Scientists from TU Delft, SoundCell and RHMDC (laboratory of Reinier de Graaf hospital) have discovered that different bacterial species produce their own characteristic sounds. Building on an earlier development from the same team, they have now shown that bacteria can be identified and their antibiotic susceptibility determined simultaneously, based solely on their sound. This combined approach delivers results within hours instead of days, offering a major step forward in the diagnosis and treatment of bacterial infections. The study is published in ACS Sensors .
-In previous work we discovered that nanodrums made of graphene were able to capture the subtle sound of a single bacterium. We used this concept to determine which antibiotics would be effective, which eventually resulted in TU Delft spin-off SoundCell,- explains associate professor Farbod Alijani. -With this new study, we take a significant leap forward: we show that each bacterial species has its own nanomotion signature.- They developed a machine learning model that can recognise a bacterial species by analysing the unique nanoscale vibrational pattern it produces.
-By combining SoundCell’s existing antimicrobial testing prototype with this machine learning model, we can identify the bacterial infection and determine which drug is effective at the same time, based purely on the sound of a single bacterium,- says SoundCell CTO, Aleksandre Japaridze. Leo Smeets, physician microbiologist RHMDC adds: -This approach eliminates the need for culturing, which normally takes days. And because the diagnostic steps are no longer performed sequentially, we can save even more time.-
Vibrational footprints
Over the years, SoundCell and TU Delft have gathered an enormous amount of vibrational data of bacteria. -At some point we asked ourselves: why not also use these data to see whether identification is possible?- Alijani explains. The researchers focused on three bacteria that are common in hospital settings: E. coli, S. aureus and K. pneumoniae. When attached to a graphene membrane, each bacterium generates nanoscale vibrations which are recorded in real time and transformed into time-frequency spectrograms, a sort of vibrational footprints. PhD candidate Santiago Mendoza Silva developed a Machine Learning model that acts as a classifier: by training on large amounts of data, it learns how to distinguish these different vibrational signatures. -That is how we discovered that different bacteria really do make different sounds-, says Mendoza Silva.
Fighting antimicrobial resistance
According to the team, the dual identification and antibiotic sensitivity testing has enormous potential. SoundCell is currently conducting studies at RHMDC and Erasmus Medical Center with its prototype. -We have already shown that we can reduce antimicrobial susceptibility testing to one hour,- says Japaridze. -If we can combine that speed with species classification using the new machine learning model, we could create a globally unique device that dramatically accelerates diagnosis and treatment. And that would be highly valuable in the worldwide fight against antimicrobial resistance.-
Unique collaboration
Alijani emphasises the strength of the collaboration. -This close partnership between scientists at TU Delft, a start-up and a hospital is quite unique. We have the entire knowledge chain working together.- The next step is to implement the machine learning tool with SoundCell’s prototypes and to test it in hospital settings, bringing this technology another step closer to real world impact.
Antibiotic screening with bacterial sound
When a bacterial cell attaches to a extremely thin, graphene drum, its biological activity produces nanoscale oscillations that can be measured and converted into sound. These vibrations indicate that the bacterium is alive. This discovery, made by TU Delft in 2022, has major potential for detecting antibiotic resistance: drug susceptible bacteria fall silent after exposure to antibiotics, while resistant bacteria keep vibrating.
