File Name : fig. s1.tif

Caption : fig. s1: (a) schematic diagram of the micro- raman setup: 785 nm diode laser, bpf-band pass filter, be-beam expander, l2, l3, l4 and l5-focussing lenses, dm-dichroic mirror, path selector, edge filter, spectrometer with ccd, ccd-camera, pc-personal computer, mo-microscope objective and sample cell, (b) typical raman spectra of (i) lyz and sers of lyz (ii).

File Name : fig. s2.tif

Caption : fig. s2: uv-absorption spectra of (a) gold nanoseed colloidal solution and (b) uv-absorption spectra of gold nanostar colloidal solution.



File Name : fig. s3.tif

Caption : fig. s3: (a) sers spectra of lyz (100 nm) recorded at five different spots on the substrate (laser power: 6.2 mw, acquisition time: 10 s), showing the reproducible peak at 1664 cm⁻¹ and (b) peak intensities at 1664 cm⁻¹ from the five spots, with mean value and standard deviation.

File Name : fig. s3 _c_.tif

Caption : fig. s3: (c) sers (average of 25 spectra) of lyz from five different spots.

File Name : fig. s4.tif

Caption : fig. s4: (a) results of leave-one-out cross-validation for the knn classifier. the misclassification curve displays correct predictions (green at baseline) and errors (red above baseline), showing that only a small number of scattered misclassifications occurred and (b) the prediction plot compares true classes with predicted outputs, illustrating that classes c and poag were classified with high accuracy.

File Name : fig. s5.tif

Caption : fig. s5: (a) results of leave-one-out cross-validation for the knn classifier. the misclassification curve displays correct predictions (green at baseline) and errors (red above baseline), showing that only a small number of scattered misclassifications occurred and (b) the prediction plot compares true classes with predicted outputs, illustrating that classes c and poag were classified with high accuracy.