
can
do
something
else,
because.
A
D
A
A
Someone
who's
worked
on
this
extensively,
but
this
is
a
couple
of
papers
where
they
talk
about
genotyping
and
type
mappings
more
generally,
and
then
they
talk
about
in
the
context
of
early
life.
So
the
first
paper
is
on
from
genotypes
to
organisms,
state
of
the
art
and
perspectives
of
a
Cornerstone
in
evolutionary
Dynamics
and
they're,
a
whole
host
of
authors.
Here.
A
B
A
Of
a
physics
of
Life
take
on
it,
the
abstract
reads:
understanding
how
genotypes
map
onto
phenotypes,
Fitness
and
eventually
organisms
is
arguably
the
next
major
missing
piece
in
a
fully
predictive
theory
of
evolution.
So
those
people
are
kind
of
taking
the
extended
synthesis
approach
to
evolutionary
theory.
But
in
general
you
know.
The
notion
of
Fitness
in
some
of
the
details
are
quite
controversial
in
terms
of
how
Fitness
is
measured
and
how
it's
heritable
and
so
forth
also
note
that
they
talk
about
phenotypes
and
fitness
and
then
organisms
as
separate
things.
A
So
a
phenotype
is
not
an
organism
per
se.
The
fitness
is
not
an
organism,
they're
all
totally
different
things
and
we're
going
to
get
into
how
those
are
related.
So
we
generally
refer
to
this
as
the
problem
of
the
genotype
to
phenotype
map.
So
if
we
move
into
conceptual
land
for
a
minute-
and
we
talked
about
this
in
the
meeting-
we
have
these
networks
of
changing
interactions
and
protein
protein
interactions,
and
we
can
collectively
call
this
the
genotype.
A
A
So
but
this
whole
concept
of
G
or
the
genotype
is
quite
broad.
It
could
be
any
set
of
genes
in
the
genome.
It
could
be
a
specific
set
of
genes.
A
specific
circuit
mapped
to
something
else:
the
map
to
some
function.
It
could
be
the
entirety
of
the
genome,
mapped
to
the
entirety
of
the
phenotype.
And
so
then
they
say
they
talk
about
P,
which
is
the
phenotype,
and
so
this
is
the
phenotype.
Again,
it's
a
very
broad
network
of
things.
A
It's
not
necessarily
the
organism,
because
the
organism
is
going
to
be
both
G
and
P
and
some
other
things
as
well.
So
this
is
not
the
organism,
it's
just
the
phenotype
and
one
other
thing
you
know
with
phenotypes.
We
basically
have
to
Define
them,
as
maybe
like
a
body
or
it
could
be
like
a
behavior,
even
but
usually
people.
Think
of
you
know:
well
you
think
of
the
behavior
as
a
phenotype,
I
guess,
but
again
you
know
you
have
the
phenotype
you
have
the
genotype
and
the
organism
which
is
the
bridge
between
these.
A
So
all
these
different,
you
know
there's
this,
there's
this
old
idea
that
one
gene
equaled
one
protein
equaled
one
piece
of
the
phenotype,
but
that's
just
simply
not
the
case.
We
have
these
Gene
Gene
interactions,
which
are
this
is
this
network
down
here
in
the
G
area,
and
then
we
have
GE,
which
is
where
we
have
influences
of
environment,
which
we're
not
putting
into
this
diagram
right
away.
A
Okay!
So
that's
what
they're
getting
at
now
this
this
this
sort
of
this
bi-level
Network.
As
you
see,
between
G
and
P.
This
is
something
we've
seen
in
epigenetic,
Landscapes
or
waddington's
Landscapes.
If
we
look
at
the
pictures,
I
don't
have
a
picture
right
now,
but
where
you
look
at
the
underside
of
the
landscape
and
you
see
all
these
lines
coming
out
of
little
cubes,
that's
his.
C
A
A
This
genotype
phenotype
Mac
we're
still
far
from
achieving
a
complete
picture
of
these
relationships.
Our
current
understanding
of
simpler
questions,
such
as
a
structure
induced
in
the
space
of
genotypes
by
sequences,
mapped
to
the
molecular
structures,
as
revealed
important
facts
that
deeply
affect
the
dynamical
description
of
evolutionary
processes,
so
they're
looking
at.
A
With
something
with
Dynamics,
so
they're
almost
looking
as
a
dynamical
system,
but
they
don't
think
they're
cool,
maybe
they're,
not
quite
going
there
and
whatever
empirical
evidence
supporting
the
fundamental
relevance
of
features
such
as
phenotypic
bias
is
mounting
as
well
yeah.
