|
Ants might not be particularly well known for their expertise in analytical chemistry, but two US researchers have successfully employed them to analyse mass spectrometry (MS) data. Their aim was to try to find a way to overcome some of the problems that can beset such analyses, especially when trying to spot biomarkers in MS data. 'There are a number of things that can hinder a successful data analysis of mass spectrometry data,' explains lead researcher Cristina Davis at the University of California, Davis. 'First, mass spectra generally are composed of extremely high dimensional data. Second, it is almost unavoidable to get mass spectra contaminated with a variety of noises.' So together with her colleague Weixiang Zhao, Davis decided to try out the ant colony optimization (ACO) algorithm. First developed in the early 1990s, the ACO algorithm takes inspiration from the way that ants collectively work out the shortest path from their colony to a source of food. This process involves an ant finding a food source and then leaving a trail of an odorous chemical known as a pheromone as it returns to the colony. Other ants can then follow this smelly trail to find the food, depositing pheromones on the way back and thereby making the trail stronger. But the pheromones also naturally evaporate, causing the trail to dissipate. As long as ants deposit pheromones faster than they evaporate, then the trail remains in place, but if they don't then the trail gradually disappears. By chance some ants will find their own paths back to the colony, some of which will be shorter than the original path. The pheromone trail on these shorter paths will build up faster than on the longer path, as each ant travels more quickly along the shorter paths than the longer ones. This pheromone build up will attract more ants to follow the short trail, causing less ants to follow the long trail, which will eventually disappear. In this way, the ants find the shortest path from their colony to the food source. The ACO algorithm applies this approach to finding the optimum solution to a particular problem. It does this by getting a number of ant-like agents to traverse some mathematical landscape representing the problem of interest. Like real ants, these agents leave trails that other agents can follow, with these trails strengthening or weakening depending on how many agents use them. Eventually, the agents will converge on a path that represents the optimum solution. This mathematical technique has already been used to find optimum solutions in a wide rage of different fields, from protein folding to routing vehicles, but it had never before been used to analyse MS data. So Davis and Zhao decided to test its ability at identifying the features in MS data that best distinguish blood serum samples from ovarian cancer patients and healthy controls. Specifically, they used the ACO algorithm to find the optimum wavelet features for distinguishing between the two groups of sera samples. Wavelets are essentially filters that extract specific time and frequency information from MS data. After conducting the ACO analysis 100 times, they eventually came up with eight wavelet features that could distinguish between the two samples with an accuracy of 98.8%. Furthermore, by analysing the interaction between these eight wavelet features and the MS data in more detail, Davis and Zhao were able to identify the specific biomarkers primarily responsible for the differences between the two sample groups. 'The ant colony algorithm provides a very efficient data mining platform and strategy for mass spectrometry scientists to extract features and biomarkers,' concludes Zhao. Davis and Zhao are now looking to improve the ACO algorithm and apply it to other MS data. 'We believe this will lead to additional successful applications of this novel approach in a wide range of fields such as pollutant detection, disease diagnosis and industrial process optimization,' Davis told separationsNOW. Related links:
The views represented in this article are solely those of the author and do not necessarily represent those of John Wiley and Sons, Ltd. |
|
|
|
|
|
