Project plus 4

Data preprocessing was performed using the WITec Project Plus 4.0 software (WITec, Germany). Cosmic ray removal was carried out before any further analysis based on an intensity threshold set by taking into account spectral and spatial pixels adjacent to the pixel of interest. All spectra were cut to the spectral region from 300 to 1800 cm⁻¹ before background subtraction and multivariate data analysis.
In order to assess the influence of the background subtraction on the unmixing algorithm output, analyses were done for both datasets with and without previous background correction (based on fitting a polynomial of order 3 and performed in the WITec Project Plus 4.0 software). For Spruce, a rank of 4 endmembers with and without background subtraction was taken for all methods, whereas when 5EM were applied, only the comparison for background subtracted data is shown. For Arabidopsis, the results are discussed with and without background subtraction based on the VCA analysis. The comparison of all approaches on the Arabidopsis dataset is based on the data without background subtraction.

Raman images (20 × 20 µm², 400 spectra) of individual cells of a total of seven cell lines (CHO‐K1, CHO‐K1‐hDAO, CHO‐S‐Humira, CHO‐S, CHO‐S/4F11, CHO‐K11D9 and CHO‐K14F10) were background subtracted (using a polynomial function of degree 3) and k‐means cluster analyzed with the following conditions: four clusters, spectral mask 400–1800 cm⁻¹ (fingerprint region), Manhattan normalization mode (area under the spectral mask equal to 1), no data reduction and no pre‐transformation mode (Witec Project Plus 4.0, Witec, Germany). By this the hyperspectral image is processed to four images presenting the chemically most different four areas (clusters) and the corresponding average spectrum for each distinguished area. One cluster was assigned to ER due to higher intensity of the protein bands, one to background (which was discarded) and the remaining two clusters were more similar and merged as ”rest of the cell“ (RC). The average spectra of ER and RC of each cell, obtained from the clustering, were used in a ”principal component analysis“ (PCA) based on the second derivative of those spectra (Savitzky‐Golay Algorithm, 17 smoothing points) and the fingerprint region of 400–1800 cm⁻¹ (Unscrambler X 10.3, CAMO Software AS, USA).