THERE’S A COLLECTION of low-rise buildings in south-east England where
history is repeating itself. In 1939, a motley collection of mathematicians,
linguists and chess players responded to an urgent call to crack secret codes of
appalling complexity. Working long hours in a set of huts that had been rapidly
erected south-west of Cambridge, they gradually succeeded in uncovering the
secrets of their adversaries.
Now the call is going out once more—this time to researchers in fields
that were unimaginable sixty years ago: pattern recognition, terabyte data
handling, computer algorithmics. Hundreds are now working long hours in the
buildings of the European Bioinformatics Institute (EBI), a few miles from the
city. They describe their task rather prosaically as “genome annotation”. What
they are really doing is breaking the 3-billion-letter code of life.
Their efforts are being paralleled in the US, Europe and around the world.
Cracking the genome is suddenly very big business, and the skills these people
command are in high demand, everywhere from science-driven institutes like the
EBI to drugs companies with their sights on billion-dollar profits. When it
comes to biology, there’s never been a better time to be a physicist.
Advertisement
The code the researchers are poring over is written in just four
letters—the nucleotide bases A, C, T and G. These are grouped into “words”
of three bases—such as ACG—which form the long “sentences” called
genes. And it is these genes that form the instructions cells use to make
life-giving proteins. There are around 3 billion letters in the complete humane
genome, and the announcement in June of the code’s “rough draft” signalled that
researchers had obtained around 90 per cent of the coded message. Now they have
to crack that code.
The crucial next step is to find which are the parts of the code that
constitute genes, the sequences which tell cells how to combine the amino acids
we take in via our food to produce the proteins we need to live.
The international effort that was trumpeted in June handed genetic
code-breakers the equivalent of an intercepted message so long it would fill 350
encyclopedia volumes. Deciphering it is a formidable task, says David Kulp,
chief technology officer of Neomorphic, a Berkeley-based company with a
reputation for having some of the best genetic code-breaking techniques
around.
Though most code-breakers would be delighted to have so much cipher text to
work on, the trouble in this case is that perhaps as much of 95 per cent of it
is meaningless “static”: base sequences that don’t seem to do anything. “The
genes themselves aren’t continuous sequences of bases,” says Kulp. “They’re
broken up into exons, which are the instruction sequences we’re interested in,
and introns, which we’re not.” In humans the useless introns tend to be
thousands of bases long—ten times the extent of the exons. “Exons can be
spread across as many as 100,000 bases,” says Kulp, “so intelligently telling
where exons are in all this stuff is a difficult challenge.”
And even when that message has been extracted from the static, it’s still in
code. Exactly what the exons do has to be worked out somehow. But the genetic
code-breakers are nothing if not resourceful, and are calling in help from as
many different fields as they can.
Among their most valuable allies are human geneticists, who have spent the
past 20 years pioneering ways to locate the genetic defects responsible for
particular diseases. They look for differences in the genomes of closely related
people, some of whom suffer from a genetic disease. Differences that appear
consistently can help locate the gene responsible to a region of around a
million bases in the human genome.
That may not sound exactly like pinpoint precision, but it’s allowed genes to
be found for many diseases, including breast cancer, colorectal cancer and
Alzheimer’s. But the fanfare that has greeted these findings has tended to gloss
over the fact that the genes have typically been for relatively rare versions of
the diseases. Indeed, the genes were usually only spotted because their defects
have unusually strong effects. The discoveries are, in the language of the
genetic code-breakers, the “low-hanging fruit”, the easy victories. True, they
are great for morale, but they are also downright misleading for gauging the
effort needed to find more.
Since the 1980s, an impressive total of around 10,000 genes have been picked
off in this way. But there are many tens of thousands still waiting to be found.
No one knows the precise figure—guesses range from 30,000 to more than
100,000. What is certain is that tracking them down is getting tougher. Clues
from studies of people suffering from particular diseases are getting fainter,
and in many of them several closely related genes may each contribute to the
overall disorder.
That’s where the genetic code-breakers come in. In another example of history
repeating itself, the gene hunters are recruiting practitioners in a field
tailor-made to make the most of these faint clues, just as they were for wartime
code-breaking: statisticians. Their task is to decide when an apparent link
between some human disorder and quirks in the genome is just a fluke, and when
it’s genuine. “It’s an exercise in what statisticians call hypothesis
testing—comparing what you actually observe with what you’d get assuming
chance alone were responsible,” says Peter Donnelly of Oxford University.
Closing in
Donnelly heads the statistical advisory board of deCODE Genetics, a
Reykjavik-based company which is trawling the genomes and genealogies of
thousands of Icelanders, searching for significant correlations between genes
and common illnesses. The statistical techniques wielded by Donnelly and his
colleagues have already helped the company finger several promising regions of
the human genome containing genes linked to serious disorders, including
osteoarthritis, stroke and late-onset Alzheimer’s disease, which affects around
five per cent of people over 65. “Statistical tests allow us to narrow down the
location of the gene from anywhere in the 3 billion bases to regions of just a
few million—which is a thousand-fold improvement,” says Donnelly. “Some
methods can narrow down the region by a further factor of perhaps hundreds, to
just a few tens of thousands of bases”
But it’s not just promising regions people want: it’s the genes themselves.
