Highly sensitive and precise SERS quantification of P. gingivalis in complex saliva matrices via AgNPs@4-MPBA substrates coupled with the NiPLS-AMLP algorithm.

Precise quantitative detection of Porphyromonas gingivalis (P. gingivalis, Pg) is important for developing a potential non-invasive screening platform for key periodontal pathogens. To address the challenge of specific P. gingivalis detection in complex saliva matrices by surface-enhanced Raman scattering (SERS), this study developed an intelligent quantitative sensing platform combining a SERS substrate based on 4-mercaptophenylboronic acid-functionalized silver nanoparticles (AgNPs@4-MPBA) with a nested interval partial least squares-attention mechanism multilayer perceptron (NiPLS-AMLP) framework. Based on the enhancement of the C-S-related characteristic peak at 674 cm-1 and the bacterial control experiments, 4-MPBA may interact with the O-glycosylated gingipain glycans on the P. gingivalis outer membrane through its terminal boronic acid group, thereby promoting bacterial capture. Meanwhile, interactions between 4-MPBA and gingipain-related cysteine residues may contribute to the enhancement of the C-S-related Raman signal. For data analysis, the NiPLS-competitive adaptive reweighted sampling (CARS) strategy identified the 650-699 cm-1 core spectral region, providing biochemical interpretability, whereas the AMLP model improved the quantitative accuracy for low-concentration P. gingivalis signals. To mitigate matrix effects caused by artificial and real human saliva matrices, semi-supervised transfer learning was further introduced to reduce cross-matrix domain shift. The platform achieved accurate P. gingivalis quantification in complex saliva matrices over the range of 2 x 102 to 2 x 109 CFU/mL, with an R2 of 0.98 and a limit of detection of 200 CFU/mL. These results support the use of this platform as a proof-of-concept approach for potential non-invasive P. gingivalis screening in complex saliva matrices and provide a strategy for intelligent SERS-based quantification of pathogens in complex biological samples.