Spectroscopic sensing · 2026
Validation practices in label-free microbial SERS
Reviewing 129 microbial SERS studies to examine how sample preparation, modelling and validation affect generalisability.
Evidence review
Approach
How are machine learning models for label-free microbial SERS validated, and does the evidence support generalisation?
The review tracked six stages, from sample preparation and spectral acquisition to validation and deployment. It audited how studies split spectra and biological units, whether evaluation used external cohorts and devices, and what evidence supported generalisation.
- Collect 129 microbial SERS studies
- Audit each stage of the workflow
- Assess validation and generalisation
Reported results
The audit found that 68.2% of studies lacked validation that kept biological samples independent across splits. Only 6.2% reported external or major-domain validation.
My contribution
I designed the review, led the corpus audit and evidence synthesis, and prepared the figures, tables and manuscript.
Study scope
The percentages describe the studies in the review corpus and the information those publications reported.