Analysis of Nutritional Content in Rice Seeds Based on Near-Infrared Spectroscopy

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper proposes a rapid analysis method for the nutritional components of single rice seeds based on near-infrared spectroscopy. This research is of great significance for rice breeding. However, in terms of modeling, the focus was on preprocessing and modeling methods, while wavelength selection was not covered. The specific comments and suggestions are as follows:

1. In Introduction, need to cite most recent references for wide reader's interest. Especially, it needs to be added contents and references about spectral wavelength selection.

2. Line 48，the NIR range is 780-2526 nm.
3. Regarding the chemical composition analysis of seeds, how are the main components of a single seed detected using traditional chemical methods?
4. It is recommended to specify the spectral range of the light source and the detector.
5. As shown in Figure 1, in the transmission spectroscopy acquisition mode, when the spectral data is collected from a single seed as a sample, is it the rice without husks or the rice grains with husks? How is the collimation of the transmission spectrum guaranteed during the spectral acquisition process?
6. It is suggested to add the relative RMSECV indicator.
7. What is the purpose of introducing the classification model evaluation indicators ROC and AUC into regression problems?
8. Line 393-394, it is suggested that the figures be numbered with Arabic numerals4, 5 and 6.
9. It is suggested that RMSECV be introduced in table1 and table2 as the basis for choosing the best preprocessing method, so as to ensure the applicability of the method in this paper to unknown sample data.
10. It is suggested that the wavelength selection method be introduced in this paper to further improve the modeling accuracy.

Author Response

Thank you for pointing out the shortcomings of the article. We have made targeted modifications and will respond to each of your suggestions one by one.

Comments 1:In Introduction, need to cite most recent references for wide reader's interest. Especially, it needs to be added contents and references about spectral wavelength selection.

Response 1：Thank you very much for pointing out the shortcomings in method selection and literature research in the article. Your suggestions have proven to be very constructive. After our research, we found that the spectral wavelength selection method can indeed optimize the predictive performance of the model and improve computational efficiency. We attempted to introduce the spectral wavelength selection method of SPA before establishing the model in the article, and have now updated the latest conclusions in the abstract, introduction, results, conclusion, references, and other sections of the article.

Comments 2:Line 48，the NIR range is 780-2526 nm.

Response 2：Thank you for the accuracy of the data in the near-infrared spectral band, which we have updated on line 48 of the article.

Comments 3：Regarding the chemical composition analysis of seeds, how are the main components of a single seed detected using traditional chemical methods?

Response 3：Thank you for your comment. In traditional chemical testing of rice seeds, the principle of realizing batch testing and single seed testing is the same, both are done by grinding and sampling a large number of rice seeds or single seeds for further testing of their nutrient content. In traditional chemical testing, the difference between batch testing and single seed testing is only in the number of samples taken, and single seed testing is time-consuming and laborious compared to batch testing. Section 2.1 of this paper describes chemical tests for protein, fat and starch. These methods are time-consuming, complex, expensive, polluting, and do not retain the rice seeds being tested.

Comments 4：It is recommended to specify the spectral range of the light source and the detector.

Response 4：Thank you for your response to the lack of clarity on the spectral range in our article, which we have corrected in lines 134-137 of the article.

Comments 5：As shown in Figure 1, in the transmission spectroscopy acquisition mode, when the spectral data is collected from a single seed as a sample, is it the rice without husks or the rice grains with husks? How is the collimation of the transmission spectrum guaranteed during the spectral acquisition process?

Response 5：Thank you for pointing out that the article did not fully describe the design of our instrument, and it is unclear whether floating shells were used for rice seeds. We chose rice seeds with shells and added a description in lines 150-158 of the article that transmission spectroscopy is more suitable for testing rice seeds with shells. We also added detailed information on how to design instruments to achieve light collimation as described in lines 163 to 170 of the article. We also added a schematic diagram of the collimation part of the mechanical model to the detection system diagram in Figure 1. Specifically, we use a collimating lens to adjust the straightness of the light beam, and use optical fibers to collect the transmitted light and transmit it to the spectrometer to complete near-infrared spectrum acquisition. When designing the instrument, the input and output ends of two optical fibers are placed on a vertical line to ensure that the light passes through the seed vertical line.

Comments 6：It is suggested to add the relative RMSECV indicator.

Response 6：Thank you for suggesting the RMSECV evaluation metric, it was an oversight on our part not to include this evaluation metric in the article. The RMSECV evaluation metrics have now been added to Tables 1, 2, and 3 as model evaluation metrics.

Comments 7：What is the purpose of introducing the classification model evaluation indicators ROC and AUC into regression problems?

Response 7：We are especially grateful for the questions you raised about our ROC curve. In our initial thinking, we mistakenly thought that the model had some classification problems, and wanted to introduce the ROC curve and AUC value for model evaluation in order to make the model evaluation more comprehensive and precise. Based on further research, we found that these two indicators did not contribute much to the evaluation of the model we built. Therefore, we canceled this evaluation index and introduced the RMSECV evaluation index to evaluate the model. Once again, we would like to express our sincere thanks for your comments!

Comments 8：Line 393-394, it is suggested that the figures be numbered with Arabic numerals4, 5 and 6.

Response 8：The figures have been changed in accordance with your corrections.

Comments 9：It is suggested that RMSECV be introduced in table1 and table2 as the basis for choosing the best preprocessing method, so as to ensure the applicability of the method in this paper to unknown sample data.