Well,
the
synthesis
of
conceptual
and
experimental
progress
leads
to
questioning
current
assumptions
and
the
nature
of
evolutionary
Dynamics
cancer
progression
models
are
synthetic
biology,
approaches
being
notable
examples
so.
D
A
This
work
delves
in
with
a
critical
and
constructive
attitude
into
our
current
knowledge
of
how
genotypes
map
onto
molecular
phenotypes
and
organismal
functions
and
discusses
theoretical
and
empirical
Avenues
to
broaden
and
improve
this
comprehension
as
a
final
goal.
This
community
should
aim
at
deriving
an
updated
picture
of
evolutionary
processes,
suddenly
relying
on
the
structural
properties
of
genotype
spaces,
as
revealed
by
modern
techniques
and
molecular
and
functional
analysis.
A
So
it's
interesting
that
they
mentioned
molecular
phenotypes
because
in
the
work
of
Gunter
Wagner,
which
we
will
also
won't
talk
about
right
now,
he
has
this
model
of
genotype
to
phenotype
mapping,
which
is
basically
a
DNA
sequence,
mapping
to
an
RNA
structure,
so
RNA
structures
are
basically
a
transcription
of
the
phenotype
or
of
the
genotype
to
a
phenotype
that
has
some
sort
of
secondary
structure.
So.
C
A
Is
a
loop:
this
is
not
based
in
anything
in
the
real
world
necessarily,
but
this
is
basically
your
RNA
molecule.
This
is
your
DNA
molecule.
They
have
these
hairpin
turns
and
other
types
of
secondary
structure.
It's
not
the
secondary
structure
of
a
protein
which
is
much
more
complex
but
they're
different,
secondary
structures
that
exist,
and
so
you
map
the
DNA
sequence
to
the
RNA
sequence.
And
then
you
have
the
secondary
structure
as
well.
So
you
have
this
genotype
and
you
have
this
phenotype
and
the
RNA
phenotype.
A
Of
course,
then
that
leads
us
to
an
analysis
using
hypercube
in
terms
of
a
neutral
space.
So
we
can
look
at
the
evolution
of
this
RNA
phenotype
through
the
analysis
of
a
neutral
space
and
I'll
just
use
a
cube
for
illustrative
example
purposes
here,
but
basically
we
have
this
Cube.
We
have
I,
think
it's
well
I,
think
we'll
have
three
states
here
and
three
states
here.
A
So
this
this
mapping
will
map
mutations
and
dna-based
mutations
in
RNA
space,
and
then
you
can
have
further
mutations
in
the
phenotype
and
or
in
the
genotype
that
that
transfer
to
the
phenotype
to
lead
you
across
the
space.
And
so
when
we
say
neutral
we
mean
that
there's
no
selective
Force
but
of
course,
selection
can
always
act
on
this
to
guide
it
down
the
shortest
path.
In
this
case,
every
path
is
equivalent,
so
selection
doesn't
necessarily
help
you.
A
So
every
path
is
like
three
steps:
three
mutational
steps,
and
that's
you
know
something
that
if
you
had
a
more
complex
space
here
with
more
you
know
with
shorter,
clearly
shorter
paths
and
clearly
longer
paths,
then
selection
might
help
you
get
there
faster.
But
in
this
case
it's
the
neutral
space,
because
every
possible
path
has
the
same
length
same
number
of
mutational
steps.
A
So
this
is
the
abstract
of
this
paper.
This
basically
talks
about.
You
know
this
relationship
between
genotyping
phenotype
and,
of
course,
if
you
have
an
RNA
molecule
as
your
phenotype,
that's
great
that's
convenient,
but
a
lot
of
real
world
phenotypes
are
a
lot
more
complicated
than
that
they're
complex
they're,
not
discrete.
They
have
a
lot
of
discontinuities.
A
There
are
a
lot
of
things
in
between,
say,
like
a
DNA
sequence
and,
like
an
embryo,
a
mammalian
embryo,
say,
there's
a
lot
of
stuff
in
between
there
that
we
need
to
put
into
that
model,
so
we
might
have
say,
for
example,
a
genotype
to
prodium
type,
to
structure
type
to
phenotype
map,
or
something
like
that
you
know
we
can
have
these.
These
models
could
be
several
levels
deep.
So
this
is
something
I've
been
thinking
about
with
respect
to
some
of
the
things
that
we're
talking
about
in
the
meeting.
A
So
we
have
this
genotype
here,
which
of
course,
is
just
a
kind
of
a
nondescript
wake
with
a
network
in
it
right.
We
don't
really
know
what
this
network
looks
like
it's
just
kind
of
like
interactions
or
Gene,
Gene
and
Gene
environment
interactions.
Here
we
have
the
environment,
my
environmental
effect
here
then
we
might
have
well.
