Comparison of an Ultrasound-Assisted Aqueous Two-Phase System Extraction of Anthocyanins from Pomegranate Pomaces by Utilizing the Artificial Neural Network–Genetic Algorithm and Response Surface Methodology Models

Round 1

Reviewer 1 Report

I have read and reviewed the article titled "Ultrasound-assisted aqueous two-phase system extraction of anthocyanins from pomegranate pomaces by utilizing artificial neural network-genetic algorithm and response surface methodology models: Modeling, optimization, comparison, and characterization". In order for this article to meet the required standards set by the food journal, I would like to suggest the following considerations:

1. Familiarize yourself with relevant literature: Given that the title of the article mentions keywords such as artificial neural network-genetic algorithm and response surface methodology models, it is crucial to thoroughly review existing literature in the field of food industry, particularly focusing on studies related to your own work. This will help establish a strong theoretical foundation and demonstrate your knowledge of the subject matter.

2. Explain the relevance of neural network and genetic algorithm: In the introduction, provide a clear explanation of why you have chosen to employ artificial neural network and genetic algorithm in your study. Discuss their significance in the context of the food industry and highlight their potential applications and benefits. This will help readers understand the motivation behind your research approach.

3. State your research hypotheses: Towards the end of the introduction, clearly state your research hypotheses or objectives. Outline the specific aims of your study and how you expect the utilization of artificial neural network-genetic algorithm and response surface methodology models to contribute to answering your research questions. This will provide a roadmap for the readers and set the context for the subsequent sections of the article.

4. To enhance the reader's understanding and facilitate reproducibility of your work, it is recommended to provide an outline of the general process at the beginning of the Materials and Methods section. This outline will serve as a roadmap, giving readers a high-level overview of the steps involved in your research. Subsequently, you can delve into each step in detail within the Materials and Methods section.

5.While I appreciate the alternative approach you have taken by incorporating neural network and genetic algorithm methods, I have identified several concerns that require your attention. I kindly request that you address and make the necessary corrections in the text of the article.

   a. Use of Neural Network: I found it difficult to understand the rationale behind your use of neural networks, especially considering that the CCD method is already a comprehensive statistical approach widely recognized as a reference. It would be helpful if you could provide a clearer explanation of why you chose to incorporate neural networks in your study, highlighting the specific advantages they offer in addition to the CCD method.

   b. Basis of the Work: It appears that the foundation of your work is based on the CCD statistical method, which was obtained through 30 tests with 6 repetitions of the central point. It is crucial to emphasize the importance of the CCD method as the basis of your study. Please provide a detailed discussion on how the CCD method influenced the design and implementation of your research.

  c.  Data Size and Neural Network Usage: Given the small amount of data available, I have concerns regarding the justification for using neural networks and dividing the data into three sets (training, validation, and testing) alongside the inclusion of 6 repeated points. I kindly request a thorough explanation of how you addressed these limitations and why you believe the neural network approach remains appropriate in this context. Additionally, discuss any measures taken to mitigate the potential impact of limited data on the neural network training and validation process.

  d.  MLP Neural Network Architecture: The use of a multi-layer perceptron (MLP) neural network in your study raises questions, particularly in light of the General Approximation Theorem, which suggests that a single hidden layer is often sufficient. Please justify your choice of an MLP architecture, taking into account the specific requirements and characteristics of your research.

 e.  Explanation of Genetic Algorithm Usage: The reasoning behind your use of genetic algorithms requires clarification. The current explanation provided in the text does not adequately cover the purpose and benefits of incorporating genetic algorithms in your study. Please provide a comprehensive explanation of how genetic algorithms complement the neural network approach or contribute to the optimization process.

 f.   Relationship among the Methods: It is important to clearly specify and discuss the relationship among the CCD method, neural network, and genetic algorithm in your manuscript. Explain how these methods are integrated, how they interact or complement each other, and provide a cohesive overview of the overall workflow employed in your research.

6. Focus on CCD Statistical Method: It appears that your emphasis in the results and discussion section is primarily on the CCD statistical method. However, it was expected that you would utilize the neural network to generate the response surface diagrams. This discrepancy needs to be addressed to ensure consistency and accuracy in your methodology and results presentation.

