►
From YouTube: DevoWorm (2021, Meeting 43): Diatom Modeling III, Tissue Physics, Mutational Diversity
Description
Presentation on tissue timelines (multiple scales of temporal effects) and the role of physical stresses in measurement. Additional data analysis of Bacillaria colony velocity and acceleration. Presentation on genetic switches in synthetic biology. Papers on genetic mutations across cell lines and with respect to the germline and embryonic development. Attendees: Susan Crawford-Young, Richard Gordon, Karan Lohaan, and Bradly Alicea.
B
A
A
A
A
I
have
a
lovely
paper
on
that,
if
you're
interested
in
seeing
it-
and
then
I
found
this
graph
smooth
strong
graph
on
well
there's
just
one
paper
about
this-
I
can't
find
anything
else.
So
you
can
see
this
goes
from
milliseconds
to
seconds
and
you
can
get
hypoxia
due
to
stopping
blood
flow
because
you're,
compressing
it
and
and
before
that
you
can
get
what's
called
hyperelasticity.
A
A
A
A
A
So
if
you're
trying
to
measure
live
tissue,
you
have
to
be
very
aware
of
where
the
tissue
range
is
that
you're
dealing
with,
because
you,
like,
I
say,
you've
got
softer
tissue-
seems
to
be
more
sensitive
to
compression,
for
instance,
and
then
you've
got
other
tissue
like
muscle,
and
it's
used
to
being
ripped
apart
and
shunted
around
and
stretched
and
like
that's.
How
you
build
muscle
is
that
you
actually
cause
some
damage
to
that
muscle
and
then
it
rebuilds.
A
A
D
A
A
A
D
A
A
A
Yeah
anyways
did
I
put
in
an
another
slide
here:
okay,
because
then
I've
explained
myself,
so
you
have
elastic
properties,
so
there's
no
hysteresis.
It
goes
up
the
graph
linearly
and
back
down,
but
you
have
viscoelasticity
and
it'll
flow
like
a
fluid
under
the
right
conditions,
plastic
and
that
deforms
permanently.
A
A
Yeah
but
then
you
can
get
real
paddock,
and
that
means
it
becomes
stiffer
as
force
is
applied
and
their
synovial
fluids
in
our
joints
is
actually
somewhat
like
that
so
anyways.
This
is
I'm
having
to
think
about
all
of
these
things,
so
I
can
get
a
proper
experiment
to
measure
the
viscoelasticity
of
something.
A
So
it
you
almost
need
to
know
what
you're
measuring
before
you
can
decide
what
parameters
you're
going
to
use
yeah
most
likely.
That's
what
I
think.
So,
if
I'm
going
to
measure
an
insect
sphericals,
then
I
need
to
know.
I
need
to
know
that
and
then
figure
out
like
what
type
of
forces
I
should
use
to
measure
that
their
viscoelasticity.
A
D
A
A
A
E
D
D
A
D
A
D
A
A
C
A
A
D
D
A
B
I
had
a
question
about
the
you
mentioned.
Something
briefly,
and
I
know
a
little
bit
about
this
area-
is
that
you
have.
If
you
apply,
forces
to
cells
or
different
mechanical,
stresses
and
strains,
you
can
actually
trigger
cell
differentiation
or
you
can
trigger
you
know
some
cancer
pathway
or
something
like
that
as
well
as
apoptosis.
D
D
D
A
I'm
not
supposed
to
be
just
thinking
of
embryos,
I'm
supposed
to
be
thinking
of
all
tissue
and
I'm
going
well.
If
I'm
going
to
do
any
experiment
with
this,
I
have
to
know
the
tissue
ahead
of
time
in
order
to
set
up
the
the
experiment,
and
I
also
need
to
know
whether
you
want
hyperelastic
horses
or
or
not.
By
the
way,
have
you
looked
at
bone
yeah
people
like
to
study
bone.
I
yeah
in
my
biomechanics
class.
A
B
I
know
there
are
people
interested
in
looking
at
something
called
brain
multiphysics,
so
it's
like
where
they,
you
know.
Usually
people
study
the
brain.
They
study
neurons
and
some
of
the
supporting
tissues,
but
they're
people
who
model
the
brain
or
the
nervous
system
is
like
a
set
of
different
types
of
tissues
and
they
study,
like
the
forces.
C
B
A
The
researcher
did
a
multiple
experiments
on
brain
tissue
and
there's
a
definite
difference
between
white
and
gray
matter,
and
she
also
tried
to
apply
shear
stress.
