Impact of Imperfect Data on the Performance of Algorithms to Compare Near-Ultraviolet Circular Dichroism Spectra.

There is growing interest in the use of algorithms to objectively compare near-UV spectra of protein biopharmaceuticals in a regulated environment. Such use will require that the methods be validated, with International Conference on the Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH) Q2(R1) currently being the key document. A key aspect of such validation is to understand how robust the method is to experimental variation. Noise-free simulated spectra, obtained by fitting multiple Gaussian peaks to experimental data obtained from a pharmaceutical protein, were used to assess the robustness of several algorithms in response to spectral data "imperfections". Sources and magnitudes of these imperfections were derived from published inter-laboratory studies. Spectral noise, wavelength calibration errors, intensity variation, and spectral offset errors were "titrated" into the noise-free simulated spectrum and imperfect data sets were compared with the simulated data using a variety of published algorithms, including Pearson, Prestrelski, and derivative correlation algorithms, and spectral overlap, spectral difference and weighted spectral difference methods, to understand how robust outputs are to imperfect data. Algorithm was assessed by comparing their sensitivity to imperfect data against the pairwise statistical variation between 20 replicate spectra.