 7.   Statistical Analysis in Table 1: The statistical method employed in Table 1 does not appear to be adequate, as you have not utilized a step-by-step method. It is important to use appropriate statistical techniques to support your findings. Additionally, the significance of x2 is not mentioned, and it is unclear whether it is statistically significant. Please provide a detailed explanation and statistical analysis for Table 1 to strengthen the validity of your results.

  8.  Quality and Thoroughness of Neural Network and Genetic Algorithm Results: Your results and discussion regarding the neural network and genetic algorithm approaches are described as having minimal quality and being overly general and superficial. It is imperative to thoroughly examine all aspects of your work and provide a comprehensive analysis of the neural network and genetic algorithm results. Please revise and expand upon these sections to address the concerns raised.

Author Response

Dear editor and reviewers,

Thank you for your careful work. We would like to express our great appreciation to you and reviewers for comments on our manuscript entitled “Ultrasound-assisted aqueous two-phase system extraction of anthocyanins from pomegranate pomaces by utilizing artificial neural network-genetic algorithm and response surface methodology models: Modeling, optimization, comparison, and characterization” (foods-2783106).

We have studied comments carefully and have made revision which marked in red in the manuscript. We have tried our best to revise our manuscript according to the comments. The response to you and reviewers are given as bellows.

Best regards,  

Dr. Ling Dong and Linyan Zhou

*Corresponding author:

Ling Dong, Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunming, Yunnan Province, 650500, China

E-mail address: [email protected]

Phone: +86-13810562442

Linyan Zhou, Faculty of Food Science and Engineering, Kunming University of Science and Technology, Yunnan Engineering Research Center for Fruit & Vegetable Products, Yunnan Key Laboratory for Food Advanced Manufacturing and International Green Food Processing Research and Development Center of Kunming City, Kunming, Yunnan Province, 650500, China.

E-mail address: [email protected]

Phone: +86-15011406984

Response to Reviewer 1

Reviewer #1: I have read and reviewed the article titled "Ultrasound-assisted aqueous two-phase system extraction of anthocyanins from pomegranate pomaces by utilizing artificial neural network-genetic algorithm and response surface methodology models: Modeling, optimization, comparison, and characterization". In order for this article to meet the required standards set by the food journal, I would like to suggest the following considerations:

1.Question (Q): Familiarize yourself with relevant literature: Given that the title of the article mentions keywords such as artificial neural network-genetic algorithm and response surface methodology models, it is crucial to thoroughly review existing literature in the field of food industry, particularly focusing on studies related to your own work. This will help establish a strong theoretical foundation and demonstrate your knowledge of the subject matter.

Answer (A): Thank you for your suggestion. We will try to read more relevant literatures during revision the manuscript and in future work.

2.Q: Explain the relevance of neural network and genetic algorithm: In the introduction, provide a clear explanation of why you have chosen to employ artificial neural network and genetic algorithm in your study. Discuss their significance in the context of the food industry and highlight their potential applications and benefits. This will help readers understand the motivation behind your research approach.

A: Thank you for your comment. We have added the relevant of artificial neural network and genetic algorithm in the introduction, and emphasized the advantages of application of ANN-GA used for the extraction of anthocyanins from pomegranate pomace in this study as “The integration of ANN and GA methods was aimed to leveraging their respective strengths to tackle the complex non-linear relationship modeling. The application of ANN bonding with GA (ANN-GA) for optimization of extraction process can help discover the most optimal conditions with less effort. Moreover, the GA added benefits of rectifying imperfections in optimized neural networks, circumventing the pitfalls of local extreme values, and achieving superior convergence and adaptability.” (line 116-122) and “Those studies showed that the profound integration of ANN and GA significantly enhanced the efficiency of identifying optimal extraction parameters during the extraction process, and generally yielded more accurate prediction results as compared to traditional response surface methodology.”(line 126-129), In addition, we also discussed the significance, benefits and potential applications of ANN-GA in the food industry as “Recently, some studies have reported that ANN-GA could be used to optimize the ex-traction parameters for target compounds from food raw materials or by-products, including ellagitannins from black raspberry seeds, punicalagin from pomegranates, and polyphenols from dragon fruit peel. Those studies showed that the pro-found integration of ANN and GA significantly enhanced the efficiency of identifying optimal extraction parameters during the extraction process, and generally yielded more accurate prediction results as compared to traditional RSM. The utilization of ANN-GA has emerged as a trending approach for modeling and optimization in the food processing industry.” in the introduction part of revised manuscript (line 122-130).

