Brito Anna Luiza B, Brüggen Carlotta, Ildiz Gulce Ogruc, Fausto Rui
CQC-IMS, Department of Chemistry, University of Coimbra, P-3004-535 Coimbra, Portugal.
Biochemistry Center, Heidelberg University, P-69120 Heidelberg, Germany.
Spectrochim Acta A Mol Biomol Spectrosc. 2022 Jul 5;275:121175. doi: 10.1016/j.saa.2022.121175. Epub 2022 Mar 21.
The ending of estrogen production in the ovaries after menopause results in a series of important physiologic changes, including hair texture and growth. In this study we demonstrate that Raman spectroscopy can be used successfully as a tool to probe menopause-induced changes on hair, in particular when coupled with suitable chemometrics approaches. The detailed analysis of the average Raman spectra (in particular of the Amide I and νS-S stretching spectral regions) of the hair samples of women pre- and post-menopause allowed to estimate that absence of estrogen in post-menopause women leads to an average reduction of ∼12% in the thickness of the hair cuticle, compared to that of pre-menopause women, and revealed the strong prevalence of disulphide bonds in the most stable gauche-gauche-gauche conformation in the hair cuticle. From the analysis of the νS-S stretching spectral region it could also be concluded that the amount of α-helix keratin is slightly higher for post-menopause than for pre-menopause women. A series of statistical models were developed in order to classify the hair samples. Outperforming the traditional PCA-LDA (principal component analysis - linear discriminant analysis) approach, in the present study a GA-LDA (genetic algorithm - linear discriminant analysis) strategy was used for variable reduction/selection and samples' classification. This strategy allowed to develop of a statistical model (L16), which has exceptional prediction capability (total accuracy of 96.6%, with excellent sensitivity and selectivity) and can be used as an efficient instrument for the hair samples' classification. In addition, a new chemometrics approach is here presented, which allows to overcome the intrinsic limitations of the GA algorithm and that can be used to develop statistical models that use GA as the variable reduction/selection method, but superseding its stochastic nature. Three suitable models for classification of the hair samples according to the menopause status of the women were developed using this novel approach (LV17, BLV20 and PLS7 models), which are based on the Fisher's and Bayers' LDA approaches and the PLS-DA method. The followed new chemometrics approach uses the results of a large set of GA-LDA runs over the full data matrix for the selection of the reduced data matrices. The criterion for the selection of the variables is their statistical significance in terms of number of occurrences as solutions of the whole set of GA-LDA runs.
绝经后卵巢雌激素分泌的终止会导致一系列重要的生理变化，包括头发质地和生长。在本研究中，我们证明拉曼光谱可以成功用作探测绝经引起的头发变化的工具，特别是与合适的化学计量学方法结合使用时。对绝经前和绝经后女性头发样本的平均拉曼光谱（特别是酰胺I和νS-S伸缩光谱区域）进行详细分析后估计，与绝经前女性相比，绝经后女性体内雌激素的缺乏导致头发角质层厚度平均减少约12%，并揭示了在头发角质层中最稳定的左-左-左构象中，二硫键占主导地位。通过对νS-S伸缩光谱区域的分析还可以得出结论，绝经后女性α-螺旋角蛋白的含量略高于绝经前女性。为了对头发样本进行分类，开发了一系列统计模型。在本研究中，一种GA-LDA（遗传算法-线性判别分析）策略用于变量约简/选择和样本分类，其性能优于传统的PCA-LDA（主成分分析-线性判别分析）方法。该策略使得能够开发出一个统计模型（L16），该模型具有出色的预测能力（总准确率为96.6%，具有出色的灵敏度和选择性），可作为头发样本分类的有效工具。此外，本文提出了一种新的化学计量学方法，该方法可以克服GA算法的固有局限性，并且可以用于开发以GA作为变量约简/选择方法的统计模型，但取代了其随机性。使用这种新方法（LV17、BLV20和PLS7模型）开发了三种根据女性绝经状态对头发样本进行分类的合适模型，这些模型基于费舍尔和贝叶斯的LDA方法以及PLS-DA方法。随后的新化学计量学方法使用在完整数据矩阵上进行的大量GA-LDA运行结果来选择约简后的数据矩阵。变量选择的标准是它们作为整个GA-LDA运行集的解在出现次数方面的统计显著性。
