Utilization of random forest approaches to obtain information on biomolecule structure and interaction from SERS experiments

表面增强拉曼散射(SERS)是一种在没有标记、标记或报告的情况下获得有关生物分子样品组成和结构的全面和多样化信息的有效方法。由于SERS的性质是探测分子与纳米结构金属衬底的不同相互作用以及特定官能团的接近程度，因此很难利用产生的高度复杂的数据。生物样品具有许多不同的生物分子和变化的条件，影响了所获得的SERS光谱，这对生物样品提出了具体的挑战。这就是为什么需要有效和公正地利用SERS数据的原因。在这个项目中，基于随机森林(RF)的方法被证明在应用于真实的SERS数据时非常稳健，将被用于在明确定义的实验条件下获得的模型数据的分析。这里分析的系统的复杂性将从单个分子组件，如脂膜或药物分子的积木，超过两个组件的组合，逐步增加到培养细胞内溶酶体囊泡的复杂环境。RF分析将以这样的方式建立，它可以服务于三个不同的目的：(I)选择重要的光谱特征用于直接结构解释，(Ii)识别共现的光谱特征以分析不同分子的相互作用，以及(Iii)整合来自先前实验的先验知识，包括与本项目中的各个其他模型进行的实验。这种多功能框架的发展将取决于SERS实验的不同复杂程度，以及一套以系统方式界定和修改的实验条件。另一方面，为模拟特定实验条件的影响而专门生成的模拟数据将以迭代的方式与实际实验进行比较。这一项目的成功完成将标志着在利用表面增强拉曼光谱对明确定义的生物物理模型以及复杂生物系统中的分子进行结构表征方面迈出了重要的一步。

Surface-enhanced Raman scattering (SERS) in the absence of labels, tags or reporters is a powerful method to obtain comprehensive and diverse information about the composition and structure of biomolecular samples. Because of the nature of SERS that probes the varying interaction of the molecules with a nanostructured metal substrate and the proximity of specific functional groups, it is difficult to exploit the highly complex data that are generated. Specific challenges are posed by biological samples that are characterized by many different biomolecules and changing conditions, influencing the obtained SERS spectra. This is why efficient and unbiased approaches for the utilization of SERS data are needed. In this project, random forest (RF) based methods, which have been shown to be very robust when applied to real SERS data, will be adapted and used for the analysis of model data obtained under well-defined experimental conditions. The complexity of the systems that are analyzed here will be increased step-wise from individual molecular components such as building blocks of lipid membranes or a drug molecule, over the combination of two components, to the complex environment of an endolysosomal vesicle inside cultured cells. The RF analysis will be established in such a way that it can serve three different purposes: (i) the selection of important spectral features for direct structural interpretation, (ii) the identification of co-occurring spectral features to analyze the interaction of different molecules, and (iii) the integration of a priori knowledge from previous experiments, including those conducted with respective other models in this project. The development of such a multi-purpose framework will rely on the different degrees of complexity of the SERS experiments as well as on a set of experimental conditions that are defined and modified in a systematic way. As a further aspect, simulated data that are specifically generated to imitate the impact of a particular experimental condition will be compared with actual experiments in an iterative fashion. The successful completion of this project will mark an important step towards utilizing SERS for the structural characterization of well-defined biophysical models as well as of molecules in complex biological systems.