Biology Reference
In-Depth Information
of their implications are deeply rooted in evolutionary and developmental per-
spectives. I will explore the role they can have in ameliorating the problem of
incomplete and unreliable data in the next two sections.
5. ROBUSTNESS AND THE MANAGEMENT OF UNCERTAINTY
The following strategies and constraints seem reasonable ways of dealing with
uncertainty in data:
(1) Particularly if data is important, try to determine it in more than one way.
That is, incorporate robust designs to increase reliability into your experi-
mental methodology. This not only reduces errors through cross-checking,
but can also be used to detect systematic differences that may lead to tech-
nological improvements to reduce errors and to have better knowledge about
when data can and cannot be trusted (Wimsatt, 1981; Levins, 1968). The
late Sylvia Culp (1995) provided powerful and revealing examples in her
analyses of diverse methods in molecular genetics.
(2) Models of smaller circuits or systems require less data to keep the included
errors to reasonable levels. Learn how these circuits behave, and their sensi-
tivity and robustness to changed structure and parameter values. If they are
robust, use them as 'seeds', taking their outputs as given, and investigate
the circuits including and intersecting them.
(3) When modeling larger circuits, look particularly for their robust properties.
(4) Take the values and behaviors emerging from such simulations with a grain
of salt. Regard the simulations as exploratory rather than definitive.
(5) After finding a behavior that is somewhat robust, try specifically to 'break' it,
determining the conditions under which it fails. These might be informative:
it might be a tunable switch or threshold device, breakdown conditions may
indicate other variables that must be maintained or other dimensions in
which it is designed to be robust.
(6) If you find a property that appears to be biologically important, and it is not
robust, be suspicious of your model or the assumed parameter values. This
is the complement to the old maxim of adaptive design that 'Nature does
nothing in vain'.
11
The more important something is, the more important
it is to guarantee its presence. Nature does not guarantee anything, but it
is a good working hypothesis. So if the property is fragile in your model,
explore the possibility that it is not important, that the model is wrong, or
that you have misidentified its function and what it is doing.
11
The primary application of that principle in this context is the reverse engineering one: the more complex
is a mechanism, the more important is its function or functions. This may fail to be true if the mechanisms
and function have been recently co-opted from another functional system, or 'kluged'.