Then
I
took
a
look
at
the
experiment
and
I
said
okay
complicated
stress
because
she
compressed
it
and
then
she
shared
it
and
I'm
going.
No,
that's
not
sheer
stress,
that's
complicated
stress.
D
A
D
A
D
D
D
A
B
B
Yeah,
so
let's
see,
let's
hear
this
acceleration
analysis,
so
this
is
dick
sent
me
this
acceleration
noise
biography.
This
is
on
acceleration
noise,
so
we're
measuring.
Last
week
I
was
showing
the
measurement
of
velocity
and
acceleration
in
a
different
in
a
single
basil
area
colony,
and
he
sent
me
all
these
references
on
acceleration,
noise
and
gear,
and
it
looks
like
they've
done
a
lot
of
work
on
this
in
traffic
engineering
or
in
vehicle
engineering
yeah.
So
this
is.
D
F
D
F
D
B
So
dick
gave
me
some
ideas
of
how
we
might
do
this
plot
acceleration
noise
in
your
spike
frequency
versus
position
or
velocity,
compare
reversal,
velocity
and
acceleration
to
same
during
vessel
area,
reversal
plot
velocity
pair
length
versus
overlap
of
the
two
cells,
so
I've
gotten
a
little
bit
on
made
a
little
bit
of
progress
on
one
or
something
similar.
Oh
okay,
good,
I'm
good,
and
then
two
I
you
know
I
looked
at
the
data
last
week.
I
was
looking
at
it
and
I
I
was
trying
to
figure
out.
B
It
looks
like
if
you
take
the
vector
of
velocities,
it
looks
like
the
velocities
like
they'll
go.
You
know
basically
they're
positive
going
from
one
end
to
the
other,
but
there
are
some
points
that
are
negative
and
I
don't
know
if
it's
measurement
error,
if
there's
some
variation
there,
but
I
I'm
trying
to
figure
out
how
to
characterize
it.
I
haven't
been
able
to
get
at
it
yet,
but
I
just
looking
at
the
data,
I
can
tell
that
there
are
some
points
where
there
are
some
reversals
in
the
velocity.
D
B
D
With
point
three
because
bacillaria
again,
I
don't
think
anyone's
measured
this
for
single
diatoms,
but
if
a
single
diatoms
moving
and
then
suddenly
reverses
how
sudden
is
that
reversal?
Or
is
it
similar
in
time
scale
to
the
time
scale
for
reversal
bacillary
right,
okay?
In
other
words,
there
might
be
a
distinction
between
the
colonial
behavior
and
the
single
cell,
but
we've
already
shown
that
it's
not
in
the
noise.
D
A
B
B
B
This
is
just
the
the
points,
so
you
can
see
their
outliers
out
here,
and
so
one
of
the
things
I
did
was
to
get
to
create
like
a
a
filter
where
I
took
like
a
threshold
of
about
maybe
0.25
and
everything
above
0.25
or
below
point
negative
0.25
were
considered
spikes,
so
this
is
sort
of
spike
classification
for
this
trace,
okay
and
then
looking
at
this,
then
you
can
take
the
acceleration
curve
and
then
this
is
the
acceleration
curve
as
continuous
data.
But
these
are
the
spikes
in
this
blues.
E
B
F
C
D
B
F
B
That
would
be
good
yeah,
so
those
are
just
some
examples.
Yeah
I
mean
if
we
had
more
data,
I
could
do
more
with
it,
but
yeah
I'm
trying
to
figure
out
some
of
these
dynamics
within
the
time
series.
So
we
have
a
couple
of
different
from
last
week.
We
have
a
couple
of
different
comparisons
between
different
cells
in
the
colony,
so
we
have
yeah.
So
we
have
single
cells
and
then
relative
movement
of
two
cells,
and
so
we
have
a
fair
amount
of
data
for,
like
one.
C
B
Good
so
anyways,
let's
see
I'm
going
to
move
on
to
this
thing,
I
wanted
to
show
on
gene
switches,
so
I
I've
been
talking
to
my
other
group
about
this
thing
called
switching
and
it's
kind
of
an
odd
thing.
To
kind
of
I
mean
it
doesn't,
maybe
it
doesn't
make
any
sense
to
people,
but
so
in
a
bunch
of
different
natural
systems,
you
see
this
effect
called
switching
and
what
that
means
is
that
you
have
this
sort
of
relationship
where
you
have
a
system
and
it
switches
between
two
states.
