Raman spectroscopy can identify bacteria from their molecular vibrations — each species has a spectral fingerprint. But current protocols require opening the culture plate, isolating colonies, and preparing samples. This preparation takes time and risks contamination, which limits clinical adoption.
This paper identifies bacteria directly through the unopened, inverted agar plate. The laser passes through the polystyrene lid and agar medium to reach the bacteria growing on the surface, then the scattered light returns through the same path. The interrogation traverses thick background material that generates its own Raman signals, which should bury the bacterial signature.
It doesn't. By combining density functional theory computations (to predict and model the background signals) with machine learning (to separate bacterial features from substrate interference), they achieve over 97.7% classification accuracy between two E. coli variants differing only in GFP expression. The method works across diverse media and substrate conditions. The technique enables real-time monitoring without breaking the sterile seal — you can watch bacteria grow, test antimicrobial responses, and classify species without ever opening the plate. The background that should be noise becomes a computable, subtractable signal.