We
don't
have
that
Arrow
we
have
a
proteomic
or
maybe
an
RNA
phenotype.
A
A
You
have
two
different
forms
of
complexity,
then
you
go
to
your
proteins
or
you
know
that
has
like
a
lot
of
interactions,
protein
protein
interactions
and
then
finally
go
up
to
the
top
level,
which
is
the
phenotype
which
you
know
you
have
all
the
sort
of
the
interactions
that
occur
at
these
different
levels.
You
have
transfers
upwards
and
then
those
will
be
sort
of
summarized
in
the
phenotype
and
again
this
is
going
to
be
a
hard
sell,
because
we
don't
know
what
the
contributions
are.
We
don't
know
what
maps
to
what
you
know.
A
A
Is
the
sort
of
logic
of
this
approach,
so
they
kind
of
talk
about
GP
Maps
as
if
they're
easy,
sometimes,
but
in
a
broader
ecological
context,
the
way
in
which
generic
properties
of
the
GP
map
shape
adaptation
have
been
really
explored.
So
this
interaction
between
the
environment
and
the
genotype
or
the
environment,
and
we
could
be
the
RNA
ohm
or
the
proteome
or
even
the
environment
phenotype.
There
have
been
a
lot
of
arguments
in
Evolution,
for
example,
the
level
of
selection.
A
Is
it
at
genes
or
is
it
a
phenotype
or
where
is
it
for
the
sake
of
argument?
I'm
just
going
to
say
it's
at
the
level
of
genes.
I
think
that's
a
good
way
to
sort
of
characterize
it
because
the
phenotype
it's
extremely
hard
to
know.
You
know
what
the
effects
of
selection
look
like
when
you're
selecting
on
a
phenotypic
trait.
So
if
you
were
to
domesticate
dogs-
and
you
were
to
read
for
specific
traits-
you're-
not
breeding
on
the
phenotypic
traits
per
se,
the
phenotypic
traits
are
the
outcome.
You're
actually
manipulating
the
genotype.
A
D
A
Is
where
you
know
you,
you
kind
of
get
into
this
issue
of
you
know
like
artificial
selection
and
what
are
the
outcomes
of
artificial
selection?
So
it's
a
very
interesting
set
of
issues,
so
they
taught
this
paper
is
really
long
and
I'm
not
going
to
get
into
the
whole
paper.
This
review
has
a
number
of
Parts
there's
the
introduction,
there's
section
two,
which
puts
in
the
perspective
how
the
relevant,
how
relevant
the
generation
of
variation
is
in
the
evolutionary
process
and
introduces
important
biases
arising
from
the
inherent
structure
of
genotype
spaces.
A
The
second
part
comprises
sections
three
to
six.
If
we
discuss
conceptual
approaches,
the
static
properties
of
GP
maps
and
their
dynamical
consequences,
so
these
GP
maps
are
both
static
and
dynamic.
If
we
look
at
like
this,
it's
pretty
much
a
static
representation,
I'm
showing
you
the
levels,
but
there
are
also
dynamical
systems
in
the
sense
that
they
evolve
and
that
they
evolve.
Sometimes
they
evolve
independently.
A
A
A
We
will
succinctly
present
computational
GP
Maps,
so
this
has
been
done
in
computer
science
and
a
life
and
in
genetic
algorithms.
Quite
a
bit.
They've
worked
out
from
sort
of
a
genomic
representation
of
a
problem,
so
they'll
take
a
problem
and
they'll
encode
it
as
a
binary
set
of
binary
strings
and
we'll
treat
that
as
a
genotype
and
that
genotype
is
expressed.
A
So
certain
bits
are
turned
on
certain
bits
are
turned
off
and
that
results
in
a
phenotype,
and
so
you
can
have
these
computational
maps
that
lead
to
certain
phenotypic
outcomes,
and
then
this
is
of
course,
can
lead
to
topological
properties
of
the
space
of
genotypes.
So
you
can
actually
look
at
a
genetype
space.
You
know
all
the
different
states,
so
it
looks
a
lot
like
this
Cube,
which
I
showed
in
seven
slide.
Seven.
A
They
give
a
mean
field
description
that
incorporates
Essentials
of
GP
map
topology
that
clarify
major
dynamical
features.
Then
it
finishes
with
a
derivation
of
equilibrium
properties
in
the
context
of
statistical
mechanics
and
some
applied
examples,
and
then
section
6
discusses
the
evolution
of
GP
Maps
themselves.
By
means
of
two
examples:
one
is
a
scenario
where
a
multifunctional
quasi-species
emerges,
which
is
something
that
you
see
in
this
is
like
evolutionary
Dynamics
jargon
and
I'm.
