Spectroscopic instruments are increasingly being implemented in the search for extraterrestrial life. However, microstructural spectral analyses of alien environments could prove difficult without knowledge on the molecular identification of individual spectral signatures. To bridge this gap, we introduce unsupervised K-means clustering as a statistical approach to discern spectral patterns of biosignatures without prior knowledge of spectral regions of biomolecules. Spectral profiles of bacterial isolates from analogous polar ice sheets were measured with Raman spectroscopy. Raman analysis identified carotenoid and violacein pigments, and key cellular features including saturated and unsaturated fats, triacylglycerols, and proteins. Principal component analysis and targeted spectra integration biplot analysis revealed that the clustering of bacterial isolates was attributed to spectral biosignatures influenced by carotenoid pigments and ratio of unsaturated/saturated fat peaks. Unsupervised K-means clustering highlighted the prevalence of the corresponding spectral peaks, while subsequent supervised permutational multivariate analysis of variance provided statistical validation for spectral differences associated with the identified cellular features. Establishing a validated catalog of spectral signatures of analogous biotic and abiotic materials, in combination with targeted supervised tools, could prove effective at identifying extant biosignatures.

中文翻译：

---

用多元统计研究拉曼光谱特征：一种微生物生物特征分类方法

光谱仪器越来越多地用于寻找外星生命。然而，如果不了解个体光谱特征的分子识别，外星环境的微观结构光谱分析可能会很困难。为了弥补这一差距，我们引入了无监督的K-意味着聚类作为一种统计方法来辨别生物特征的光谱模式，而无需事先了解生物分子的光谱区域。用拉曼光谱测量来自类似极地冰盖的细菌分离物的光谱分布。拉曼分析确定了类胡萝卜素和紫罗兰素色素，以及包括饱和和不饱和脂肪、三酰基甘油和蛋白质在内的关键细胞特征。主成分分析和靶向光谱积分双图分析表明，细菌分离物的聚类归因于受类胡萝卜素色素和不饱和/饱和脂肪峰比率影响的光谱生物印记。无监督K-均值聚类突出了相应光谱峰的普遍性，而随后的监督置换多变量方差分析为与已识别的细胞特征相关的光谱差异提供了统计验证。建立一个经过验证的类似生物和非生物材料的光谱特征目录，结合有针对性的监督工具，可以证明有效地识别现存的生物特征。