Response 9：Thank you for suggesting the RMSECV evaluation metric, it was an oversight on our part not to include this evaluation metric in the article. The RMSECV evaluation metrics have now been added to Tables 1, 2, and 3 as model evaluation metrics.

Comments 10：It is suggested that the wavelength selection method be introduced in this paper to further improve the modeling accuracy.

Response 10：Once again, thanks to your constructive suggestions, we have introduced the SPA spectral feature wavelength selection method

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

The research is very interesting; however, there are a few points that, in my view, should be further investigated.

Conducting a more thorough bibliographic search, thereby increasing the references and considering other instruments.

Insert a space between the tables and the text

Line 47 - The authors should better specify this point:  this technology based on transmittance or reflectance. Please explain the differences

Line 212 - Observations also on the Artificial Neural Network (ANN)

Line 333 - Has an evaluation using ANN been considered?

 

 

Comments for author File: Comments.pdf

Author Response

Comments 1:Insert a space between the tables and the text

Response 1:Thank you for pointing out the formatting issue in the article. We have made modifications to the table section in the text.

Comments 2:Line 47 -The authors should better specify this point:  this technology based on transmittance or reflectance. Please explain the differences

Response 2:Thank you for pointing out the insufficient description of the transmission method in the article. We have supplemented it in lines 150-158 of the text.

Comments 3:Line 212 - Observations also on the Artificial Neural Network (ANN)

Response 3:Thank you for providing the ANN modeling solution. We have not conducted any research on this modeling method before. After our research, we believe that artificial neural network models are also a good solution for predicting the nutritional composition of rice seeds. We have introduced the ANN algorithm in the introduction, results, references, and other sections of the article. And Section 2.4.3 of this article provides a supplementary introduction to the ANN model.

Comments 4:Line 333 - Has an evaluation using ANN been considered?

Response 4:Thank you again for providing the ANN model solution. After research, we have adopted the artificial neural network model you suggested to establish a prediction model for the dataset. After several days of work, the predictive performance of the model has been updated to Tables 1, 2, and 3 in this article. In addition, we also introduced wavelength selection methods for further optimization. It has been proven that ANN models indeed have impressive predictive performance. As this is our first time using ANN models for modeling, we may not have achieved the best results. I will delve deeper into my studies in the future. Thank you for your suggestion.

 

 

 

 

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

I still have doubts about how a single seed can complete traditional chemical detection methods. What kind of high-end equipment can determine the protein, fat and starch content of a single rice seed? Regarding the determination of fat and starch content, when our team uses traditional chemical methods, the detection error is still relatively large when using a large number of samples for determination. The author did not provide a convincing answer on how to achieve precise detection of the chemical indicators of individual seeds. Precise chemical detection is the cornerstone of constructing a high-precision spectral detection model, which makes me skeptical about the research results of the paper.

Author Response

Comments：I still have doubts about how a single seed can complete traditional chemical detection methods. What kind of high-end equipment can determine the protein, fat and starch content of a single rice seed? Regarding the determination of fat and starch content, when our team uses traditional chemical methods, the detection error is still relatively large when using a large number of samples for determination. The author did not provide a convincing answer on how to achieve precise detection of the chemical indicators of individual seeds. Precise chemical detection is the cornerstone of constructing a high-precision spectral detection model, which makes me skeptical about the research results of the paper.

Respond：

Dear reviewer:

I'm sorry for the confusion caused by my previous response. Due to our laboratory's focus on optical and mechanical testing, we do not have the ability to perform chemical testing.The chemical detection and calibration methods were entrusted to the Molecular Biology Laboratory of Hebei Medical University. After further communication, we found that our answer to your last question was incorrect. Due to communication issues with the testing agency during the consultation, we mistakenly believed that the accuracy of a single particle in traditional testing methods could achieve the effect of large-scale testing. We apologize for our misunderstanding of your question.

The single particle detection mentioned in the article refers to the multiple spectral data detection of a single particle, rather than the single particle chemical detection of seeds. We provide a large number of labeled rice seeds measured by single particle spectral data for testing, which is divided into two parts: one part is used for reagent kit detection, and the other part is used for large-scale detection and calibration by traditional chemical methods. In theory, single particle detection has already covered the detection limit of the reagent kit. To ensure that the accuracy of single particle detection is not lost as much as possible, we use 3 rice seeds to grind for multiple detections and take the average value. After removing outliers, we retain the data with an error of no more than 8% as the dataset. The method used is as described in our article. The BCA protein concentration assay kit used for protein detection is characterized by high sensitivity, with a detection concentration lower limit of 25ug/L (with a good linear relationship in the concentration range of 20-1000 μ g/mL) and a minimum protein detection amount of 0.2ug. For fat detection, the triglyceride (TG) test kit is used, with a linear range of 0.3-11.4 mmol, r2>0.995。 Starch is detected using Solarbio's starch assay kit. The grinder adopts a high-speed low-temperature tissue grinder SCIENTZ-48L, the centrifuge adopts the Centrifuge 5810 R model, the handheld homogenizer, and the enzyme-linked immunosorbent assay (ELISA) uses a multi-mode microplate detector. And we have repeatedly confirmed with the chemical testing agency that the chemical data is true and valid.

We apologize again for our negligence and wish you a smooth work and happy life.

 

Author Response File: Author Response.docx