A
A
A
So
this
is
something
that,
if
you're
interested
in
this
area,
it's
a
good
paper
to
read
through,
even
if
you're,
not
I,
think
it's
good
to
think
about
some
of
these
problems
and
Stephen
J
Gould,
of
course,
thought
about
these
problems
so
he's
basic,
but
he's
who's
very
critical
of
the
Viewpoint
that
a
large
part
of
the
variation
in
organisms
is
isotropic.
In
other
words,
it's
not
biased
in
One,
Direction
or
another.
So
Stephen
J
gold
was
critical
of
this
Viewpoint
and
expressed
it
as
follows.
A
Variation
becomes
raw
material
only
and
as
an
isotropic
Sphere
for
potential
about
the
model.
Modal
form
of
a
species,
only
natural
selection
can
manufacture
a
substantial
directional
change.
So
this
is
something
like
we're
talking
about
with
the
bias
in
these
models,
so
you
could
just
have
a
genotype
to
phenotype
model
where
one
gene
mapped
to
one
piece
of
the
phenotype.
A
So
this
is
something
that
you
know:
it's
not
necessarily
directed
evolution
in
the
lamarckian
sense,
but
it's
this
sort
of
bias
that
results
from
these
different
levels
being
pushed
in
different
directions.
Sort
of
almost
I
can
think
of
it
as
path
dependence,
maybe
rather
than
directional
evolution.
A
But
the
outcome
is
looks
like
something
maybe
like
directional
evolution,
so
they're
different
models
that
people
have
introduced
me
Maynard,
Smith
introduced
and
the
notion
of
mapping
from
a
genetic
space
to
a
molecular
structure.
I
think
we
talked
about
that
a
couple
weeks
ago
in
a
classic
paper
and
the
ideas,
a
network
that
links
viable
genotypes.
A
This
is
a
resolution
of
an
evolutionary
Paradox
pointed
out
by
Salisbury,
where
the
number
of
possible
amino
acid
sequences
exceed
by
many
orders
of
magnitude
the
number
of
proteins
that
has
ever
existed
on
Earth
since
the
origin
of
life.
So
you
need
to
have
a
subset
of
those
amino
acid
potential
amino
acid
sequences.
The
information
is
in
restricting
that
to
a
certain
alphabet
or
grammar,
and
so
that's,
and
so
how
does
it
happen
in
evolution?
A
And
so
this
is
kind
of
the
attempt
to
make
this
non-limarkian,
and
so
this
is
where
this
comes
from
part
of
it
as
well.
They
talk
about
neutral
networks,
which
is
this
Ensemble
of
connected
genotypes,
with
the
same
Fitness
and
as
I
saw
with
the
RNA
could
be
a
phenot,
an
RNA
phenotype,
including
those
with
identical
phenotypes.
So
you
can
actually
have
set
up
your
neutral
Network
as
a
series
of
genotypes
that
map
to
phenotypes
that
look
the
same
so
they're
these
phenocopies
of
each
other,
but
they
have
different
underlying
genotypes.
B
A
A
Then
we
have
these
simple,
effective,
GP
models,
which
is
where
you
just
have
like
these
strings
that
are
expressed
to
different
points.
You
have
stop
codons
at
different
points
that
truncate
the
sequence
in
different
places,
or
you
have
this
with
actual
nucleotide
bases,
so
you
can
actually
edit
them
in
different
ways.
The
polynomial
polyomia
Amino,
it's
like
a
domino
but
a
PO
yeah.
So
this
is
where
you
have
these
different
configure,
geometric
configurations
with
different
sides
faces
of
different
values,
and
you
connect
them
together
in
a
way.
That's.
C
A
And
that
kind
of
restricts
that
possibility
space
down,
and
then
you
can
examine
what
kinds
of
sequences
you
get
from
that.
So
not
every
configuration
is
allowed
in
the
game,
so
you
have
to
fix
I
guess
you
have
to
interact
sequential
numbers,
so
one
fifths
to
two
two
fits
or
I
guess
two
fits
to
three.
We
don't
have
that.
We
have
a
three-phase
fitting
Bill
four
face
and
a
two-phase
fitting
dual
one
face.
So
there's
a
transition
here
you
can't
combine
non-likes,
so
you
can't
combine
a
zero
and
a
zero.
A
So
you
end
up
with
a
subset
of
these
shapes
and
you
can
characterize
the
sequence
here
is
the
genotype,
and
this
is
the
phenotype
and
then
biomorphs,
which
are
these
models
G.
This
is
Doc
Richard
Dawkins
biomorphs,
where
it's
a
genotype
with
a
few
parameters
that
Define
the
generative
rules
of
the
structure.
So
you
take
this
genotype
on
the
right.
You
can
generate
these
phenotypes
on
the
left.
