IN BIOLOGY, you can鈥檛 claim to know how things work until you鈥檝e seen them in
action. Other disciplines have the luxury of 鈥減redictive modelling鈥�. For
example, physicists postulate the existence of the Higgs boson鈥攚hich gives
other particles mass鈥攅ven though nobody鈥檚 seen it. But in biology, the
variables are too numerous for predictive modelling to be of much use.
That鈥檚 about to change, says Sydney Brenner, president of the Molecular
Sciences Institute in Berkeley, California. Biologists will soon be able to
build powerful computer simulations of organisms that may forecast, say, whether
a strain of corn can survive a drought, or predict the human body鈥檚 response to
a new drug.
Making all this possible are the heaps of data now being gathered鈥攏ot
just the 3 billion bases of the human genome, but also reams of information on
what genes do and how proteins interact. The challenge is to make sense of it.
鈥淎t the moment, everybody is just packing their computers full,鈥� says Brenner.
鈥淎s soon as the data binge is over, the rate-limiting step will be ideas.鈥�
Advertisement
Brenner suggests beginning with simulations of simple systems such as
bacteria. Given information about the way the proteins in a cell behave,
including enzyme rate constants and data about protein interactions, these
models may yield testable predictions. Researchers can then feed back
experimental results into the models to refine them.
Drew Endy of the Molecular Sciences Institute has already written a program
that simulates T7, a virus that attacks Escherichia coli bacteria. The
program simulates the location and activities of each of T7鈥檚 fifty or so genes,
and can reproduce its growth and infectious life cycle. It can predict how T7鈥檚
genes, RNA and proteins react when nutrients are scarce and model the growth of
genetic variants.
Ultimately, such models may become sophisticated enough to simulate complex
organisms, perhaps even humans. But could such models ever reveal what an
organism is like from its DNA sequence alone? Brenner thinks not, at least not
without extensive knowledge of protein behaviour.
Researchers still can鈥檛 predict how a protein will fold from its DNA or amino
acid sequence. Yet a protein鈥檚 shape dictates its function. If shapes cannot be
calculated, every protein鈥檚 shape will have to be worked out by, say, X-ray
crystallography. 鈥淲e鈥檒l have to accept there are certain things we can only get
by measurements,鈥� Brenner says.
Predictive modellers have many other obstacles to overcome. Knowing when and
where a gene is active does not prove what it does, he says. 鈥淓volution only
produces things that work, not necessarily things that are perfect,鈥� says
Brenner. So genes may be switched on even when they are not needed, just like
lights in an empty room.
Not all biologists will leave the lab bench behind. But some will use
computers at least as often as a pipette. And soon they may be telling the folks
back in the lab what鈥檚 going to happen even before they鈥檝e seen it.