B
D
Okay,
I
ignored
it,
but
just
yesterday
I
ran
across
a
paper
suggesting
that
bacteria
can
have
what
they
call
a
chaotic
state.
Okay
and
the
chaotic
state
is
vulnerable
to
different
kinds
of
antibiotics
than
normal
states
yeah
and
since
I'm
not
interested
much
in
antibiotics,
I
can
get
the
paper.
But
if
I.
E
D
B
It
again
I'll
send
it
to
you.
Okay,
that
sounds
good.
Okay,
yeah,
so
yeah
you
can
have
like
cells
can
have
different
states.
Sometimes
they
can
be
chaotic.
Sometimes
they
can
be.
You
know
some
other
state
and
this
this
might
be.
You
know,
differentiation
is
essentially
a
switch
too
so,
but
at
the
genetic
level
you
have
these
genes
or
these
gene
circuits
that
you
can
build
using
synthetic
biology
techniques,
and
you
can
build
these
switches
that
allow
you
in
synthetic
biology.
B
People
are
mainly
interested
in
like
say,
creating
different
biochemical
or
compounds
or
biochemicals
using
something
like
bacteria.
So
you
have
a
bacterial
genome.
You
you,
you
knock
in
these
cells
that
they
design
and
the
cell
might
have
a
switch
with
at
the
from
at
its
promoter
region.
Where
then,
it
switches
between
two
different
states
or
more
and
it
it
can
generate
things
if
you
give
it
like
a
stimulus
like
some
sort
of
food
source,
or
you
know
light
or
something
like
that,
it
can
flip
that
switch.
B
You
can
flip
on
and
generate
what
you
want
so
there
you
know
this
is
in.
This
is
a
the
synthetic
biology
version
of
the
switch
and
so
how
it
works.
In
this
simulation,
I
think
I
talked
about
lac,
z
promoters
on
it
a
few
weeks
ago,
but
basically
they
used
this
lac,
lac,
z,
gene,
which
is
something
you
find
in
bacteria.
B
It's
a
black
case
production
gene
and
you
can
there's
a
system
of
like
operons
that
turn
these
genes
on
and
off.
So
this
gene
will
produce
this
enzyme,
but
you
have
to
have
the
right
promoter
in
place
to
do
this,
and
so
this
is
a
very
famous
system,
this
black
z
system
in
bacteria,
and
they
can
use
they
can
manipulate
this.
So
this
this
is
the
geometric
circuit
that
they
propose
it's
a
black
operator,
which
is
this
lactase
gene
with
a
promoter.
B
Then
there's
this
other
weiss
institute,
which
is
a
biological
institute.
I
think
they're
out
of
boston,
they've
done
a
lot
of
work
with
these
kind
of
switches
in
actual
bio
biological
organisms,
so
the
gene
offs,
which
is
tool
up
synthetic
biology.
It's
the
title
of
this
article.
This
is
some
of
the
institute's
propaganda
as
it
were,
because
you
know
their
press
release
so
they're.
B
I
wanted
to
point
out
that
you
have
these
different
types
of
switches
that
people
are
working
on,
not
just
with
black
z,
but
all
sorts
of
different,
interesting
types
of
elements
that
they're
working
on.
So
you
can
program
a
repressor
element,
for
example
that
allows
you
to
repress
certain
genes
at
certain
times.
C
B
B
They
talk
about
ribo
computing
devices,
which
are
kind
of
a
larger
class
of
technology.
Here,
ribo
computing
devices
can
sense.
Multiple
biological
rna
signals
simultaneously
and
act
as
molecular
logic
boards.
So
these
are
you
know
things
that
you
can
use.
If
you
can
string
these
switches
together,
you
can
build
logic,
a
series
of
logic
gates
that
you
can
do
computations
with.
B
C
B
So
any
desirable
regulatory
element
for
synthetic
biology
would
be
a
device
that
can
do
exactly
the
opposite
of
this
thing
here,
which
is
only
if
a
certain
combination
of
input
signals
is
present.
Does
the
device
produce
a
desired
output
protein
so
in
synthetic
biology?
But
we
also
want
to
do
that,
but
we
also
want
to
shut
down
the
expression
of
protein
in
response
to
a
stimulus
when
it
is
no
longer
wanted.
