Alagna Nicolò, Dúzs Brigitta, Dietrich Vincent, Younesi Ali Tayefeh, Lehmann Livia, Ulbricht Ronald, Köppl Heinz, Walther Andreas, Gerber Susanne
Institute of Human Genetics, University Medical Center, Mainz, Germany.
Department of Chemistry, Johannes Gutenberg University, Mainz, Germany.
Commun Chem. 2025 May 15;8(1):153. doi: 10.1038/s42004-025-01541-y.
Model-based analysis is essential for extracting information about chemical reaction kinetics in full detail from time-resolved data sets. This approach combines experimental hypotheses with mathematical and physical models, enabling a concise description of complex system dynamics and the extraction of kinetic parameters like kinetic pathways, time constants, and species amplitudes. However, building the final kinetic model requires several intermediate steps, including testing various assumptions and models across multiple experiments. In complex cases, some intermediate states may be unknown and are often simplified. This approach requires expertise in modeling and data comprehension, as poor decisions at any stage during data analysis can lead to an incorrect kinetic model, resulting in inaccurate results. Here, we introduce DLRN, a new deep learning-based framework, designed to rapidly provide a kinetic reaction network, time constants, and amplitude for the system, with comparable performance and, in part, even better than a classical fitting analysis. We demonstrate DLRN's utility in analyzing multiple timescales datasets with complex kinetics, different 2D systems such as time-resolved spectra and agarose gel electrophoresis data, experimental datasets as nitrogen vacancy and strand displacement circuit (using photoluminescence and transient absorption techniques), even in scenarios where the initial state is a hidden, non-emitting dark state.
基于模型的分析对于从时间分辨数据集中全面详细地提取化学反应动力学信息至关重要。这种方法将实验假设与数学和物理模型相结合，能够简洁地描述复杂系统的动力学，并提取诸如动力学途径、时间常数和物种振幅等动力学参数。然而，构建最终的动力学模型需要几个中间步骤，包括在多个实验中测试各种假设和模型。在复杂情况下，一些中间状态可能未知且常常被简化。这种方法需要建模和数据理解方面的专业知识，因为数据分析过程中任何阶段的错误决策都可能导致错误的动力学模型，从而产生不准确的结果。在此，我们介绍DLRN，一种基于深度学习的新框架，旨在快速为系统提供动力学反应网络、时间常数和振幅，其性能相当，部分甚至优于经典拟合分析。我们展示了DLRN在分析具有复杂动力学的多个时间尺度数据集、不同的二维系统（如时间分辨光谱和琼脂糖凝胶电泳数据）、实验数据集（如氮空位和链置换电路，使用光致发光和瞬态吸收技术）方面的效用，甚至在初始状态为隐藏的、不发光的暗态的情况下也是如此。
