Understanding the causal origins and
mechanistic principles for the behavior

of an evolving system is one of the major
challenges of our time.

For example, how a protein may fold to
become a functional piece in your body

or how a drug may help against
some disease.

A causal system can be described by an
algorithmic model evolving over time.

The length of the shortest computer model
is called

the algorithmic information content of 
the system.

That is how much computer code is needed
to reproduce the object itself.

The shorter the description, the more
likely the system is causally generated.

The longer its description, the less
likely it is to be causally generated.

A dynamic system can usually be
represented

as a network of interacting elements,

such as interacting cells or interacting
genes turning on and off other genes.

What we want is to figure out the first
principles driving a system,

such as a genetic network representing a
cell

or the causes for the cell to behave in
one way or another.

Power approach consistis in finding a set
of algorithmic models that can explain

the structure of the system.

When the system is just a random process
and, therefore, not causally generated

it can only be represented by a long
descriptive model.

However, systems rich in causal content
can be represented by a short model,

because they have a generating mechanism
that evolves

into any observable state of the system.

To determine the causal content of 
an evolving system,

we peform perturbations.

In a network, for example, a perturbation
can be deleting a node or deleting a link.

We examine the effects of that
perturbation, and evaluate how much the

system became more or less random.

In a random system, for example,

a change will not have a major impact,
because no part of the system can explain

any other part of the system, and so
the perturbation goes unnoticed.

We can thus safely say that these 
systems cannot be reprogrammed.

But, in non-random causally generated
systems,

some changes will render them
unrecognizable,

causal interventions in these systems make
them reprogrammable.

Moreover, if we remove an element and the
system gets further away from randomness,

we can conclude that the element is not
part of the algorithmic content,

and is unlikely part of the causal
generating mechanism of the network.

It is thus likely to be noise or part of
another system.

Alternatively, if we remove an element and
the system moves towards randomness,

it means that the element is part of the
causal model that explains

other parts of the system's evolution.

But, if we remove an element and the
system does not approach either

randomness nor simplicity, the deleted
element is non-essential in the

explanation of the system and likely an
element produced by

the normal course of its evolution.

Returning to gene regulation, what we do
is to apply this concept

and evaluate the contribution of
every gene and gene interaction to the

original network, ranking the elements by
causal contribution.

When this ranking is biologically
interpreted,

these tools have demonstrated the ability
to pinpoint markers of

cell function, cell differentiation, and
cell fate.

This shows how these tools can properly
profile systems elements, and steer,

and reprogram systems, such as biological
cells.

If these systems have some elements that
move the network towards randomness,

and other elements that move the network
towards simplicity,

we say these systems are more
reprogrammable.

But, if a system can only move in one
direction, then it is less reprogrammable.

Stem cells will be able only to move
towards a single direction

towards a differentiated cell,

while completely differentiated cells will
be able only to move

to the opposite direction.

This causal calculus is better equipped to
tackle the general challenge of

causal discovery and science than more
traditional tools,

and is helping scientists better
understand, and even steer,

biological and synthetic systems.