B
B
Gates,
they've
been
able
to
implement
15
of
these
in
parallel,
and
so
you
can
have
them
all
working
in
parallel
to
do
that,
produce
different
things
to
shut
different
things
off,
and
so
this
is
a
paper
in
nature,
chemical
biology,
where
they've
published
this
and
then
they've
also
designed
something
called
a
toe
hold
switch,
which
is
a
different
type
of
switch.
This
is
where
you
have
rna
strands
that
detect
trigger
rnas,
so
these
are
rnas
that
trigger
these.
These
promoter
regions,
with
virtually
arbitrary
sequence,
is
to
activate
translation
of
a
linked
protein-coding
rna
sequence.
B
So
this
is
like
where
you
can
control
the
transcription
of
a
gene,
and
this
is,
I
think,
allows
for
greater
control
than
the
the
transcript
the
purely
transcriptional
switch,
which
is
where
the
promoter
merely
gets
turned
on
and
off.
This
is
triggering
the
things
that
sort
of
trigger
the
promoter,
so
this
is
a
in
in
a
cell.
This
is
important
because
this
machinery
is
sensitive
to
the
sequence
of
the
rna,
and
so
if
the
rna
is
the
right
sequence,
it
will
trigger
the
promoter
region.
But
if
you
know
you.
C
B
B
Methods,
oh
yeah
yeah,
so
I
mean
gene
deletion
is
yeah
and
they've
been,
and
people
worked
out
ways
to
sort
of
make
that
more
and
more
precise.
So
you
know
with
a
lot
of
the
stuff
with
crispr
and
where
they
can
just
cut
out
a
piece
of
dna.
But
this
is
where
you're
controlling
sort
of
the
expression
of
genes
you
can
put
genes
in
and
you
know,
build
actual
logic
gates.
So
this
means
that
you
can,
under
certain
conditions,
you
can
express
a
gene.
The
way
you
want
it,
yeah
yeah.
G
D
B
B
So
this
I
had
a
paper
that
I
let's
see,
this
is
the
one
that
I
didn't
get
last
week,
because
I
misplaced
it,
but
there's
a
new
type
of
method.
Here,
we've
talked
about
cellular
automata
in
the
group
here
and
there's
someone
from
university
of
manitoba.
So
so
this
is
called
learning's
graph.
Cellular
automata-
and
this
is
we've
talked
about
cellular
automata
in
the
group
before
as
a
way
to
model.
B
So
the
abstract
reads:
cellular
automata
are
a
classic
computational
model
that
exhibit
rich
dynamics
emerging
from
local
interactions
of
cells,
arranging
a
regular
lattice.
So
in
this
work
we
focus
on
a
generalized
version
of
a
typical
ca
called
graph
cellular
automata.
So
this
is
a
where
we
have
the
graph
part,
the
soil
or
automata
part
in
which
the
lattice
structure
is
replaced
by
an
arbitrary
graph.
C
B
In
this
case,
they're
taking
the
struct
lattice
and
they're,
replacing
it
with
a
graph
in
particular,
we
extend
previous
work
that
used
convolutional
neural
networks
to
learn
the
transition
rule
of
conventional
ca,
and
we
use
graph
neural
networks
to
learn
a
variety
of
transition
rules
for
gca,
so
they
are
using
graph
neural
networks
here.
So
in
this
previous
work,
they
used
a
different
type
of
neural
network
to
learn
transition
rules.
So
basically
in
cellular
automata,
you
have
these
rules
that
they
follow.
B
I
think
we've
talked
about
the
like
rule
120,
which
will
give
you
something
that
looks
like
a
seashell
pattern,
or
you
know
different
rules
that
will
give
you
different
patterns,
and
so
they
follow
these
rules.
The
you
know
the
rules
based
on
their
interactions
will
produce
this
macro
pattern.
In
this
case,
what
they're
doing
is
they're
taking
a
neural
network
to
learn
these
rules
but
they're
replacing
the
conventional
neural
network
with
a
graph.
C
B
Network
so
now
they're
learning
a
variety
of
transition
rules
for
this
type
of
cellular
automata.
So
this
paper,
I
think,
just
goes
through
how
this
is
done.
They
present
a
general
purpose
architecture
for
learning,
a
graph
cellular
automata.
Then
they
show
it
can
represent
any
arbitrary
graph,
cellular
automata
with
finite
and
discrete
state
space.
So
this
graph,
this
graph
neural
network,
will
learn
enough
rules
and
enough
about
the
rule
structure
to
reproduce
anything
that
that
cellular
automata
might
produce.
B
B
You
know
so
there's
that
kind
of
tessellation:
it's
imitating
the
behavior
of
a
group
of
flocking
agents.