So
then
we
have
genotype
networks.
This
is
generated
by
assigning
viable
genotypes
at
random,
with
probability
with
a
certain
probability
to
Binary
sequences
of
length.
A
L
equals
eight,
so
a
certain
length.
A
certain
probability,
this
shows
a
network
of
different
genotypes,
says
actually
shows
like
lethal
genotypes
here,
viable
genotype,
saying,
green
and
then
recombination
centers
in
blue.
So
you
can
connect
genotypes
together
and
Network.
This
shows
kind
of
how
this
happens
in
different
simulations
and
different
mutation
rates
and.
A
A
Networks
or
something
I
can't
remember
the
name
right
now,
but
one
important
motivation
for
these
type
of
networks
has
been
the
finding
that
robustness,
synchronization
or
cooperation
lead
to
different
behaviors
when
studied
and
isolated
or
in
interconnected
networks.
So
if
you
connect
things
together,
if
you
connect
networks
together
into
their
own
Networks,
you
can
actually
look
at
things
like
synchronization
and
cooperation,
whereas
if
you
just
studied
interconnected
networks,
one
versus
the
other,
you
wouldn't
be
able
to
understand
those
things.
A
The
main
reason
for
the
change
in
perspective
has
been
to
realize
that
many
natural
systems
Beyond
displaying
a
network-like
organization,
are
also
made
of
interacting
and
competing
networks
of
very
different
scales.
So
we
have
these
different
networks
in
our
organism.
We
have
molecular
networks.
We
have
you,
know
protein
protein
protein
interactions,
Gene
Gene
interactions.
We
have
molecular
interactions
and
when
we're
building
so
I
can
tissue
or
some
sort
of
phenotype,
we
have
organismal
organismal
networks.
A
So
those
are
those
all
those
networks
are
embedded
in,
an
organism
which
is
then
embedded
in
and
it
works.
So,
to
get
all
that
all
these
net
levels
of
network,
we
need
to
have
the
same
type
of
model
that
we
have
for
our
phenotype
to
genotype
mapping.
So
we
have
this
these
sort
of
layers
and
it's
vertically
organized-
and
this
is
one
of
the
reasons
why
we
do
this
because
it
allows
us
to
get
at
some
of
these
interactions.
A
So
this
is
a
multi-level
Network,
as
opposed
to
just
like
single
singular
networks
when
I
said
meta
networks
before
I'm
at
multi-level,
Networks,
anyways.
So
the
extent
to
which
network
science
can
foster
a
knowledge
and
comprehension
of
the
evolution
and
adaptation
of
heterogeneous
populations
in
an
ever-changing
biosphere
is
a
relevant
open
question.
In
particular,
the
theory
of
competing
networks
can
be
used
to
analyze
the
evolutionary
dynamics
of
populations
and
its
basic
genotypes
that
can
be
regarded
as
a
network
of
Networks.
A
So
when
this
Viewpoint
population
evolution
is
described
as
a
competition,
competition
of
resources
for
resources
of
a
certain
kind
where
the
competitors
or
whole
networks
instead
of
independent
nodes.
So
again,
you
know
if
we
think
about
like
a
model
of
population
genetics
where
we
have
a
single
node,
an
organism
being
that
single
node
or
a.
A
Being
a
single
node,
we
can
actually
think
of
that
node,
then,
as
this
network
within
that
node.
So
the
network
changes.
If
there's
mutation
in
that
Network
or
if
there's
mutation
in
that
node,
we
can
represent
it
more
clearly.
We
can
really
understand
kind
of
what's
going
on
inside
the
node
and
then,
of
course,
you
know
understand
how
competition
unfolds,
so
this
is
kind
of
what
they're
doing
here.
A
So
finally,
I'd
like
to
talk
about
virtual
cells
and
evolutionary
Dynamics
in
the
RNA
World
model,
described
here
in
molecular
phenotypes,
were
determined
for
pairs
of
complementary
sequences,
positive
and
negative
strands
based
on
specific
structure
motifs
under
Revolution
the
following
phenotypes
emerge.
So
if
replicases
parasites
helpers,
installers
and
junks,
so
these
are
different
classes
phenotypes.
This
is
something
you
kind
of
find
in
Cellular
automata
in
The,
Game
of
Life,
where
you
see
colliders
and
other
signs
of
structures.
A
So
this
is
not
something
that
this
is
very
compatible
with
that
approach,
but
they're
also
building
these
virtual
cells
and
so
they're
trying
to
get
a
sense,
I,
think
of
the
virtual
cell
and
what
it
looks
like
that
this
is
at
the
origin
of
life,
so
you're
trying
to
build
different
classes
of
things
that
do
different
that
will
fit
into
a
sort
of
a
functional
Network
within
the
cell.
