What questions should I
ask before adding a new compartment or process to a model?
- If this compartment or process is eliminated, what
portion of the experimental data can I no longer explain
even if I am allowed to change the values of all the
remaining rate constants?
- If this compartment or process is eliminated, what else
could be done to the model so that it again accounts for
the experimental data? If I can find an alternative model
that accounts equally well for the data, what experiment
can I, or someone else, do to distinguish between them?
- If the rate constant for this process is made very fast,
do the model predictions become inconsistent with my data
in a way that cannot be reversed by changes in the values
of other rate constants?
- If the rate constant for this process is made very slow,
do the model predictions become inconsistent with my data
in a way that cannot be reversed by changes in other rate
constants? Questions 3 and 4 are the qualitative
equivalent of asking whether the rate constant is
identifiable (has a coefficient of variation less than
0.3).
- What is the physiological meaning of this compartment or
rate constant? If it is counter to current dogma, can I
explain in detail why current dogma fails to account for
my experimental data?
- If a compartment or process is included in the model
because it has been resolved in previous investigations,
but is not resolved by the current experiment, then say
so and fix the value(s) of the parameter(s). If a
parameter or compartment is new to the field of enquiry,
all of its quantitative features must be resolvable from
my data. There is no justification for NEW processes or
compartments whose values are fixed rather than
adjustable. If you introduce something novel, you must be
able to resolve it from your own data.
- If I were to impose on this model the experimental
protocols used by other investigators, would my model be
able to account for their results?
These questions raise a formidable barrier
to addition of unnecessary or unwarranted compartments and
processes. They also militate against removal of widely accepted
compartments and processes. This is how modeling supports
scientific method. By answering these questions systematically,
starting with the compartments and processes nearest to the site
of your experimental perturbation and working toward the most
remote parts of your model, you will accumulate exactly the
information that belongs in your published paper on this work.
Return to the
Computational Biology FAQ