Clostridioides difficile infection (CDI) has been identified as one of the leading causes of healthcare-associated infections (HAI) and antibiotic-associated diarrhoea worldwide. Additionally, it is of increasing concern in the community [1]. The most recent European survey conducted in 2018–2019 reported a median incidence of CDI cases in hospital facilities of 4.1 per 10,000 bed-days in Europe and 2.3 in France [2]. CDI is associated with significant morbidity, mortality and economic burden [1]. Early and accurate diagnosis and prevention of CDI are therefore essential.

Some C. difficile strains are toxigenic, meaning they carry toxin genes that enable the organism to produce both toxin A (TcdA) and toxin B (TcdB), which act synergistically to mediate the pathogenicity of the organism [1]. Current diagnosis of CDI relies on both clinical criteria (such as diarrhoea or pseudomembranous colitis) and microbiological criteria (including the presence of (TcdA) and (TcdB) in stools detected by immuno-enzymatic tests, or the presence of a toxigenic strain detected by toxigenic culture or molecular methods) [3].

The use of matrix-assisted laser desorption/ionisation time-of-flight mass spectrometry (MALDI-TOF-MS) has increased worldwide over the past decade as an innovative alternative to conventional microbial identification based on biochemical properties [4]. In addition to rapid, accurate and cost-effective bacterial identification, MALDI-TOF MS has been used for other applications, including the detection of specific virulent strains and antimicrobial resistance, as well as epidemiological analysis [[5], [6], [7]]. Artificial intelligence techniques, including machine learning (ML) and deep learning (DL) using neural networks, have shown promise in analysing and designing algorithms (also called classifiers) to improve the results generated by MALDI-TOF MS. For instance, Calderaro et al. reported on the rapid typing of circulating C. difficile strains using ML-based algorithms. The study found an advantage in turnaround time compared to polymerase chain reaction (PCR)-ribotyping results [8]. Recently, another team has successfully evaluated (with greater than 95 % accuracy) the use of MALDI-TOF MS in conjunction with ML to distinguish binary toxin-producing strains from non-binary toxin-producing strains [9]. In addition, Godmer et al. showed that MALDI-TOF coupled with DL-based algorithms could accurately exclude ToxB−A− spectra in 95 % of cases. The specificity for identifying ToxB + ToxA + CDT + strains was >96 % [10]. All of these studies analyzed MALDI-TOF MS spectra of laboratory-cultivated strains and compared them with results from multiplex PCR for detecting the genes of the main virulence factors (tcdA, tcdB, and cdt).

This study aimed to evaluate the performance of MALDI-TOF MS combined with ML and DL-based algorithms for identifying toxigenic C. difficile strains, and compared it to whole genome sequencing and toxin gene detection.