3.Q: State your research hypotheses: Towards the end of the introduction, clearly state your research hypotheses or objectives. Outline the specific aims of your study and how you expect the utilization of artificial neural network-genetic algorithm and response surface methodology models to contribute to answering your research questions. This will provide a roadmap for the readers and set the context for the subsequent sections of the article.

 A: Thank you for your comment. The research objective has been added in the introduction part as “As a result, it was important to build an environmentally-friendly and sustainable technique for extracting biologically active compounds. The objective of this study was to extract ACN from PP in a green and efficient technique.” (line 62-65) and“In this study, in order to optimize the extraction process of ACN from PP, we firstly built a UA-ATPS for the extraction of ACN from PP by demonstrating the phase diagrams of ethanol and ammonium sulfate ratios, then the optimization of CCD-RSM and ANN-GA for UA-ATPE were compared, including ACN yield, antioxidant activity, and monomeric ACN content.” (line 131-135)

4.Q: To enhance the reader's understanding and facilitate reproducibility of your work, it is recommended to provide an outline of the general process at the beginning of the Materials and Methods section. This outline will serve as a roadmap, giving readers a high-level overview of the steps involved in your research. Subsequently, you can delve into each step in detail within the Materials and Methods section.

A: Thank you for your suggestion. We now added an outline of the general process at the beginning of the Materials and Methods section as below.   

5.While I appreciate the alternative approach you have taken by incorporating neural network and genetic algorithm methods, I have identified several concerns that require your attention. I kindly request that you address and make the necessary corrections in the text of the article.

a.Q: Use of Neural Network: I found it difficult to understand the rationale behind your use of neural networks, especially considering that the CCD method is already a comprehensive statistical approach widely recognized as a reference. It would be helpful if you could provide a clearer explanation of why you chose to incorporate neural networks in your study, highlighting the specific advantages they offer in addition to the CCD method.

A: Thank you for your comment. As you said, CCD-RSM has been widely used in prediction and optimization research. However, the CCD-RSM has some limitations. for example, it is limited in its ability to explore non-linear, highly interactive or complex relationships and may not capture all influencing factors and patterns of change. While ANN is a non-linear computational modeling method that allows self-learning and training to solve practical problems without knowledge of specific mathematical models by simulating biological neural structures. Notably, a well-trained neural network can predict the output values under different input variables and obtain accurate results in a large amount of complex data. ANN models have shown a high degree of flexibility and capability in data fitting, prediction, and optimization. Nowadays, ANN has been applied in a wide range of fields, such as extracted ellagitannins from black raspberry seeds, punicalagin from pomegranates, and polyphenols from dragon fruit peel. Based on the aforementioned studies, we believed that Artificial Neural Network (ANN) is highly advisable for extracting anthocyanins from pomegranate pomace. In addition, the GA was a search method that has been shown to effective methodology for tack-ling numerous optimization problems, because it mimics the adaptation process of real biological systems. The application of ANN bonding with GA (ANN-GA) for optimization of extraction process can help discover the most optimal conditions with less effort, which also have the advantages of rectifying the imperfections of optimized neural networks, avoiding the trap of local extreme value, and obtaining good convergence and adaptability. The advantages of ANN and GA have been added as “The application of ANN bonding with GA (ANN-GA) for optimization of extraction process can help discover the most optimal conditions with less effort. Moreover, the GA added benefits of rectifying imperfections in optimized neural networks, circumventing the pitfalls of local extreme values, and achieving superior convergence and adaptability.” in the introduction part of the revised manuscript (line 118-122).

