Nanoscopy

Scattering-type scanning near-field optical microscopy (s-SNOM)-based nano-FTIR spectroscopy was employed to distinguish individual SARS-CoV-2 and influenza A virus particles. These viruses exhibit similar morphological features but possess distinct biochemical compositions, enabling their classification through nano-FTIR spectral signatures. However, nano-FTIR throughput is constrained by instrumental drift and optical alignment requirements, limiting the number of spectra reliably acquirable per measurement session, unlike conventional vibrational spectroscopic techniques, where automated large-scale data collection is routinely feasible. Integrating spectral information across multiple demodulation orders within a multivariate analysis and data fusion framework therefore represents a strategy to maximize the biochemical information extractable from a limited spectral dataset, without requiring large-scale data collection. It was addressed by chemometric spectral data fusion of the nano-FTIR spectra, where the spectral demodulation orders n=2,3,4 were utilized to construct a PLS model for virus classification. The PLS models were trained separately for the phase and near-field absorption spectra (Absorption=Amplitude × sin (Phase)). Each demodulation order was analyzed separately, and in addition, a data fusion model was trained using all the demodulation orders. The data fusion model of phase and near-field absorption spectra demonstrated high performance for single spectral analysis (i.e., analysis performed on individual spectra) with balanced accuracies of 96.5 and 98.6% for near-field absorption and phase, respectively, using a majority voting approach. Particle-level analysis (i.e., aggregation of spectra belonging to the same particle) using the mean spectrum achieved balanced accuracies of 97.5 and 100% for near-field absorption and phase, respectively. By fusing spectral data across all nano-FTIR demodulation orders, we achieved robust virus classification by integrating biochemical information from each spectral order. This approach provides a detailed characterization of both surface and subsurface chemical signatures, enabling comprehensive analysis at the single-virus level.