So
this
is
where
you're
learning
this
transition
rule
on
what
is
essentially
a
grid
and
you're
tessellating
a
space
and
you're.
You
know
allowing
the
cells
to
you
know
evolve
these
matter.
These
pattern,
these
global
patterns,
two,
is
where
you're
imitating
the
behavior
of
a
group
of
flocking
agents.
B
C
B
F
D
B
Portuguese,
I
don't
think
so.
Well,
I
don't
think
so.
I
think
tom
actually
does
where
they
d,
he
does
the
cellular
automata
with
different
neighborhood
sizes,
and
then
he
uses
a
neural
network
at
the
end
to
classify
them,
but
this,
I
think,
is
actually
where
they're
they're
actually
trying
to
learn
the
rules
directly.
I
don't
know
if
I
don't
know,
if
he's
doing
that
step,
I
I
think
he's
just
kind
of
classifying
the
outputs
yeah.
B
I
remember
discussing
that,
but
yeah
don't
think
it
came
to
a
resolution
yeah,
but
I
think
I
think
the
point
here
is
that
the
like
they're
trying
to
improve
upon,
like
you
know,
people
have
been
using
neural
networks
for
that
purpose,
to
learn
the
rules
themselves,
but
you
know
it's.
It's
not
easy
to
do,
of
course,
because
it's
a
huge
state
space
and
you
need
to
you,
know
yeah,
but
so
these
graph
neural
networks
apparently
can
do
this.
B
B
Because
it
gives
you
some,
you
know,
past
work
on
this,
where
people
are
kind
of
building
up
to
the
this.
This
work,
so
so
a
seminal
work
on
the
subject
is
wolf
and
hertz
who's
successfully.
Treating
a
small
neural
network
imitate
one
and
two
dimensional
binary
cellular
automata
with
chaotic
or
complex
dynamics.
B
Then
people,
if
you
also
use
the
evolutionary
algorithms
to
identify
neural
network
transition
rules
that
would
generate
these
patterns.
So
this
is
what
they
call
morphogenesis,
but
this
is
you
know,
I
don't
know
if
that's
yeah,
so
they're
just
identifying
here,
they're
just
identifying
these
transition
rules.
So
it's
like
how
do
you
get
a
rule
that
produces
a
meat,
some
sort
of
pattern
that
looks
like
it's,
you
know
coherent.
B
They
also
use
the
neat
genetic
algorithm,
which
is
this
neural
evolution
topology
it's
a
little
bit
different
type
of
genetic
algorithm.
It
sort
of
moves
in
the
direction
of
neural
networks
and
then
compositional
pattern
producing
nns,
which
are
the
cpnns
as
well.
So
there's
been
past
work
on
this
and
then
you
know,
people
have
used
encoders
things
like
that
and
then
more
recently
people
have
looked
at
neural.
B
This
is
interpreted
as
an
image
and
then
this
they
train
the
neural
cellular
automata
to
converge
to
a
specific
image.
So
you
have
this
thing
that
you're
defining,
which
is
different
than,
of
course,
maybe
morphogenesis,
where
maybe
you
don't
have
a
target?
You
just
have
rules
that
give
you
the
target
consistently,
but
in
this
case
they're
actually
saying
there's
a
target
that
you
have
to
reach
and
I've
seen
people
do
this
before
with
in
different
ways
like
you
know,
they'll
use
a
target
as
a
training.
B
You
know
as
a
training
set
where
they'll
say
we
want
this,
this
pattern
of
a
barbell,
and
so
then
the
the
neural
network
will
produce
this
pattern
of
barbell,
but
that
isn't
necessarily
morphogenesis.
It's
just
kind
of
evolving.
This
cellular
automata
to
a
target.
So
that's
a
very
subtle
difference,
but
it's
important.
B
B
So
this
is
actually
here,
where
they're
able
to
train
this
cellular
automata
to
form
a
rabbit,
so
they
have
those
there
are
like
20,
I
guess
at
t
equals
20,
which
is
their
time.
They
get
a
lot
of
error
here
over
different
steps,
but
they're
able
to
get
something
that
looks
similar
to
a
rabbit,
they're
able
to
form
letters
gsp
in
the
same
way.