Reference：

Heri Septya Kusuma, Robby Ginanjar Margo Sudrajat, David Febrilliant Susanto, Selfina Gala, Mahfud Mahfud, Response surface methodology (RSM) modeling of microwave-assisted extraction of natural dye from Swietenia mahagony: A comparation between Box-Behnken and central composite design method. AIP Conf. Proc. 2015, 1699 (1), 050009.                            https://doi.org/10.1063/1.4938345

Aung T, Kim S J, Eun J B. A hybrid RSM-ANN-GA approach on optimisation of extraction conditions for bioactive component-rich laver (Porphyra dentata) extract. Food chemistry. 2022, 366, 130689. https://doi.org/10.1016/j.foodchem.2021.130689

Chen, B. Wang, J. Li, J. Xu, J. Zeng, W. Gao and K. Chen. Comparative study on the extraction efficiency, characterization, and bioactivities of Bletilla striata polysaccharides using response surface methodology (RSM) and genetic algorithm-artificial neural network (GA-ANN). International journal of biological macromolecules. 2023, 226, 982-995. https://doi.org/10.1016/j.ijbiomac.2022.12.017

b.Q: Basis of the Work: It appears that the foundation of your work is based on the CCD statistical method, which was obtained through 30 tests with 6 repetitions of the central point. It is crucial to emphasize the importance of the CCD method as the basis of your study. Please provide a detailed discussion on how the CCD method influenced the design and implementation of your research.

A: Thank you for your comment. The CCD-RSM method is a widely recognized statistical analysis technique, particularly in parameter optimization. Therefore, we incorporated this method into our research methodology from the outset and considered it an essential component of our study. Prior to ANN-GA experimentation, eight parameters as response values were used for optimization using the CCD-RSM model, including ACN yields, antioxidant activity (DPPH, ABTS, FRAP) and monomeric ACN content (Cyanidin-3-glucoside, Cyanidin-3,5-O-diglucoside, Pelargonidin-3-O-glucoside, Delphinidin-3-O-diglucoside). The optimization results showed that the response value of ACN yield had the highest R² value, which was further used for building ANN-GA model. The relationship between CCD and ANN had been added as “The target values of ACN yield for ANN optimization and prediction were deter-mined based on the results obtained from CCD-RSM screening.” in the revised manuscript (line 258-259).

c.Q: Data Size and Neural Network Usage: Given the small amount of data available, I have concerns regarding the justification for using neural networks and dividing the data into three sets (training, validation, and testing) alongside the inclusion of 6 repeated points. I kindly request a thorough explanation of how you addressed these limitations and why you believe the neural network approach remains appropriate in this context. Additionally, discuss any measures taken to mitigate the potential impact of limited data on the neural network training and validation process.

A: Thanks for the insightful review and concerns regarding the data size and the use of neural networks in our manuscript.

As the ANN excels in capturing complex non-linear relationships within data, even when the dataset size is small, we opted for a MLP network to model the relationship between the ACN yield and the process parameters. We added some discussion of any measures taken to mitigate the potential impact of limited data on the neural network training and validation process as “To address the challenge posed by the limited dataset, we implemented some strategies to mitigate its impact. For example, firstly, we applied data normalization techniques to standardize the distributions of our training samples, thereby minimizing errors in ANN. Secondly, we employed weight decay and dropout to prevent overfitting and ensure the model's generalizability on the test set. Thirdly, 3-fold cross-validation is employed to mitigate the limitation on dataset.” in line 264-269 of the revised manuscript.

d.Q: MLP Neural Network Architecture: The use of a multi-layer perceptron (MLP) neural network in your study raises questions, particularly in light of the General Approximation Theorem, which suggests that a single hidden layer is often sufficient. Please justify your choice of an MLP architecture, taking into account the specific requirements and characteristics of your research.

A: Thank you for your inquiry regarding the use of a multi-layer perceptron (MLP) neural network in our study.

Our work involves the intricate nonlinear relationships between the ACN yield and the process parameters, while MLP is adept at capturing and learning these complex relationships effectively. Throughout our experiments, we observed that a single hidden layer network encountered challenges in capturing these intricate feature representations compared to a configuration with two hidden layers, we attribute this discrepancy to the inherent complexity of the domain-specific data.

e.Q: Explanation of Genetic Algorithm Usage: The reasoning behind your use of genetic algorithms requires clarification. The current explanation provided in the text does not adequately cover the purpose and benefits of incorporating genetic algorithms in your study. Please provide a comprehensive explanation of how genetic algorithms complement the neural network approach or contribute to the optimization process.

