Toward smart and in-situ mycotoxin detection in food via vibrational spectroscopy and machine learning

Abstract

Recent advances in vibrational spectroscopy combined with machine learning are enabling smart and in-situ detection of mycotoxins in complex food matrices. Infrared and spontaneous Raman spectroscopy detect molecular vibrations or compositional changes in host matrices, capturing direct or indirect mycotoxin fingerprints, while surface-enhanced Raman spectroscopy (SERs) amplifies characteristic mycotoxins molecular vibrations via plasmonic nanostructures, enabling ultra-sensitive detection. Machine learning further enhances analysis by extracting subtle and unique mycotoxin spectral features from information-rich spectra, suppressing noise, and enabling robust predictions across heterogeneous samples. This review critically examines recent sensing strategies, model development, application performance, non-destructive screening, and potential application challenges, highlighting strengths and limitations relative to conventional methods. Innovations in portable, miniaturized spectrometers integrated with cloud computation are also discussed, supporting scalable, rapid, and on-site mycotoxin monitoring. By integrating state-of-art vibrational fingerprints with computational analysis, these approaches provide a pathway toward sensitive, smart, and field-deployable mycotoxin detection in food.

Keywords: Cloud computing; Mid-infrared spectroscopy; Miniaturized sensors; Mycotoxins; Near-infrared spectroscopy; Raman spectroscopy; Surface-enhanced Raman spectroscopy.

Fig. 1

Characteristics and future prospects in smart and in-situ mycotoxin detection in food by vibrational spectroscopy.

Fig. 2

Spectra of soybean meal collected using different spectroscopic techniques: (a) MIR, (b) NIR, and (c) Raman Spectrometers (Rodriguez-Saona et al., 2020); Typical settings for mycotoxins screening of (d) attenuated total reflection (ATR) in MIR spectroscopy, (e) diffuse reflection in NIR spectroscopy, and (f) transmission in NIR spectroscopy; Representative spectra of direct aflatoxins screening by (g) MIR and (h) Raman spectroscopy.

Fig. 3

Common data processing and transformation techniques, unsupervised and supervised machine learning algorithms for screening mycotoxins using vibrational spectroscopy.

Fig. 4

Schematic of the proposed SERS mechanisms, utilizing (a) bottom-up and top-down fabrication approaches (Logan et al., 2024), with (b) Ag nanoparticles (J. Yuan et al., 2017b), (c) cauliflower-inspired 3D substrates (Li et al., 2019b), (d) aptamers engineered via SERS nanotags and MRS nanoprobe (Cao et al., 2024), (e) integrated into an end-edge-cloud computing framework (Y. Wang et al., 2024) and (f) addressing challenges in IoT-based smart sensors for automated farming (Rajak et al., 2023) and (g) portable spectroscopy sensors (Rodriguez-Saona et al., 2020).