The Guardian view on protein modelling: the answer to life, the universe and everything | Editorial

‘Designing medicines to target diseases requires knowing what proteins are involved and their form.’ Scientists have identified a protein which is a key driver for the growth and spread of breast cancer. Photograph: Rui Vieira/PA

When Eliezer Yudkowsky, one of the world’s top artificial intelligence theorists, mused about how superintelligent robots might wipe out humans he speculated that perhaps they would solve one of the science’s holy grails: predicting protein structure from DNA information. In Mr Yudkowsky’s words these robots would then “synthesise customised proteins ... building even more sophisticated molecular machines. Imagine tiny invisible synthetic bacteria, with tiny onboard computers, hiding inside your bloodstream and everyone else’s. And then, simultaneously, they release one microgram of botulinum toxin. Everyone just falls over dead.” Mr Yudkowsky’s apocalyptic scenario rests on something science has pondered with no answer for decades: why can’t we say what determines a protein’s shape?

This is not some idle speculation. Proteins are the bedrock of living systems, intimately involved in every physiological process from triggering an immune response to thinking. Good health requires a fine balance of proteins. An imbalance, and disease often strikes. Cancer is traced to an overproduction of proteins. Misfolding proteins have been linked to type 2 diabetes, while the strange bundling of them is thought to be behind the death of brain cells in Parkinson’s disease. Proteins’ function is dependent on their form, which is the result of a folding up of hundreds of amino acids – its constituent parts – into a specific and complex 3D structure. That configuration determines what the protein does: whether it becomes an enzyme to accelerate a chemical reaction; or a receptor passing signals to a cell’s molecular machinery. Crucially, a drug can alter a protein’s function by binding to it in a particular spot. Designing medicines to target diseases requires knowing what proteins are involved and their form. After a half century we can identify 100,000 protein shapes. But we have a database of 100m proteins. That is why we have few molecular keys capable of picking the lock to understanding disease-causing proteins.

Why has protein structure proved so hard to crack? They can be probed with x-rays, but that means first purifying proteins and then growing them as crystals in a laboratory. It’s a lengthy process. Some do not seem to crystallise at all. There are glimmers of hope. David Jones at Britain’s Francis Crick institute, which has just been awarded a £2m European grant, uses new computational techniques to predict novel protein structures. But the real prize is the one Mr Yudkowsky identified: by looking at DNA, could one predict the shape of the proteins it released? Since DNA encodes the amino-acid building blocks of an organism’s proteins, we know their composition. This is not much help with their structure. Human proteins can fold up in an astonishing number of ways: about a googol cubed or 10 to the power of 300. There’s not enough computing power to work out all these possibilities and thus find the optimum. Less than 10% of human DNA codes and regulates proteins. But we have no idea how altering the gene sequences changes proteins’ forms and functions. If we did understand it so that we could tamper with it to our advantage, then it would likely lead to all sorts of ethical dilemmas such as growing older without ageing. Maybe that is why Mr Yudkowsky considered it a task only solvable by a superintelligence so clever that its very existence might spell our end.

By using Yahoo you agree that Yahoo and partners may use Cookies for personalisation and other purposes