B
B
B
B
This
article
is
so
the
abstract
reads
over
the
course
of
an
individual's
lifetime
norm
in
human
cells
accumulate
mutations,
and
so
we
know
this
from
like
just
studies
that
in
your
lifetime
you
accumulate
mutations
by
interacting
with
your
environment,
but
your
germline,
which
is
the
thing
that
you
that
your
offspring
will
inherit
as
protected
from
those
sort
of
somatic
mutations
there.
Those
cells
are
sequestered
in
a
general
line,
pretty
early
in
development,
and
then
they
get,
they
don't
suffer
those
type
of
mutations.
C
B
Two
ubiquitous
mutational
signatures,
sps-1
and
sps-540-
accounted
for
the
majority
of
acquired
mutations
in
most
cell
types,
but
their
absolute
relative
contributions
very
substantially
sps-18,
which
potentially
reflects
oxidative
damage,
and
several
additional
signatures
attributed
to
exogenous
and
endogenous
exposures
contributed
mutations
to
subsets
of
cell
types.
The
rate
of
mutation
was
lowest
in
sperm
spermatogonia,
which
is
in
the
germ
line.
B
The
stem
cells
from
which
sperm
are
generated,
okay
and
from
which
most
genetic
variation
in
the
human
population
is
thought
to
originate.
So
this
is
a
germline
cell.
It's
experienced
a
very
low
rate
of
mutation.
Now,
germline
cells
will
experience
mutation,
but
they
won't
be
environmental
they'll
just
be
like
random
mutations,
so
you'll
see
very
low
rates
of
mutation
in
the
germline,
and
so
these
stem
cells,
from
which
sperm
originate,
are
protected
by
and
large
for
mutation,
and
so
this
is
where
a
lot
of
our
genetic
variation
originates.
But
it's
not
deleterious.
B
So
this
was
due
to
low
rates
of
ubiquitous
mutational
processes
and
may
be
partially
attributable
to
a
low
rate
of
cell
division
in
basal
spermatogonia.
So
every
time
the
cell
divides
you
get
an
opportunity
for
mutation,
you
get
an
opportunity
for
recombination
and
so
forth,
and
so
low
rates
of
cell
division
can
also
contribute
to
that.
B
These
results
highlight
similarities
and
differences
in
the
maintenance
of
the
germline
and
soma.
So,
as
I
said,
the
the
germline
is
something
that
you
know
gets
protected
from
a
lot
of
environmental
mutation.
It's
and
this
studies
actually
does
a
lot
of
this
sort
of
they.
They
kind
of
go
over
a
lot
of
what's
going
on
in
these
different
cell
types
in
terms
of
its
mutational
landscape,
so
they
use
next
generation
sequencing
for
this
and
they
they
align
these
next
gen
sequencing.
B
So
the
next
gen
sequencing
here
is
a
large
scale,
transcriptional
analysis
and
sequencing.
So
you
can
actually
look
at
like
the
genes
that
are
expressed
in
their
frequencies
of
expression,
but
also
the
sequences
that
they
come
from,
and
so
you
can
get
all
these
data
and
it'll
show
mutations
between
cell
lines.
You
can
identify
them,
and
so
this
just
kind
of
shows
the
donor
here
where
they
take.
You
know
I
guess
they're
harvesting
from
a
dead
body
or
they're
doing
biopsies
and
they
basically
yeah
the
warm
autopsy
and
additional
donors.
B
B
Then
they
can
do
these
samples-
and
this
is
these-
are
some
of
the
sort
of
the
histology
of
these
different
cells.
They
can
stand
with
antibodies,
so
they
show
some
of
the
expression
of
some
of
these
monoclonal
and
clonal
and
polyclonal
antibodies
in
this
in
the
histology
here,
so
they're
able
to
identify
different
mutations
in
the
in
the
tissues.
B
So,
as
we
were
talking
about
before,
you
know
we're
talking
about
forces
on
tissues
and
their
effects,
these
are
mutational
effects
on
tissues.
So
it's
quite
a
bit
different,
but
it's
still,
you
can
see
a
lot
of
variation
across
cell
types.
You
can
see
that
you
have
pancreas
and
kidney
and
prostate
gland
cells,
stomach
gland
cells,
all
these
different
areas
of
the
body-
and
you
can
see
the
sort
of
mutational
signatures
in
these,
and
so
so
yeah
mutation
rates
are
interesting
because
you
can
calculate
them.
We
don't.
B
B
B
B
So
yeah
this
and
then
there
are
different
mechanisms
that
underlie
this
mutation
rate.
Differences
in
the
mutation
rate
there's
also
dna
repair.
