Transcript CIN_W2_Presentation_Wednesday_Session_1

Causality in Complex Systems
The Probem of Modularity
Mt. Tamborine, July 8, 2009
Sandra D. Mitchell
Department of History and Philosophy of Science
University of Pittsburgh
1
Dynamic
Compositional
Human Brain
Dictyostelium – slime
mold life cycle
Evolved
Honey Bee Colony
2
Complex Biological Systems





Evolved contingency
Multi-level organization
Multi-component causal interactions
Modularity compositional structure
Robustness in relation to internal and
external changes
3
A manipulationist account of
causation


Mill’s “Method of Difference” “If an instance in which
the phenomenon under investigation occurs, and an
instance in which it does not occur, have every
circumstance in common save one, that one occurring
only in the former; the circumstance in which alone the
two instances differ, is the effect, or the cause, or an
indispensable part of the cause, of the phenomenon.”
(Mill, 1888, page 280).
The paradigmatic assertion in causal relationships is that
manipulation of a cause will result in the manipulation of
an effect. … Causation implies that by varying one factor
I can make another vary. (Cook & Campbell, 1979, p. 36,
emphasis in original.)
4
What would have happened if X
had been different?



Jim Woodward Making Things Happen
Explanation requires only invariance, not
universal truth. Invariance comes in
gradations or degrees.
Relation between variables F (X,Y) is not
universal. Under certain “ideal interventions”
where the value of X changes, the function
will describe the value of Y. Hence X
explains Y for those ranges where the
functional relationship is invariant.
5
Features of causality

Invariance



Modularity or Separability



Relation between dependent and independent variable
Relation between purported cause and effect
Is a relation among multiple functional equations
describing a single system
i.e. A relation of independent disruptability or
contribution to overall effect of casual components
Insensitivity

Stability of causal relation across variations in
background, context, or conditions external to system
6
Modularity essential to Causality
“..this is implicit in the way people think
about causation…this sort of independence
is essential to the notion of causation.
Causation is connected to manipulability
and that connection entails that separate
mechanisms are in principle
independently disruptable.” (Hausman
and Woodward 1999)
7
Modularity beyond equations


Modular equations represent distinct,
autonomous, context-insensitive causal
mechanisms subject to independent
disruptability.
Modularity in biology, especially evo-devo
is held out as potential theoretical
unification.
8
Modularity and Exportation


Modularity reflects one form of invariance
– distinct, autonomous, internally contextinsensitive causal mechanisms subject to
independent disruptability.
A module’s behavior in one system is
sufficiently stable and insensitive to be
exportable other systems.
9
Is Modularity essential to causation?

Some causal networks have elements that
do not behave in modular ways



Do we want to infer that the elements are
NOT causes?
Do we have to move to a finer or coarser
granularity to satisfy the
modularity/independence required?
Do we want to infer that modular causes do
not exhaust causality?
10
Genetic knock-out experiments
11
Genetic Knockouts


Principle assumption: normal function of a gene
can be inferred directly from its mutant
phenotype.
Results are hard to interpret



Sometimes “intervention” on one gene  lethal
Sometimes “intervention” on one gene  change in
phenotype
Sometimes “intervention” on one gene  virtually
no change in phenotype
12
Problem of Inference

“The big surprise to date is that so many
individual genes, each of which had been
thought important, have been found to be
nonessential for development”
Robert Weinberg

“I don’t believe in complete redundancy. If
we knock out a gene and don’t see
something, we’re not looking correctly”

Mario Capecchi
13
14
Redundancy and Degeneracy

Redundancy


When a gene is “knocked out” other elements
of the same structure are activated. Built in
“fail-safe”.
Degeneracy or Robustness

When a gene is “knocked out” elements of
other structures respond flexibly to issue in a
similar functional outcome.
15
What do we say of this system?

The causal structure reorganized?


With intervention – new causal roles for
elements which are non independent or
context dependent
The initial representation was not
complete.

Response to intervention indicated that not
ALL the causal relations were initially
represented. Existing ones became active.
16
Responses



1. “bite the bullet” strategy the genes in the
normal genetic pathways are not causes
2. “make the world fit your theory”
redescribe the network in finer or coarser
granularity
3. “pluralism of causes” modular causes do
not exhaust all the types of causality
17
Actual Causes versus Causal Laws



Domain of biology is historically and
currently existing organic life – not what is
possible, or potential given the constraints
of physics and chemistry.
Biology studies ACTUAL causes i.e. a
subset of what is biologically possible.
And…actual causes may not behavior
modularly
18
Biological Modularity

“Speaking loosely, biological modules are
consortia that act autonomously to produce
a single form or function and are
redeployed within and across species,
thereby creating novelty and fueling the
development and evolution of biological
complexity.”

Meyers 2004 Nature
19
MODULARITY in biology

Apparent paradox:



same gene or gene complex  different
structures in same organism
same gene or gene complex  different
structures in different taxa
Module is part of a system whose internal
constituents are strongly interactive while it
is as least partially independent of other
modules – weakly integrated.
20
Evolutionary and developmental
modularity

Evolutionary Modules


selectable units, i.e. change can occur in one
module without causing disruption in the
others, thus increasing viability while
permitting adaptive variation in complex
organisms.
Developmental Modules

Morphological units i.e. discrete interactive
systems of causes with stable context
insensitive effects in contributing to
development of an organism.
21
Problem: Modules are system
and grain dependent




“Although the Notch signaling pathway is contextindependent module, in the sense that the molecular
interactions between its members are conserved and
invariant, the outcome of Notch signaling is highly
context dependent” Celis, (produces neuron, hair, dermis)
“Shh (super sonic hedghog) signalling is a contextindependent module with conserved functon during
vertibrate muscle development” Borycki
Sometimes – invariance is structure,
sometimes function.
Evolutionary modules do not necessarily
map onto developmental modules
22
Methodological implications


Singe perturbation experiments even if they
indicate causal contribution of gene, assume a
fixed genetic background. If there is interaction
with the background genes, the results may not be
applicable to other contexts.
Multifactorial experiments – use existing
variation to generate multiple backgrounds and
compare effects of gene perturbation across all
realistic genetic variation
23
Conclusion and choices



Accept that what is “well-behaved” causally
varies with embedded context OR
Accept that there are non-modular, context
sensitive actual causes that explain the behavior
of biological systems.
The consequence is that exporting knowledge
from one system to another requires more
than generalization and instantiation – need to
use more local knowledge
24