Computer models that mimic the patterns of human social behaviour are helping chemists to determine the molecular structures of powders. 

Scientists routinely work out how atoms are arranged inside single crystals by looking at the diffraction patterns created when X-rays pass through the material. But powders made from microscopic crystals pose a much greater challenge - the crystals' small size effectively compresses a three-dimensional X-ray pattern into a one-dimensional series of peaks, losing a lot of structural information in the process. 

One tactic to get the most from these data involves using computer models to make an educated guess about what the structure of the crystal might be, and then predicting what the powder diffraction pattern should look like. If the prediction matches the experimental data, the computer's structural guess is probably right. 

These predictions can be improved by a process similar to evolution by natural selection. Two reasonable estimates of the crystal structure can be combined to produce an 'offspring'. If this child provides a better fit to the experimental data, it is allowed to live on and breed with other good matches. But if it fails to make the grade, the child gets the chop. 

Maryjane Tremayne and Samantha Chong of the University of Birmingham, UK, have now boosted this evolutionary approach with a method that reflects social or cultural selection. 'We were inspired by the idea that human social behaviour can also be modelled in an evolutionary algorithm that adapts to a changing environment much quicker than genetic principles,' Tremayne said.

The key difference is that structural models that are a long way from the average are excluded from the next generation: 'the freaks don't get to breed,' explained Tremayne.

 'This can be compared to a new fashion catching on in a human society,' she added. While social factors can guide the evolution of the models, the pressures of natural selection still drive that process. This means that fashion can't over-ride the genetic factors, helping to 'avoid any fashion disasters'. 

The scientists found that this strategy reduced the time it took to solve the structure by 40 to 50 per cent, and hope that it might prove to be much more efficient than commercially available programs.

'This is a really neat idea in a hot area,' said crystallographer Chick Wilson of the University of Glasgow, UK. 'Structure determination from powder data is high profile within the structural chemistry community, being relevant for high throughput compound discovery and polymorph screening.' 

Tremayne said that the technique has 'potential applications in many other optimisation problems in chemistry', but that her next step will be to use it to study protein folding problems, in collaboration with her Birmingham colleague Roy Johnston. By representing each amino acid group as a single hydrophilic or hydrophobic entity, they hope to predict the complex folded structures of proteins.

Neil Withers