Machine learning assisted optical biosensing of bacteria on a monolayer MoS2 platforms

Bacterial infections continue to impact billions of lives worldwide. Current detection methods are struggling to provide both rapid and accurate bacterial detection. Therefore, innovative approaches to rapid disease detection are urgently required. Recently, there has been much interest in the space of 2D material-based sensing. In particular, transition metal dichalcogenides (TMDs) have emerged as suitable biosensor candidates, due to their unique properties. These materials are relatively easy to make, non-toxic and have distinct charge characteristics. This study explores the integration of Molybdenum disulfide (MoS 2 ) into a monoclonal antibody (mAb) functionalized biosensor which successfully detects methicillin-resistant Staphylococcus aureus (MRSA) using both Raman and photoluminescence (PL) spectroscopy. 2D nanoflakes of MoS 2 were deposited onto a Silicon oxide (SiO 2 ) substrate via chemical vapour deposition (CVD). The resulting MoS 2 chips were functionalised with F598, a mAb that binds to the polysaccharide poly -N- acetyl-glucosamine (PNAG) found in various microbes, including many diverse species of bacteria. MRSA was incubated onto the MoS 2 chips for 30 min, before rinsing with phosphate buffered saline (PBS), then PL and Raman spectroscopy were performed on the samples. Our results show accurate and rapid detection of MRSA using both PL and Raman spectroscopy, based on spectral changes that occur because of bacterial adhesion onto the chip surfaces. Machine learning was then utilized in a convolutional neural network (CNN) to distinguish between the different classes of spectra (controls and MRSA). Using a CNN an accuracy of 86% and 96% was found for the Raman and PL data, respectively. Thus, with a platform of MoS 2 nanoflakes, generalized detection of PNAG-expressing bacteria (in this case MRSA) using spectroscopic techniques is successful. As this platform's detection is based on the antibody used, the platform can be modified to detect various species of bacteria by using different monoclonal antibodies. • Antibody-functionalised monolayer MoS₂ enables rapid, label-free MRSA detection. • Raman and photoluminescence spectroscopy reveal clear optical signatures of bacterial adhesion. • Convolutional neural network achieves 86–96% classification accuracy from spectra. • Platform adaptable for diverse PNAG-expressing bacteria via antibody substitution.