Answers are grounded in this paper and cite their sources.
The stability of a spectral reference for external measurements cannot be inferred from the variance explained within that reference alone. We investigated reference sensitivity using Fourier-transform infrared spectra of gingival crevicular fluid from 17 adults and 25 children acquired under different protocols. Principal component analysis (PCA) was fitted exclusively to adult spectra, and the same pediatric spectra were projected into models with different reference compositions and retained dimensionalities (K = 2–5). Three components explained 94.1% of adult variance. Omitting one adult record changed the retained PCA subspace (maximum principal angle, 73.1°) and increased the median pediatric residual fraction from 0.440 to 0.769. At higher dimensionalities, similar aggregate residual fractions did not ensure agreement in individual rankings. At K = 3, the largest baseline mean squared residual occurred near 991 cm−1, whereas the largest local increase following record omission occurred near 2922 cm−1. Algebraic decomposition showed that changing the reference mean alone did not explain the increase. Descriptive age and sex summaries also changed with reference composition. Together, these findings demonstrate that reference composition can substantially alter the representation of unchanged external spectra, highlighting the importance of assessing sensitivity at aggregate, individual, and spectral-variable levels.
No takes yet. Share an insight, caveat, or question.
Litvinova et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: