A team of researchers is using a potent machine-learning system to study an infection that's highly resistant to antibiotic therapies. With the work already yielding positive results, it could lead to improved understanding of bacteria, and ultimately the discovery of new treatments.
A recently-developed algorithm known as a denoising autoencoder was originally designed to pick out prominent patterns or features in large sets of data, without first being told specifically what to look for. The technique has been used for various purposes in the past, including analyzing random collections of YouTube images to identify common trends or features (unsurprisingly, cat videos were found to be popular).
UPGRADE TO NEW ATLAS PLUS
More than 1,200 New Atlas Plus subscribers directly support our journalism, and get access to our premium ad-free site and email newsletter. Join them for just US$19 a year.UPGRADE
Now, a group of University of Pennsylvania researchers are looking to utilize the technique for biological science, using it to uncover new information about organisms and their genes. Specifically, they used a specially-designed system known as Analysis using Denoising Autoencoders of Gene Expression (ADAGE), to study a bacterium called Pseudomonas aeruginosa, which is associated with cystic fibrosis and other chronic lung conditions. It's a particularly interesting candidate for study, as it exhibits high levels of resistance to antibiotic therapies.
The analysis was conducted using information gathered from 109 separate datasets, showing the identities of some 5,000 genes, with their expression levels detailed for each different experiment. The goal of the work was to discover how effective the algorithm is at pinpointing patterns in gene expression, and how those patterns change under different circumstances, such as in the presence of an antibiotic.
The researchers describe the complexity of ADAGE to be roughly equivalent to a brain containing only a few dozen neurons. However, despite the system's relatively simple nature, it was able to identify sets of genes that work together, and was even able to spot the differences between strains of P. aeruginosa taken from patients and those grown in the lab.
With those early positive results in mind, the researchers are confident that the system will prove useful in finding effective new therapies for attacking cystic fibrosis lung infections. On a broader scale, the use of such systems could lead to significant breakthroughs in the field.
"We think that the proliferation of 'big data' provides an opportunity, through the use of unsupervised machine-learning, to find completely new things in biology that we didn't even know to look for," said team member Casey Greene.
The findings of the study are published in the journal mSystems.
Source: University of Pennsylvania