So
when
you
get
in
when
you
have
say,
like
a
skin
cell,
that's
exposed
to
the
sun
too
long,
you
can
have
mutations
in
that
cell,
but
then
the
dna
can
also
be
repaired.
So
it's
not
something.
That
is
a
long-term
thing.
So
you
get
these.
You
know
they
get
these
dynamics
of
mutations.
Sometimes
they
drive
a
cell
to
a
cancer
pathway.
Sometimes
they
get
repaired
and
it
depends
on
the
cell
type.
B
So
I'm
going
to
go
to
this
paper,
which
is
this
clonal
dynamics
in
early
human
embryogenesis
inferred
from
somatic
mutation.
So
now
we're
talking
just
about
somatic
mutations
and
the
abstract
carried
cellular
dynamics
and
fate
decision
in
early
human
embryogenesis
remain
largely
unknown,
owning
owing
to
the
challenges
of
performing
studies
in
human
embryos.
B
B
The
cell
bio
tissue,
biopsies
and
they're
getting
cellular
material
for
this
study
using
somatic
mutations
is
an
intrinsic
barcode,
so
they
have
this
technique
called
bar
coding,
where
they
can
identify
some
of
these
mutations,
and
you
know
they
can
use
it
as
a
way
to
identify
different
cell
types,
and
things
like
that.
We
can
reconstructed
early
cellular
phylogenys
that
demonstrate
so
a
cellular
phylogeny
is,
I
think,
we've
talked
about
phylogenys
before
there
are
these
trees
that
show
how
different
tax
that
diverge
from
one
another
historically.
B
B
These
cellular
phylogenys
demonstrated
five
things
actually,
so
the
first
is
an
endogenous
mutational
rate
that
is
higher
in
the
first
cell
division,
which
is
the
first
cell
division
after
its
differentiates,
but
decreases
to
approximately
one
cell
per
cell
division
later
in
life.
So
this
is
where
one
per
cell
per
cell
division.
So
this
is
where
the
mutation
rate
is
high,
as
it
first
is,
that
cell
type
is
born
or
that
cell
is
born
and
then,
as
the
cell
divides,
it
gets
fewer
and
fewer
mutations.
B
Interestingly
cells,
of
course,
unless
they're
stem
cells
they
age
as
they
divide,
so
they
have
this
limit
where,
if
they
make
enough
divisions,
they
undergo
this
apoptosis
process
and
die,
and
so
as
as
they're
getting
closer
to
that
apoptotic
point
end
point
they're,
actually
getting
fewer
mutations
in
there
in
their
genome.
So
this
is
an
interesting
sort
of
paradox.
I
guess
two
universal
unequal
contribution
of
early
cells
to
the
embryo.
B
There
is
a
universal
unequal
contribution
of
early
cells
to
the
embryo,
proper,
resulting
from
early
cellular
bottlenecks.
That's
stochastically
set
outside
epiblast
cells
within
the
embryo,
so
this
means
that
they're,
the
not
all
cell
types
are
equally
represented
in
the
embryo.
So
this
is
just
something
that
they're
identifying
from
the
mutational
signature
number
three
examples
of
varying
degrees
of
early
clonal
imbalances
between
tissues
on
the
left
and
right
sides
of
the
body.
B
So
there's
this
difference
in
terms
of
laterality,
your
left
side
of
your
body
and
your
right
side
of
your
body
are
supposed
to
be
symmetrical,
but
sometimes
they're
asymmetrical,
and
this
is
one
example
of
that.
Also
different.
Germ
layers
and
specific
anatomical
parts
and
organs
also
exhibit
this
clonal
imbalance.
B
So
this
means
that
they're
just
asymmetries
between
the
tissue
types
and
or
mutational
signatures
and
different
tissues
in
different
parts
of
the
body,
depending
on
where
we're
looking
four
emergence
of
a
few
ancestral
cells
that
will
substantially
contribute
to
adult
cell
pools
in
the
blood
and
liver.
So
you
have
these
cells
that
are
just
a
few
ancestral
cells
that
go
on
to
form
the
blood
and
liver
proliferate
and
form
these
these
different
things
in
the
body
and
then
five
presence
of
mitochondrial
dna
heteroplasmy
and
a
fertilized
egg.
B
B
B
So
this
is
something
that
they
go
back.
You
know
they
look
at
like
embryogenesis
from
the
adult
cells
in
the
body.
They
trace
these
cells
back
to
their
embryonic
origins,
and
they
can
make
these
inferences
through
the
mutational
signatures
and
they
can
find
out.