So
here
you
have
a
diagram,
so
this
is
virtual
cell
and
evolutionary
Dynamics.
A
All
right
here
is
the
scheme
of
a
virtual
cell.
So
you
have
some
sort
of
ion
pump.
That's
pumping
things
into
the
cell,
you
have
these
different
reactants
in
the
middle
of
the
cell.
You
have
the
end
product
and
then
you
have
transcription
factors
that
bind
to
genes
and
express
them,
and
so
this
is
a
whole
sort
of
metabolic
cycle
within
the
cell.
This
shows
the
genome
size
and
gray.
So
this
is
the
gray
function.
A
You
know,
underneath
the
function
is
the
gray
and
then
this
red
function,
which
is
showing
common
ancestry
Through
Time
Redline,
is
the
average
fitness
in
three
standard
environments.
So
this
is
the
fitness
versus
The
genome
size,
shaded
area
gray
area
predicts
genome
size,
showing
initial
genome
inflation,
which
is
here
followed
by
a
streamlining
of
her
Gene
loss.
A
So
maybe
you
know
what's
happening
there
is
that
you
have
a
lot
of
genes
and
they're
encoding
for
a
lot
of
things
in
the
virtual
cell,
a
lot
of
different
components
and
then
those
maybe
get
reduced
as
it
becomes
more
efficient.
So
there
are
things
that
are
maybe
shared
genes
and
they're
expressed
in
different
ways.
A
This
leads
us
from
an
initial
Fitness
distribution,
the
Shaded
gray
area,
to
a
more
pronounced
u-shape,
which
is
this
fraction
of
mutants
under
this
residual
Fitness
fraction.
So
it
kind
of
goes
down
and
then
up
with
from
going
from
zero
to
one
as
a
residual
Fitness
reaction,
so
I
guess
the
residual
Fitness
for
Action
is
neutral.
Mutations
are
one
deleterious.
Mutations
are
zero
if
I'm
understanding
this
graph
correctly,
like,
as
you
can
see
some
of
the
evolutionary
Dynamics
here
now
onto
the
paper
on
early
life.
A
So
this
is
Christoph
hadami,
Natasha,
CG
and
they're
from
Michigan
State,
and
they
do
a
lot
of
stuff
with
stuff
does
a
lot
of
stuff
with
physic
biophysics
and
a
life,
especially
like
the
Evita
platform.
He
was
one
of
the
people
to
develop
that
so
this
kind
of
talks
about
this,
this
original
paper
that
we
talked
about
previously
so
in
the
Target
article,
the
officer
brought
overview
of
computational
and
theoretical
investigations
into
the
nature
and
properties
of
complex
fitness
landscapes.
A
Fitness
Landscapes
are,
as
the
authors
emphasize,
throughout
complex
maps
and
Regina
type
space
to
a
phenotype
space.
This
map
is
necessarily
complex
because
for
the
most
part,
there
are
many
more
distinct
genotypes
in
a
sense
exponentially
more
in
their
existing
phenotypes.
So
there
are
a
lot
of
genotypes
that
exist
that
might
be
inviable
as
phenotypes
or
you
have
these
phenocopies,
where
the
phenotype
is
essentially
the
same,
but
they're
different
genotypes
underlying
them.
So
you
have
like
different
types
of
neutral
mutations
that
don't
contribute
to
the
phenotype.
A
A
Makes
this
map
even
harder
to
where
we
can
understand?
One
consequence
of
this
feature
of
the
genotype
phenotype
map
is
that
the
fraction
of
functional
sequences
does
it
have
a
particular
phenotype
is
very
small
compared
to
the
number
of
possible
phenotypes.
In
fact,
this
density
of
functional
sequences
can
be
used
to
quantify
the
information
content
of
the
sequences
is
pointed
out
by
a
showstack.
A
You
define
the
functional
information.
The
molecule
has
about
a
particular
phenotype
e
in
terms
of
the
fraction
of
sequences,
f,
d,
f
e
that
carry
the
phenotype
at
level
or
higher
as
this.
So
this
is
the
information,
and
then
this
is
log
F.
So
this
is
basically
an
information
metric
on
this.
So
it's
a
fraction
of
sequence
as
a
carry
the
phenotype
and
there's
a
threshold
for
the
phenotype
being
carried
so
it
has
to
be
like
you
know,
the
phenotype
might
be
cryptic
or
it
might
be
very
distinct.
A
It
has
to
be
at
a
certain
level
and
then
so.
This
just
shows
us
information
measure
that
they're
using
for
this.
It's
basically
negative
log,
so
that
tells
us
something
about
the
information
and
then,
if
the
sequences
are
written
in
an
alphabet,
size
d,
so
that
means
like
the
bases
that
you
have
so
for
proteins.
