The previous E. coli gene regulatory

network example was mostly a static

example, although by way of original

analysis, one can think of artificially

evolving the network one step at a time

to see how the network can evolve over

time or as a result of interacting or

being perturbed by some other system from

the outside. But we can move to a more

dynamic example. In this case the network

associated to a type of cell that is

called T naive cell, which is associated

to the immune system and this is a cell

that has not yet been fully differentiated
.

In this new regulatory network, we

started from a network representing this

undifferentiated state and followed it to

its differentiated state when it becomes a

differentiated TH17 cell, which is a
specific

cell of the adaptive immune

system that helps keeping the organism

from getting, for example, an infection.

And it is related to all sorts of

interesting immune processes. So this cell

was analyzed in detail in paper published

in Nature, in 2013 and the data was made

available from data source. What we did

was to do exactly what we did for E coli

but now on three networks representing

the cell as a dynamical system evolving

from a non-differentiated to a differenti-
ated

state. You can see on the figure A

how the algorithmic information of genes

distribute and how many of the genes are

negative and positive, meaning that they

can move the cell towards or away from

randomness. When we talk about the cell,

we are talking about its genetic network

representation, which is a good represent-
ation

of the cell, as in it reminds many

if not most of its most basic processes,

leading to protein protection at metabol-
ic

pathways. But of course it may not

represent every aspect of the cell. So on

figure A you can see how genes distribute

differently at every step. With the two

first cases closer to each other than its

final configuration. Now as you may

remember, we have been saying that systems
that

have negative and also positive

elements are more reprogrammable because

as we found out, they allow us to move

the dynamical system qualitatively,
namely

to the modified and attractor landscape by

increasing or decreasing the attractors

and therefore, the depth of such 
attractors. One can see how in the

differentiated state, colored in green, 
the cell distributes all its genes on the

positive side, meaning that it has 
reached some sort of, both dynamical but

also algorithmic stability, in line with
all we saw in the previous unit.

Moreover, if we were doing any of this by

chance, we would not find any over 
representations of the genes that we

marked as negative or positive when 
performing a gene ontology

enrichment analysis. However, we did find

that genes related to processes identify

in this kind of immune cells were
pinpointed by the algorithmic dynamics

analysis, as you can see on figure B.
So in figure B, we have three plots

corresponding to each differentiating time

indicating how many genes identify as the

most positive or most negative appear in

the literature associated to the process

of differentiation. If we had not captured
the right genes the distributions would

look flat as if we had chosen random genes
out of all possible genes in the network.

But genes that are more positive or more
negative appear with greater number in the

early stage and in the intermediate stage
genes in the positive patch were highly

informative of differentiating processes.

And finally in the last stage, again, the
distribution is far from uniform

indicating that those genes identified by 
algorithmic dynamics were involved in the

processes of differentiation of the cell.

In the next slide, we will explain how GO 
enrichment analyzing was done in some

detail. But in figure C, you can see 
that this distribution of genes that were

positive and negative and how they flipped
from stage 1 and 2 and then became almost

all positive in the last stage, colored in
blue.

Almost like locking the state of the cell
when it reached its differentiative state.

Finally, on figure D, we can check the 
overall complexity of the network,

in which we see also suggestive by the 
algorithmic dynamics that the first

network somehow, received signals to start
its differentiating process probably

captured in this increase of complexity in
the second stage where the estimation of

algorithmic complexity increases accordi-
ng to BDM and then decreases again once

differentiated.

Here is a Venn diagram in the form of a
network, showing the genes in the square

nodes that have been identified in the 
literature from nine major papers shown

in black elliptic notes covering research 
on TH17 cells.

We had to do this because traditional
gene ontology databases are not

specialized enough to look over
literature specific to immune cells like

this one. Even that this is at the 
forefront of current research.

For the creation of this network I am 
choosing the literature on which it is

based. We teamed up professional 
immunologist.

This paper covers the majority of genes
that have been associated to TH17 cells.

Linked genes in the figure are genes found in
common between two or more papers.

Black lines show the number of genes 
found in common between every two papers

with the thickness denoting the size of
the overlap that were used in the gene

enrichment analysis of the previous TH17
differentiation network.

But even more interesting to see we are 
doing what we are claiming to do, we

devise a third experiment after getting 
the expected results from highly curated

and validated gene regulatory network from
E coli and then the TH17 network.

This other experiment involved taking the 
gene regulatory network of twenty one cell

types, mostly based on experimental 
results and therefore, of high quality.

We then applied not only the algorithmic 
information dynamic but also the

reprogrammability indexes, hoping that we
would be able to tell in what stage of

differentiation every of these different 
cell types were.

The experiment makes a lot of sense if you
think about it because all cells

ultimately come from the same very first 
cell we call an Embryonic stem cell,

denoted by the letters ESC, from were all 
other cells come from.

What we found was not only that the 
Embryonic stem cell was on top of

programmability measure, as you can see
on both diagrams but also that another

type of stem cell called Hematopoietic
stem cell, here denoted by the letters

HSPC, they are stem cells that give rise 
to all other blood cells, had also high

reprogrammability as it would be expected
and for example, BMT cells followed

closely relatively close to each other
conforming to the biological expectation

and then all other differentiated cells
were on the other side of the

reprogrammability index with low
algorithmic complexity.

So by taking this information, one can
attempt to reconstruct an attractor

landscape, also commonly known as
Waddington's energy landscape.

Where, similar to what we saw in the unit
of dynamical systems, one can capture and

picture an attractor landscape of an 
Embryonic stem cell and place its

non-differentiated state as being on top 
of a hill in a stable position, where

moves can make it differentiate into
different cell types by falling into the

different attractors represented by
valleys that are also more stable than

the hills. So I hope you see how
interesting this application of

algorithmic information dynamics is to
real world, in particular to molecular

biology and genetics. In the next
lectures we will see other applications to

other areas such as evolution, cognition
and machine learning among other topics.