Artificial intelligence used in drug detective work.

Artificial intelligence used in drug detective work.

Computers running artificial-intelligence algorithms might soon be helping forensic scientists to profile illegal drugs, according to scientists at the University of Strathclyde, UK.

Niamh NicDa?id and colleagues have found a way to help classify ecstasy pills and identify which clandestine laboratory a particular sample comes from. Using inductively coupled plasma mass spectrometry (ICP-MS), a technique more famous for its application to geology, NicDa?id measures the inorganic impurities in each tablet to create a ’chemical fingerprint’. By comparing the fingerprints with each other, similar pills can be grouped together.

Until now, the main obstacle to this kind of forensic analysis has not been measuring a pill’s fingerprint, but rather analysing the data properly. Traditional chemometric techniques fail because of the huge variation in chemical composition between tablets.

The Strathclyde researchers turned to artificial neural networks which work in much the same way as the human brain. One particular advantage of neural networks is that no prior assumptions are made. To sort samples into groups, traditional chemometric techniques often require extra information, such as the number of categories that are expected. Neural networks, however, need to be trained and must ’learn’ how to analyse data correctly.

Batches of 100 neural networks were repeatedly generated, but NicDa?id selected only the best 20 performing examples for use in the final analysis. This process was repeated until 100 networks were obtained. Although every neural network is individual, this Darwinian-like survival of the fittest ensures that each is up to the job. Of the 96 tablets analysed, between 96 and 99 per cent were assigned correctly to the group they came from.

NicDa?id explained that while the work is still in its conceptual stages, there are high hopes for further developments: ’at present most of our work has been looking at amphetamines, specifically ecstasy. But we have a lot of data for other narcotics, such as heroin, and there is no reason why neural networks couldn’t be applied to these’.

Ian Farrell