You
have
20
bases
right.
You
have
20
amino
acid
types
that
can
be
combined
or
in
DNA.
You
have
D
equals
four.
You
have
four
nucleotides.
A
So
this
is
how
this
is
calculated
and
limited
to
sequence:
length
L,
then
n
equals
D
to
the
power
of
L
and
then
that's
our
information
measure,
so
we're
just
multiplying
out
into
the
into
the
the
information
n
actually
is
yeah,
so
n
equals
D
to
the
core
of
L.
So
you
have
log,
D
and,
and
then
L
minus
that
and
then
that's
your
information.
A
So
then
I'll
log
n.
Is
this
the
micro
canonical
or
coarse
grained
approximation
of
the
Shannon
entropy?
So
we've
seen
this
in
Shannon
entropy.
Basically,
it's
a
similar
thing,
an
approximation
of
Shannon
information
content
by
molecular
sequence.
So
a
lot
of
times
people
will
calculate
the
Shannon
entropy
of
some
sequence
and
then
use
that
to
characterize
the
genotype
in
its
information
content,
and
the
chistax
measure
is
an
as
a
convenient
way
to
quantify
how
much
information
about
performing
task
e
is
stored
within
a
sequence.
A
So
this
gives
us
a
good
measure
of
this
genotype
to
phenotype
mapping
and
so
then
they're
applying
this
to
a
Vita,
and
this
is
where
you're
building
the
sub
self-replicating
programs
that
have
a
certain
instruction
length.
The
instruction
length
can
be
reduced
to
these.
The
symbolic
diversity
that
you're
going
to
measure
with
this
information
picture.
A
So
using
the
number
for
n
and
two,
which
is
this
equation
up
here
and
then
using
the
phenotype
Colony
forming,
is
the
threshold,
so
it
has
to
have
a
colony
forming
attribute,
and
you
know
whatever
that
looks
like
I
guess
you
could,
in
a
computer
program,
Define
a
colony
forming
phenotype
very
precisely,
reveals
an
information
content
of
5.91
mirrors.
So
it's
like
a
mirror.
Is
this
like
the
the
number
of
characters?
So
basically
it's
5.91
characters
for
this
information
that
has
the
information
content
of
the
phenotype.
A
So,
basically
in
you
know
in
Evita,
this
is
very
it's
a
very
easy
way
to
do
this.
You
have
an
instruction
set
that
produces
sort
of
these
behaviors
or
these
competencies.
You
then
have
this.
They
can
be
reduced
to
a
sequence,
then
then
maps
to
some
behavior
in
the
program,
and
then
here
we
have
this
Colony
forming
phenotype
that
we
can
evolve
and
then
once
we've
evolved
it
we
can
characterize
a
Genet
genotype
in
terms
of
what
instruction
set
is
needed
for
a
minimal
instruction
set.
A
A
So
this
shows
a
network
of
this
largest
cluster
containing
or
94
of
the
replicators
for
the
L
equals
nine
landscape.
With
the
edges
drawn
between
one
mutant,
Neighbors
graph
consists
of
five
sparsely
connected
groups
of
tightly
connected
clusters,
as
you
can
see
here,
and
so
then
the
so
this
is
different
calculation.
A
So
the
information
content
of
the
replicators
is
5.77
merged.
Using
the
same
equation.
Replicators
differ
vastly
in
the
robustness
of
replicators,
are
sort
of
the
vehicles
for
replication.
It's
a.
They
differ
vastly
as
measured
by
the
number
of
one
mutant
neighbors
that
are
also
replicators.
Well,
the
most
robust
replicator
is
74
viable
one
mutant,
neighbors,
so
they're
using
this
network
analysis
to
understand
like
what
the
effective
replicators
are,
is
their
self-replicating.
You
can
measure
these
sorts
of
lineages.
A
We
can
test
how
information
is
encoded
in
replicator
I
by
analyzing
the
density
of
replicators
and
mutational
steps
away
from
that
sequence.
So
this
gives
us
this
equation
here
and
then
in
equation.
Five
there's
a
number
of
viable
replicators
mutation,
depth,
K
away
from
the
target
sequence
pi.
So
we
go
back
to
this.
This
Cube
or
this
hypercube
analog
that
I
mentioned
before
you're,
basically
trying
to
figure
out.
A
So
this
this
expression
is
the
number
of
viable
replicators
at
muted
Asian
depth,
K
away
from
the
target
sequence,
I.
Okay,
so
then
there's
the
number
of
this
denominator
here
is
the
number
of
possible
units
up
to
distance
n,
so
that
function
p,
I
n,
is
the
density
of
replicator.
