22
H. KAesER, and J. A. BURNS
Predictability and Understanding
We now have at our disposal values for the non-systemic, elementary
parameters (enzyme quantities, Michaelis constants etc.) and means of
altering these experimentally as well as values for the systemic variables
(pools, fluxes, coefficients, etc.). The well-known enzyme equations,
established by enzymology, together with the structural information of
our metabolic maps enables us to write down the set of equations in their
differential form which represent the whole system. These will solve for
us the questions for which we had no simple logical answer. The truth of
the matter is, however, that none but the simplest set of equations is soluble.
Our only way out of this dilemma is to obtain particular solutions. With
the advent of computers, provided one has sufficient funds to use them,
this has become a practical possibility. A system of any complexity can
now be handled and the computer can be made to run through as many
sets of data as you wish to supply it with. This is indeed the usual use of
computers in simulation (see e.g. GARFINKEL [4]). In this way we are able
to calculate the fluxes and sensitivity coefficients. We are able to show that
most coefficients are likely to be small, thus confirming the remarkable
buffering capacity in the arginine pathway already referred to (see [2]).
The computer can be used to confirm or deny suggested mechanisms or
quantitative assumptions; it can suggest experiments and predict their
outcome.
By such means we may regain predictability which we had lost because
of complexity. But we have not regained understanding. We have, so to
speak, delegated the observing of the processes to the computer and must
rest content with observing the outcome of its arithmetic. We are in fact
making empirical observations on a system-which is what we started off
by doing in our biological experiments. Of course, the computer system is
constructed in a known manner with specified elements and specified
relations. But then, provided we have made a correct and exhaustive
analysis of our 'real system', we had this knowledge before we programmed
the computer. The two systems are believed to be isomorphic and their
behaviour is what it is. We are left with what we might call the 'transparent
box' problem. Although we see everything in the box, how things are
connected and what each part is, the behaviour of the whole is largely
unknown,-except after the event. The status of the observations on the
computer is exactly the same as that of observations on the system which
it simulates. We have no insight or expectations beyond those obtained by
the empirical observations of either system. As with bird-watching, computer watching has its fascinations and advantages. In particular, it has
been fashionable for some time to claim that prediction is the only aim of
scientific endeavour. I think this view is mistaken. I think that prediction
H. KAesER, and J. A. BURNS
Predictability and Understanding
We now have at our disposal values for the non-systemic, elementary
parameters (enzyme quantities, Michaelis constants etc.) and means of
altering these experimentally as well as values for the systemic variables
(pools, fluxes, coefficients, etc.). The well-known enzyme equations,
established by enzymology, together with the structural information of
our metabolic maps enables us to write down the set of equations in their
differential form which represent the whole system. These will solve for
us the questions for which we had no simple logical answer. The truth of
the matter is, however, that none but the simplest set of equations is soluble.
Our only way out of this dilemma is to obtain particular solutions. With
the advent of computers, provided one has sufficient funds to use them,
this has become a practical possibility. A system of any complexity can
now be handled and the computer can be made to run through as many
sets of data as you wish to supply it with. This is indeed the usual use of
computers in simulation (see e.g. GARFINKEL [4]). In this way we are able
to calculate the fluxes and sensitivity coefficients. We are able to show that
most coefficients are likely to be small, thus confirming the remarkable
buffering capacity in the arginine pathway already referred to (see [2]).
The computer can be used to confirm or deny suggested mechanisms or
quantitative assumptions; it can suggest experiments and predict their
outcome.
By such means we may regain predictability which we had lost because
of complexity. But we have not regained understanding. We have, so to
speak, delegated the observing of the processes to the computer and must
rest content with observing the outcome of its arithmetic. We are in fact
making empirical observations on a system-which is what we started off
by doing in our biological experiments. Of course, the computer system is
constructed in a known manner with specified elements and specified
relations. But then, provided we have made a correct and exhaustive
analysis of our 'real system', we had this knowledge before we programmed
the computer. The two systems are believed to be isomorphic and their
behaviour is what it is. We are left with what we might call the 'transparent
box' problem. Although we see everything in the box, how things are
connected and what each part is, the behaviour of the whole is largely
unknown,-except after the event. The status of the observations on the
computer is exactly the same as that of observations on the system which
it simulates. We have no insight or expectations beyond those obtained by
the empirical observations of either system. As with bird-watching, computer watching has its fascinations and advantages. In particular, it has
been fashionable for some time to claim that prediction is the only aim of
scientific endeavour. I think this view is mistaken. I think that prediction