Antibiotic susceptibility testing has long been built around planktonic bacteria, but in the clinic, many chronic infections are driven by biofilms. These structured microbial communities behave very differently, often displaying dramatically increased tolerance to antibiotics.
A recent study from the team of Prof. Dr. Tom Coenye at the Laboratory of Pharmaceutical Microbiology, University of Ghent, published in npj Biofilms and Microbiomes, shows how alternative analytical techniques combined with machine learning can improve predictions of antibiotic susceptibility in Pseudomonas aeruginosa biofilms — a major challenge in chronic infections such as cystic fibrosis.
Traditional antimicrobial susceptibility tests (ASTs) often work well for free-floating cells, but they frequently fall short when applied to biofilms. In diseases such as cystic fibrosis, Pseudomonas aeruginosa forms persistent biofilms that can withstand treatments deemed effective by standard ASTs. Bridging this gap requires tools that can capture biofilm-specific physiology rather than relying on planktonic measurements.
To address this, the researchers experimentally evolved P. aeruginosa biofilms in a physiologically relevant synthetic cystic fibrosis medium. These biofilms were then characterized using a diverse set of analytical methods:
This multi-modal dataset provided a rich foundation for training machine learning models to predict antibiotic susceptibility.
A particularly novel aspect of the study was the use of isothermal microcalorimetry with the calScreener. IMC measures the real-time heat flow generated by microbial metabolism—without exposing the biofilms to antibiotics at all. Remarkably, a single thermogram was enough to capture subtle metabolic signatures associated with antibiotic tolerance.
More tolerant biofilm strains showed features such as:
These patterns, invisible to conventional assays, turned out to be highly informative.
When the models were validated using clinical isolates, IMC stood out. It showed the strongest predictive power among all tested techniques, particularly for predicting biofilm prevention concentrations (BPCs). Its performance was far better than random baselines, highlighting that untreated biofilm metabolism already contains valuable clues about antibiotic response.
Biofilm physiology fundamentally reshapes how bacteria respond to antibiotics, yet most diagnostic tools still ignore it. This study demonstrates that isothermal microcalorimetry offers a rapid, label-free, and highly sensitive way to capture biofilm-specific metabolic phenotypes. By combining IMC with machine learning, researchers move closer to biofilm-relevant susceptibility testing, a crucial step toward more effective, personalized treatment strategies for chronic infections.
Learn more
Read more about biofilm testing with calScreener biocalorimeter and clinical research using biocalorimetry.