Is
a
distance
up
to
n
from
sequence
I,
so
you're,
counting
up
the
density
of
replicators
across
the
diversity
of
sequences
plotting
this
information
against
n
reveals
that
information
is
encoded
within
the
sequence.
A
So
graphs
here
see
that
this
is
the
fraction
of
functional
sequences
on
the
y-axis.
The
mutational
depth
on
the
x-axis-
and
you
can
say,
is
mutational
depth.
Increases
fraction
of
functional
sequences
was
down
to
negative
six,
but
I
think
that's
that's
the
log.
So
yeah,
that's
so
it's
going
in
this
downward
slope.
A
The
red
line
is
a
theoretical
limit,
so
nothing
should
be
below
that
and
then
these
black
lines
are
the
different
examples
that
they
got
from
the
primary.
The
red
line
indicates
that
density
Decay
if
sites
do
not
interact
no
epistasis,
so
epistasis
are
these
Gene
Gene
interactions
and
this
most
of
them
are
exhibiting
epistasis.
So
you
see
them
above
the
red
line.
A
A
So
finally,
we
go
back
to
the
first
paper.
We
can
talk
about
some
of
these
open
problems
and-
and
there
are
a
lot
of
them,
listed
I'm
not
going
to
go
through
all
of
them
in
the
interest
of
time,
but
I'm
going
to
talk
about
a
few,
so
one
of
them
is
9.1.
So
if
we
understand
this
GP
Mac
architecture,
we
might
ask
the
question:
is
it
Universal?
A
So
we
might
ask
the
question:
is
it
always
a
matter
of
mapping
from
genotype
to
phenotype,
and
we
saw
in
the
mathematical
formulations
that
there
is
a
information
content,
the
genotype
that
needs
to
be
carried
over
to
the
phenotype
to
make
it
viable?
We
also
know
that
certain
combinations
are
by
there.
We
were
biased
towards
certain
combinations
of
genotype
to
produce
a
phenotype
and
only
those
certain
combinations
that
are
favored
Maybe
by
viability,
Criterion
or
by
selection,
or
by
neutral
processes
or
or
you
know,
development
are
going
to
be
viable,
but
is
it
Universal?
A
Does
this
always
need
to
happen,
and
so
they
talk
about
this
to
different
biological,
relevant
situations.
Models
such
as
toy
life
and
virtual
cells
should
be
further
studied.
If
we
want
to
explore
issues
relevant
to
synthetic
biology
and
some
other
systems
where
we
can
really
manipulate
things
and
get
a
good
account
of
this.
A
So
this
is
something
that
we
see
a
lot
of
eukaryotic
reproduction,
where
we
get
Gene
duplications
and
deletions
any
variational
properties
of
a
gene
that
is
associated
with
the
gene
being
retained
in
the
genome
and
thus
becoming
enriched
in
the
GP
map.
So
this
is
where
you
know
there's
this:
if
genes
are
retained,
it
can
become
a
larger
component
of
the
GP
mapping.
It
becomes
more
important
to
the
phenotype,
and
so
this
is
something
that
you
know
the
an
interesting
area
of
research.
A
I
think
we
can
do
it
with
simulations,
but
also
with
very
simple
model
organisms
as
well,
and
so
Gene
do
differential,
Gene
duplicability,
which
is
where
some
genes
are
replicated.
Others
are
not
replicated.
This
is
something
that
people
have
investigated
and
what
its
consequences
are
for
organismal
evolvability,
and
it
can
build
quantitative
models
to
look
at
this
very
these
variational
properties
and
in
evolutionary
computation.
They
call
this
constructional
selection,
but
you
can
look
at
it
in
other
ways
as
well.
A
And
so
finally,
there
are
these
entropic
evolutionary
forces,
and
so
we
talked
a
little
bit
about
information,
content
and
entropy
in
the
other
paper,
but
you
know
they're
these
different
entropic
phenomena,
and
so
this
results
from
Evolution
along
neutral
networks.
So
a
couple
of
these
are
sub-functionalization
where
you
get
different
functional
modules,
constructive,
neutral
Evolution,
which
is
something
we
talked
about.
Erlin
stoltzfus
has
introduced
this
General
concept
of
entropic
processes
that
produce
greater
genetic
complexity,
to
treat
simply
because
they're
happening
to
be
more
complex
ways
to
generate
a
tree,
then
there
are
simple
ways.
A
A
This
is
an
exact
analogy
to
statistical
mechanics,
which
is
something
that
some
people
call
free
Fitness,
which
is
the
sum
of
the
fitness
phenotypes
and
the
sequence
entropy.
The
analog
temperature
is
the
inverse
of
population
size.
They
use
that
type
of
statistical
Mech
model
to
kind
of
understand.