A: Thank you for your comment. We have added the benefits of incorporating genetic algorithms in the introduction as “The integration of ANN and GA methods was aimed to leveraging their respective strengths to tackle the complex non-linear relationship modeling. The application of ANN bonding with GA (ANN-GA) for optimization of extraction process can help discover the most optimal conditions with less effort. Moreover, the GA added benefits of rectifying imperfections in optimized neural networks, circumventing the pitfalls of local extreme values, and achieving superior convergence and adaptability.” in line 116-122 of the revised manuscript. 

f.Q: Relationship among the Methods: It is important to clearly specify and discuss the relationship among the CCD method, neural network, and genetic algorithm in your manuscript. Explain how these methods are integrated, how they interact or complement each other, and provide a cohesive overview of the overall workflow employed in your research.

A: Thank you for your comment. The outline of methods had been added to clearly describe the process.

6.Q: Focus on CCD Statistical Method: It appears that your emphasis in the results and discussion section is primarily on the CCD statistical method. However, it was expected that you would utilize the neural network to generate the response surface diagrams. This discrepancy needs to be addressed to ensure consistency and accuracy in your methodology and results presentation.

A: Thank you for your suggestion. The aim of our study was to compare the prediction and optimization of anthocyanin yield from pomegranate pomace between the traditional CCD-RSM and the emerging ANN-GA. We referred to some publications, it was found that the result of ANN generally exhibited in Table form (Aung et al., 2022).

Reference:

Thinzar Aung, Seon-Jae Kim, Jong-Bang Eun. A hybrid RSM-ANN-GA approach on optimisation of extraction conditions for bioactive component-rich laver (Porphyra dentata) extract. Food Chemistry. 2022, 366, 130689. https://doi.org/10.1016/j.foodchem.2021.130689.

7.Q: Statistical Analysis in Table 1: The statistical method employed in Table 1 does not appear to be adequate, as you have not utilized a step-by-step method. It is important to use appropriate statistical techniques to support your findings. Additionally, the significance of x2 is not mentioned, and it is unclear whether it is statistically significant. Please provide a detailed explanation and statistical analysis for Table 1 to strengthen the validity of your results.

A: Thank you for your comment. Based on your comment, we checked the ANOVA analysis for RSM modeling, and the results of adjusted R², predicted R² and Adequate Precision were added in Table 1. Meanwhile, we also improved the analysis of Table 1 results as “It should be noted that a high R² value didn’t necessarily indicated the adequacy of the regression model, because R² value always tends to increase with the inclusion of additional variables, even if additional variables were not statistically significant. Therefore, it is more appropriate to consider the adjusted R² for a more accurate representation. In the CCD-RSM model of ACN yield, the predicted and adjusted R² were 0.8809 and 0.9562, respectively, and the difference between these two values was within an acceptable range of less than 0.2.” (line 375-381) and “In general, the Adequate Precision represented the ratio of signal to noise, and the ratio greater than 4 indicated that the model results were desirable. The ACN yield model has an Adequate Precision value of 22.9694, indicating sufficient signal and usability for navigating the design space.” (line 395-399) in results and discussion part 3.2 of the revised manuscript, in order to strength the discussion of response surface ANOVA results. Regarding to the significance of X2, we have also made a clearer and more explicit statement as “while the other factors (X₂, X₁X₂, X₁X₃, X₂X₃, X₂X₄) had no significant effect on ACN yield at the p>0.05.”in the revised manuscript (line 405-406).

8.Q: Quality and Thoroughness of Neural Network and Genetic Algorithm Results: Your results and discussion regarding the neural network and genetic algorithm approaches are described as having minimal quality and being overly general and superficial. It is imperative to thoroughly examine all aspects of your work and provide a comprehensive analysis of the neural network and genetic algorithm results. Please revise and expand upon these sections to address the concerns raised.

A: Thank you for your comment. We have now revised the results and discussion of neural network and genetic algorithm as suggested in the results and discussion part 3.4, 3.5, and 3.6, which marked in red in the revised manuscript (line 491-505, 516-517, 530-533, 559-571).

Author Response File: Author Response.docx

Reviewer 2 Report

I suggest accepting the manuscript after the minor revision. These are the following suggestions and corrections:

-          In this form, the title is too long and complex. I suggest to change it to be shorter and more simple.

