Towards a smart blood culture bottle: Machine learning assisted electrochemical profiling to provide early in-situ identification of bloodstream infection pathogens – E-MOC

Despite all improvements in the diagnosis of infectious diseases in the last 20 years, the socioeconomic burden of bloodstream infections (BSI) is still high for every country all around the world. Approximately 20% of global mortality worldwide (11 million deaths per year) is related to sepsis. Thus any improvement to accelerate BSI diagnosis that will help to shorten time to antibiotic treatment adaptation if ineffective is of major value. Indeed, in the case of bloodstream infections, early appropriate antimicrobial treatment is critical for patient survival as every hour of ineffective treatment increases mortality. That is why collecting key phenotypic features of the pathogen at a very early time point is of crucial importance. The E-MOC project investigates the concept of i) smart blood culture bottle (BCB), which relies on the electrochemical monitoring of the liquid phase of a blood culture, and ii) instrumented portable incubator. Thanks to these two breakthrough biomedical innovations, we can take advantage of the transportation time (up to 20 h) to grow and to identify bacteria to accelerate availability of early actionable results. The E-MOC project relies on the hypothesis that the time-lapse potentiometry of the electroactive fraction of metabolome can bring an in-situ label-free identification of pathogen, at least for monomicrobial BSI (85% of total). In other words, we assume that the variability of the collected fingerprint is small enough, within a given species, to allow the pathogen identification thanks to supervised machine learning, in spite of the variance of diverse parameters (patients, temperature, inoculum, etc.).
The E-MOC consortium gathers all the required complementary skills in clinical microbiology and clinical studies (CHU Grenoble Alpes and CHU Nice), artificial learning and electrochemistry (CEA-LETI), materials science and processes optimization (LGP2).
As a first objective, the multidisciplinary E-MOC consortium will build a database of electrochemical fingerprints from bacteria growing in BCB in conditions close to reality in order to allow rapid bacterial identification in new positive BCB through machine learning algorithms. Identification performances for the most frequent species involved in bacteraemia will be assessed. The fingerprints will be collected with the electrodes we made and inserted into commercial BCB, then analyzed with up-to-date classification algorithms of machine learning.
As a second objective, the consortium will work, in parallel, on designing and optimizing prototypes of instrumented portable incubator, and instrumented BCB. A first task will be dedicated to the development of a smart portable incubator. A second task, devoted to the development of our own instrumented BCB, will investigate various materials (bottle, printed chipboard, formulation of conducting inks), as well as processes to make the electrodes (screen printing, automated drop casting).