All
these
things
is
kind
of
a
really
amazing
paper.
I
I
don't
know
how
I
I'm
not
going
to
go
through
the
entirety
of
it,
but
I
mean
you
could
read
it
yourself,
but.
B
J
J
Some
other
ideas,
you
know
that
I
wanted
to
bounce
off
because
I'm
still,
you
know
kind
of
new
to
all
this,
but
still
like.
I
can
grasp
some
things
and
you
know
something's
kind
of
like
still
in
the
dark,
so
yeah
they'll
have
to
you
know,
go
through
a
bit
more
terms
but
like
like
susan's
ppt,
you
know
which
was
discussing
viscous
elastic
properties
of
tissues,
and
you
know
the
kind
of
applications
that
we
can
use
them
for,
like
building
tensor
models.
For,
let's
say
a
physical
medium.
J
You
know
where
we
can.
You
know
calculate
heat,
stresses
or
load
stresses
and
all
that,
so
that
is
kind
of
easy,
because
you
did
a
lot
of
libraries
for
that
by
the
library
so
matlab
we
can
do
it
on
matlab
as
well.
So
when
it
comes
to
potential
applications
like
like
there's,
this
idea
like
where
we
have
you
know
piezoelectric
material,
because
essentially
what
we'll
be
doing
is
we'll
be
mapping
out
the
stress
and
strain
relationship
of
this
tissue,
which
is
in
the
form
of
a
tensor
model.
J
So
maybe
you
know
data
and
in
terms
of
like
the
amount
of
energy
that
can
be,
you
know,
transferred
to
the
work
that
can
be
done
to
maybe
power
and
or
wireless
implantable
medical
devices,
because
this
is
again
a
field
of
you
know,
papers
that
I
was
going
through.
So
the
problem
in
this
field
is
a
lot
of
devices
that
are
used
here.
They
have
to
be
powered,
you
know
wirelessly
from
magnetic
fields
or
electromagnetic
induction
and
all
that.
J
So
the
challenge
right
now
here
is
you
know
finding
viable
sites
inside
the
human
body
like
some
people
was,
you
know,
suggesting
the
cardiovascular
system
for
propelling.
Maybe
you
know
devices
that
flow
through
your
bloodstream,
so
maybe
something
along
these
lines,
but
a
bit
more
stationary
like
maybe
it
has
a
target
site
where
it
is
close
to
your
heart
like
or,
let's
say,
pacemakers
or
your
muscles,
so
it'll
kind
of
like
giving
given
this
kind
of
data.
J
B
B
J
At
least
coming
up
with
the
model
for
tensorflow
like
state-of-the-art
things,
are
there.
I
think
it
won't
take
a
lot
of
time,
and
there
was
another
thing
I
had
like.
I
was
not
able
to
find.
You
know
accurate
data
when
it
came
to
axolotl
and
zebrafish
model
like
for
doing
the
carry
on
before
pony
model
thing
right.
J
B
I'll,
take
a
look
I
I
I
know
we
have
some
data
on
it's
it's
kind
of
hard
to
find
data.
We
have
some
zebrafish
data
that
are
like
cell
tracking
data
and
then
we
have
some
raw
axolotl
data,
which
is
just
it's
actually
a
three-dimensional.
Well,
it's
two-dimensional,
but
it's
it's
three-dimensional.
It
could
be
four-dimensional
if
you
can
extract
the
data
points
from
the
images.
So
that's
a
that's
a
stumbling
block,
but.
B
J
J
B
J
Yeah,
it
will
always
be,
you
know
helpful,
because
to
meddle
around
with
those
things
you
know,
you'll
have
to
see
what
properties
I'll
have
to
consider.
You
know
with
that
thing,
and
especially
when
generating
that
4d
model
thing,
you
know
it
becomes
kind
of
heavy
on
the
computer
as
well,
because
you
know
processing
the
4d
image
across
time,
depending
on
the
again
complexity
of
the
model.
You
know
those
factors
will
also
have
to
be
considered,
but
but
yeah
definitely
like
all
hell
will
be.
J
You
know
appreciated,
because
finding
data
sets
right
now,
especially
with
regards
to
a
lot
of
data
or
zebrafish
in
you
know,
going
through
the
developmental
stages
from
the
embryo
to
the
adult
state
and
also
with
regards
to
complexity.
I
think
zebrafish
has
lesser
number
of
stages
right
compared
to
an
axolotl
like
it's
much
more.