-          In the introduction part the importance of anthocyanin extraction from pomegranate pomaces. It should be more highlighted why researching anthocyanin extraction from pomegranate pomaces is important and pose specific questions the study aims to answer.

-          It seems like the equations are not in the specific types for equation. This should be checked.

-          In the conclusion part connect conclusions to stated objectives and clearly highlight key results. Conclusions should more clearly reflect achieved research objectives, emphasizing the contribution of anthocyanin extraction from pomegranate pomaces in realizing ecological and economic benefits.

Minor editing of English language required

Author Response

Dear editor and reviewers,

Thank you for your careful work. We would like to express our great appreciation to you and reviewers for comments on our manuscript entitled “Ultrasound-assisted aqueous two-phase system extraction of anthocyanins from pomegranate pomaces by utilizing artificial neural network-genetic algorithm and response surface methodology models: Modeling, optimization, comparison, and characterization” (foods-2783106).

 We have studied comments carefully and have made revision which marked in red in the manuscript. We have tried our best to revise our manuscript according to the comments. The response to you and reviewers are given as bellows.

Best regards,  

Dr. Ling Dong and Linyan Zhou

*Corresponding author:

Ling Dong, Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunming, Yunnan Province, 650500, China

E-mail address: [email protected]

Phone: +86-13810562442

Linyan Zhou, Faculty of Food Science and Engineering, Kunming University of Science and Technology, Yunnan Engineering Research Center for Fruit & Vegetable Products, Yunnan Key Laboratory for Food Advanced Manufacturing and International Green Food Processing Research and Development Center of Kunming City, Kunming, Yunnan Province, 650500, China.

E-mail address: [email protected]

Phone: +86-15011406984

Response to Reviewer 2

Reviewer #2: I suggest accepting the manuscript after the minor revision. These are the following suggestions and corrections.

1.Question (Q): In this form, the title is too long and complex. I suggest to change it to be shorter and more simple.

Answer (A): Thank you for your suggestion. We have been revised the title to “Comparison on Ultrasound-Assisted Aqueous Two-Phase System Extraction of Anthocyanins from Pomegranate Pomaces by Utilizing Artificial Neural Network-Genetic Algorithm and Response Surface Methodology Models”.

2.Q: In the introduction part the importance of anthocyanin extraction from pomegranate pomaces. It should be more highlighted why researching anthocyanin extraction from pomegranate pomaces is important and pose specific questions the study aims to answer.

A: Thank you for your suggestion. As you suggested, we have highlighted the significance of the study on the extraction of anthocyanin from pomegranate pomaces in the introduction part, specifying the study objectives as “However, PP after juicing processing was normally discarded, which bring both the environmental pollution and economic loss. With the development of the pomegranate industry, its by-products dramatically increase and the problem become increasingly prominent. However, little research has been focused on the utilization of the by-products of pomegranate, especially by using green and sustainable methods. As a result, it was important to build an environmentally-friendly and sustainable technique for extracting biologically active compounds. The objective of this study was to extract ACN from PP in a green and efficient technique.” in line 58-65 of the revised manuscript.

3.Q: It seems like the equations are not in the specific types for equation. This should be checked.

A: Thank you for your comment. We have checked and modified all the equations as you suggested.

4.Q: In the conclusion part connect conclusions to stated objectives and clearly highlight key results. Conclusions should more clearly reflect achieved research objectives, emphasizing the contribution of anthocyanin extraction from pomegranate pomaces in realizing ecological and economic benefits.

A: Thank you for your suggestion. Based on your suggestions, we have made some revision in the conclusion part as “Thus, the study succeeded in building an efficient and green way to extract ACN from PP by optimizing and forecasting the extraction process by the ANN-GA model based on the UA-ATPE method, which provided a novel idea and methodology for the high-value utilization of abundant PP resources produced with the development of pomegranate industry. The study also offered foundation for the extraction of bioactive substances from by-products of fruit and vegetable processing, which has positive significance for exploring the potential value of more by-products and obtaining good ecological and economic benefits in the future.” in line 606-613 of the revised manuscript.

Author Response File: Author Response.docx