Researchers on the University of Cambridge have harnessed AI within the fight against antibiotic resistance.
The research team, led by Professor Stephen Baker, created a machine learning tool using only microscopy images to differentiate between bacteria proof against ciprofloxacin (a standard antibiotic) and people at risk of it.
This could dramatically reduce the time required for diagnosing antibiotic resistance, potentially transforming how we treat dangerous infections like typhoid fever.
The study, published in Nature Communications, focused on Salmonella Typhimurium, a bacterium that causes severe gastrointestinal illness and may result in life-threatening invasive disease.
Salmonella is a bacteria that commonly infects humans through contaminated food, and a few strains have gained antibiotic resistance. Source: University of Cambridge.
Dr. Tuan-Anh Tran, a key researcher on the project, explained the approach in a blog post: “The great thing about the machine learning model is that it could actually discover resistant bacteria based on a couple of subtle features on microscopy images that human eyes cannot detect.”
The research process involved several key steps:
- Bacterial sample preparation: The team grew S. Typhimurium samples in liquid nutrient media, some exposed to different concentrations of ciprofloxacin and others not.
- High-content imaging: Using a complicated microscope, the researchers took detailed pictures of the bacteria at multiple time points.
- Image evaluation: Specialized software extracted 65 different features from each bacterial cell, including shape, size, and interaction with fluorescent dyes.
- Machine learning model development: The researchers fed this data into various machine learning algorithms, training them to acknowledge patterns related to antibiotic resistance.
- Feature selection: The team identified probably the most crucial features for distinguishing between resistant and susceptible bacteria.
The results of this process were impressive. The AI system appropriately identified antibiotic-resistant bacteria about 87% of the time.
Perhaps most importantly, the researchers found that resistant and susceptible bacteria had distinct visual patterns that the AI could detect, even after they hadn’t been exposed to antibiotics.
This suggests that antibiotic resistance changes the looks of bacteria in ways which are too subtle for humans to see, but that AI can detect.
Current methods typically require several days of bacterial culture and testing against various antimicrobials. In contrast, the brand new AI-based method could potentially provide results inside hours.
Faster diagnosis allows doctors to prescribe probably the most effective antibiotics sooner, potentially improving patient outcomes and reducing the spread of resistant bacteria.
Looking ahead, the research team goals to expand their approach to more complex clinical samples like blood or urine and test them on other varieties of bacteria and antibiotics. They’re also working on making the technology more accessible to hospitals and clinics worldwide.
As Professor Baker explains: “What could be really necessary, particularly for a clinical context, could be to have the opportunity to take a fancy sample – for instance blood or urine or sputum – and discover susceptibility and resistance directly from that.”
“That’s a rather more complicated problem and one that basically hasn’t been solved in any respect, even in clinical diagnostics in a hospital. If we could discover a way of doing this, we could reduce the time taken to discover drug resistance and at a much lower cost. That may very well be truly transformative.”
Dr. Sushmita Sridhar summarized the impacts, stating, “Given that this approach uses single cell resolution imaging, it isn’t yet an answer that may very well be readily deployed all over the place. But it shows real promise that by capturing just a couple of parameters in regards to the shape and structure of the bacteria, it could actually give us enough information to predict drug resistance with relative ease.”
As antibiotic resistance continues to pose an escalating global health threat, revolutionary approaches like this AI-powered imaging technique offer latest hope.
This is an element of a broader trend of AI-driven innovations in antibiotic research. At MIT, researchers have used deep learning models to find a wholly latest class of antibiotics.
In the same vein, one other team of scientists announced in May last 12 months that that they had used AI to discover a brand new antibiotic effective against drug-resistant bacteria.
AI enables faster, more accurate identification of drug-resistant infections, paving the way in which for simpler treatments and higher patient outcomes.
The next few years will probably be crucial because the team works to translate their laboratory success into real-world clinical applications.
