Determination of monophenolase activity based on backpropagation neural network analysis of three-dimensional fluorescence spectroscopy

Tyrosinase (EC 1.14.18.1), pivotal for melanin formation, is widely present in microorganisms, plants, animals, and humans and has various biological functions, including human skin pigmentation, wound healing, insect sclerotization, and immunity (Chang, 2009). Not only is tyrosinase involved in medicine, cosmetics, food, and agriculture (Xu and Yoon, 2011), but also used to synthesize substituted o-diphenols (Min et al., 2019). It is a unique bifunctional enzyme with both monophenolase and diphenolase activities. Monophenolase catalyzes the hydroxylation of tyrosine to L-DOPA (L-3,4-dihydroxyphenylalanine), while diphenolase catalyzes the oxidation of L-DOPA to L-dopaquinone. Monophenolase activity has attracted interest because it functions as a rate-limiting step (Min et al., 2010).

Measurement of monophenolase activity might provide profound insights into the catalytic mechanism and biochemical processes involving tyrosinase, and the application of tyrosinase in the fields of clinical diagnosis, inhibitor screening, food quality control, environmental monitoring, and synthetic organic chemistry (Qu et al., 2020, Guo et al., 2020a). Therefore, it is of great urgency to develop a monophenolase assay for both fundamental research and industrial applications (Teng et al., 2015). Monitoring monophenolase activity remains a huge challenge (Ma et al., 2019, Zhang et al., 2022a). Several methods have been reported to measure monophenolase activity, including spectrophotometric, colorimetric, oximetric, chromatographic, and radiometric methods (Qu et al., 2020, Du et al., 2021a). The most widely used continuous spectrophotometric method is based on measuring the optical absorbance at 475 nm for dopachrome or 505 nm for the coupling reaction between dopaquinone and MBTH (3-methyl-2-benzotheiazolinone) (Zeyer et al., 2018). The oximetric method is based on the measurement of the volume of oxygen consumption using an oxygen electrode (Yamazaki and Itoh, 2003a). It is unlikely for the spectrophotometric and oximetric methods to correctly trace the true monophenolase reaction because they account for the sum of monophenolase and diphenolase activities (Zeyer et al., 2018, Zhang et al., 2021a). Chromatographic and radiometric methods are discontinuous assays and have cumbersome procedures (Olsovská et al.). Monitoring tyrosine consumption over time is deemed an appropriate method for the exclusive monophenolase assay (Guo et al., 2020a). However, serious spectral overlapping leads to failure in the selective detection of the tyrosine substrate in the complex system of an enzymatic mixture containing the product L-DOPA (Guo et al., 2020a). Moreover, the real-time monitoring of kinetic and nonlinear changes during dynamic processes is required for continuous determination of enzyme activity. In our previous study, the native fluorescence of L-DOPA was quenched by boric acid, and diphenolase activity was exclusively blocked (Guo et al., 2020a, Du et al., 2021a, Zhang et al., 2021b). However, boric acid may pause the catalytic cycle for the natural state of tyrosinase (Yamazaki and Itoh, 2003a, Cieńska et al., 1932).

Three-dimensional fluorescence (excitation-emission matrix fluorescence) spectroscopy has proven to be a sensitive technique for the real-time and nondestructive analysis of complicated samples by affording large-sized data (Wu et al., 2020, Xie et al., 2015). Nevertheless, the fluorescence signal is inclined to nonlinearity for three-dimensional fluorescence (Escandar et al., 2006, Chen et al., 2014). Artificial neural network (ANN) is a promising methodology for performing nonlinear multivariate regression tasks in complex systems in the presence of overlapping spectra and matrix interference (Chen et al., 2014, Hasani et al., 2007). A backpropagation (BP) ANN is a feed-forward neural network that uses the error BP method as a supervised training procedure (Fig. S1). The hidden layer forms a weight connection between the input and output layers (Hasani et al., 2007). The model is established using a gradient descent search technique to obtain the minimum sum of the squared errors. The BP ANN has the advantages of a simple structure and easy operation (Shahlaei et al., 2015a). However, it also exhibits disadvantages, including slow training speed, susceptibility to initial weight and threshold, and easy entrapment in local minima dilemmas (Cheng et al., 2020a). Principal component analysis (PCA) can extract principal component (PC) vectors. Therefore, PCA can efficiently accelerate the convergence of the ANN and decrease the error bias by lowering the data dimension (Shahlaei et al., 2015a). Genetic algorithms (GAs) achieve an optimal value through a stochastic process in which individuals are operated in the population based on fitness. GAs can effectively circumvent the drawback of the BP network by globally optimizing the weights and thresholds to provide an optimal solution (Wang et al., 2020a).

Enzymatic reactions are complex systems and dynamic processes involving diverse components that vary both temporally and nonlinearly (Zhang et al., 2018). In the current study, a hybrid artificial intelligence algorithm was developed to determine monophenolase activity using a combination of PCA, GAs, and BP neural networks (Scheme 1A). First, a total of 121 factors of the fluorescence spectra were decomposed into four PC scores using PCA as input variables for the BP. Second, GAs were used to determine the optimal structural parameters of the BP network. Finally, the network model was trained using the BP learning algorithm to predict the output of the unknown samples (Scheme 1B).

There are few reports concerning the use of ANN to resolve the issues associated with enzyme activity assays. The strategy of coupling both the neural network and molecular fluorescence has the advantages of full selectivity and high sensitivity for the real-time quantification of complex dynamic systems. The rapid acquisition of three-dimensional fluorescence spectroscopy allowed for the real-time and in-situ monitoring of dynamic processes (Damiani et al., 2007, Yin et al., 2015). The ANN successfully resolved the spectral overlap between tyrosine and L-DOPA in complex enzymatic mixtures through mathematical separation. The advantages of the three techniques in the PCA-GA-BP model were fully exploited to enhance the prediction capability of the model and the efficiency of network training (Shahlaei et al., 2015a, Wang et al., 2020a). The proposed method provides a new solution for conventional enzyme activity assays. To the best of our knowledge, this is the first ANN study to measure tyrosine in enzymatic reactions and determine monophenolase activity.

留言 (0)

沒有登入
gif