Going that extra step is compelling the genetic code-breakers to bring in
specialists from yet other fields. And among those most in demand are experts in
pattern recognition. Their challenge is to find ways of telling when a sequence
of bases is part of a real, message-carrying gene, and when it’s just part of
the “static”.
It’s very difficult to do using statistics alone, says Kulp. “The patterns
that are typical of genes are difficult to spot because the statistical
differences between genes and non-genes is pretty low.” Even so, he has
developed one of the most powerful means of doing it. In 1996, while at the
University of California at Santa Cruz, Kulp and his doctoral supervisor David
Haussler pioneered the use of a technique mathematicians call hidden Markov
models (HMMs) for finding sequences of bases likely to constitute genes.
HMMs are based on the concept of the Markov chain. Roughly speaking, this is
a sequence in which what comes next depends only on whatever immediately
preceded it. It’s a property that has made HMMs very popular in signal and
speech recognition. Words “hidden” in noise can be detected by predicting what
sound or word is likely to come next, given what’s just been said. Now the same
ability is being exploited to spot the genetic “words” forming genes hidden in
the DNA static of the human genome. Kulp has put HMMs at the heart of
Neomorphic’s gene-spotting package, with impressive results. In an international
gene-spotting competition held last year, it spotted genes in the genome of the
fruit fly with around 95 per cent reliability.
But digging the genes out of the static is, while crucial, nothing compared
with the main challenge facing the code-breakers. “What we’re really looking to
do,” says Kulp, “is to computationally infer biochemical functions, interactions
and pathways from the gene sequences.” One of the most promising ways to do that
is to take another leaf out of the standard code-breaking book, and use clues
from genetic “messages” that have already been decoded. Researchers have already
figured out the functions of the genes in dozens of bacteria, as well as a
handful of higher animals and even some mammals. The trick is to borrow this
information to find out what’s happening in humans.
Terry Gaasterland and her colleagues at Rockefeller University, New York,
have developed a suite of computer programs called Magpie, which allows
researchers to compare human gene sequences with genes whose functions are
already known in other organisms. “It’s set up as a database, with every genome
used in relation to every other,” she explains. “All the evidence it presents
then comes with a confidence level, depending on the degree of similarity it
´Úľ±˛Ô»ĺ˛ő.”
Using techniques like Magpies, genetic code-breakers are working their way
through the human genome, identifying ever more genes and their functions. But,
warns Kulp, there are already signs that even the smartest, most powerful
computer methods aren’t going to be enough. It’s a view supported by Ewan
Birney, head of the genetic code-breaking effort at the EBI. “We need a
breakthrough in the biological model of how genes behave,” he says. “There’s a
need for people with really clever algorithms to speed things up, and for people
who can understand maths and physics. But there’s got to be input from
˛úľ±´Ç±ô´Ç˛µľ±˛őłŮ˛ő.”
Just getting those involved in the code-breaking effort to understand
everyone else is a problem right now, says Birney. He recalls a recent
conference where a molecular biologist stood up in front of an audience of
computer science people and asked whether she should explain what a vector was.
“No one said anything, so she went ahead,” recalls Birney. “It took the computer
scientists quite a while to work out that what she meant by a vector—a
means of gene transfer, like a virus—was totally different from what they
mean by it, namely an array of numbers.”
Sometimes the differences can have much more serious consequences. “In
computer databases it’s crucial that the objects in it have a single, unique
identifier— like a National Insurance number—so everyone can find
it,” says Birney. “But biologists like to give genes descriptive names like
`legless’, and when they’re told that they must use a number, they say they
don’t want to make it meaningless, they want to give it a name.”
What the genetic code-breaking effort now needs above all, says Birney, is a
new breed of scientist, as happy with the terse language of computer algorithms
as with the jargon of molecular biology and the intricacies of genetics. It is
already clear that only such strange, new, hybrid researchers have much hope of
understanding the language of the genes.
In their search for the skills to help them, the genetic code-breakers are
calling in help from some pretty eclectic quarters. Until recently, George Lake
was a professor of astrophysics at the University of Washington, Seattle, where
he acquired an international reputation for devising highly efficient computer
algorithms. One such, developed for simulating the formation of planetary
systems, was a million times faster than anything that had gone before. He’s
also the founder of the Whole N-Chilada Project, a multinational effort to
create a computer system that simulates, analyses and visualises systems made up
of billions of interacting components.
Now Lake is bringing his computer skills to the genetic code-breaking effort
at the Institute for Systems Biology in Seattle. “Increasingly, it has become
clear to me that biology has some of the hardest problems out there, with some
of the most compelling reasons for finding solutions,” he says.
Lake sees plenty of scope for bringing in the sort of computer-based help
that physicists take for granted. “Computation and information technology have
become very well integrated in to the physical sciences—and they certainly
aren’t in biology,” he says. “There’s a lot of very basic things that just
haven’t been done.” Lake also sees a need to make sure those trained in the life
sciences realise just how much help is already out there for them, in the form
of computing tools that are commonplace in other fields.
Lake thinks the can-do attitude of physicists faced with daunting problems
might also help when the going gets tough. “We work on the basis of controlled
ignorance,” he confesses. But he is also keenly aware of the reputation for
arrogance that physicists have—and the need to work alongside the life
scientists who have spent years wrestling with this ultimate genetic challenge.
“Physicists have arrogance reverse-transcripted into their genes. They’re smart
and they know a lot—but they know a lot less than they think they do.”