Development of Volatile Fatty Acid and Methane Production Prediction Model Using Ruminant Nutrition Comparison of Algorithms

Round 1

Reviewer 1 Report (Previous Reviewer 2)

Comments and Suggestions for Authors

Dear Authors

I have read the revised manuscript and appreciate the authors' consideration of my previous suggestions. Authors have covered all of my observations in an acceptable manner, in such a way that I have no further observations.

Author Response

Dear Reviewer,

Thank you for reviewing the revised manuscript. I appreciate your consideration of my previous suggestions. You have addressed all of my observations in an acceptable manner, and I have no further comments or suggestions.

Once again, thank you for your efforts and thorough review.

Best regards,

Myungsun Park.

Reviewer 2 Report (New Reviewer)

Comments and Suggestions for Authors

Predictive models were developed using multiple linear regression (MLR), artificial neural networks (ANN) and KNN algorithms

The models were based on data extracted from 30 research articles and 16 in vitro rumen fermentation tests.

The data was then pre-processed, which included data normalisation and other steps to prepare the data for analysis.

Different modelling algorithms were selected to develop the prediction models. This included models based on multiple linear regression (MLR), artificial neural networks (ANN) and the KNN algorithm.

Each algorithm was chosen based on its characteristics and ability to deal with different types of data and problem complexities.

After developing the models, they were evaluated for their accuracy and performance using metrics such as root mean square error (RMSE), coefficient of determination (R²).

The models were validated to ensure that they could generalise well to new data sets and situations not observed during training.

In my opinion, this work has several limitations:

The algorithms were developed on the basis of data from a variety of studies and experiments. In my opinion, they are not representative enough and do not cover the wide range of conditions, so the models may not be able to fully capture the complexity of the rumen system.

Although the algorithms chosen (MLR, ANN, KNN) are widely used and have their own advantages, each also has its limitations. For example, multiple linear regression may not capture non-linear relationships, neural networks may be susceptible to overfitting or underfitting

The models developed may have difficulty generalising to new data sets, they may not be able to make accurate predictions in new contexts

Although the models have been evaluated for their accuracy and performance, it is important to carry out additional validations in real-world conditions to confirm that they can be successfully applied in different environments and situations.

It is therefore crucial to be aware of these weaknesses when interpreting and applying the results of the models developed and this point should be detailed in the discussion which should also have a clear explanation of the practical limitations of the results.

Author Response

Dear Reviewer,

Thank you for your efforts and for taking the time to review our work. We appreciate the time and effort you have dedicated to providing your valuable suggestions.

We acknowledge that the data used in developing our predictive model is sourced from various studies and experiments, and it may not fully encompass the Lumen system. Lumen operates under anaerobic conditions where the external environment plays a crucial role, and our model may not cover all these extensive practical conditions. Consequently, we aimed to predict Johnson production using a subset of the available data as a starting point.

Regarding the use of Multiple Linear Regression (MLR), Artificial Neural Networks (ANN), and K-Nearest Neighbors (KNN), we recognize that each method has its own strengths. Evaluating which approach best fits our specific conditions was essential to determine the most accurate predictive model. The issue is that MLR might not capture certain nonlinear relationships, while ANN could suffer from overfitting or underfitting. These considerations were part of our model evaluation process.

We also agree on the importance of fitting new datasets. Although we used metrics like RMSE and R² to assess model fit and performance, further validation is necessary to ensure applicability across diverse environments and situations.

We will integrate your suggestions and emphasize the practical implications of our results in the discussion section. This will provide a clearer understanding of the performance and applicability of the predictive model. Additionally, we have added a section in the conclusion to highlight each consideration and potential weakness, ensuring users take these into account when interpreting the results.

Thank you for your patience. Expanding on the areas you have highlighted will greatly enhance the quality and applicability of our research.

Sincerely,
Dr. Park

Reviewer 3 Report (New Reviewer)

Comments and Suggestions for Authors

The authors used three different multivariate models to predict methane production and VFA concentrations in the ruminal fermentation from dietary inputs (DMI, NDFI, etc.). Such an approach has utility for producers in that these dietary inputs are easily measured, and can yield fermentation product predictions without consideration of the complex biochemistry of the fermentation itself.

            The reviewer confesses to a limited understanding of the KNN and ANN approaches, and it is likely that many readers may likewise be interested in the subject of the analysis (ruminal fermentation), and have similar  limited familiarity with the model methodology. Consequently, the authors should provide an explanatory paragraph or two on the underlying principles of the models. They should also provide a better explanation as to how the KNN model could yield essentially perfect fit of the data. Has the clear superiority of the KNN model been observed in other systems in which such comparisons have been made?

Specific comments:

L41-42: Indicate whether these percentages are on a weight basis or a molar basis.

L44-45: More importantly, it also is the major contributor to gluconeogenesis.

L49-51: This statement is a little misleading, as there is virtually no aceticlastiic methanogenesis in the rumen. The relatively small proportion of methane not produced by hydrogenotrophic methanogenesis more likely comes from conversion of methyl groups rather than acetate.

L110: Describe the filtration (cheesecloth? type of filter?).

L119-121: Describe the types of syringes, needles, and gas packs used. Also, the method seems imprecise. With what precision could the gas values be read on the syringes?

Equations 1 and 3: What does x_min’ represent?

L190-195: It’s one thing for a model to “overfit” the data, it’s quite another thing to obtain a perfect relationship (k=1). What could account for such a radical extent of overfitting?

Figures 1, 2, 3 and 5: Typically, MLR analysis also yields a P-value, useful in assessing the probability that the relationship is statistically significant. What was the P-value in this MLR analysis? Are P-values also obtainable from the KNN and ANN analyses?

L210: Why are the units “%” for predicted nutrient intake? Percent of what?

L216-218: From Table 3, it appears that the regression coefficient of 1.57 in MC4 is for DMI, not NDFI.

K235-236: Should the units for TVFA be in mmol/L, as opposed to mmol/100 mmol?

L237: Why are the units for DMI and CPI d/kg, rather than kg/d?

L278-280: This sentence needs rewriting, as its intent is unclear.  In its current form, it essentially states that “Methanogenesis…...requires a methane prediction model”, which is certainly not true.

L307-309: This appears to be generally true, but why does the KM1 equation show a strong negative coefficient (-0.94) for methane production for MC2 (L321)?

L339-342 and L347-349: The authors are being a bit vague here. Are they implying that methane is produced from acetate and butyrate? This would go against the generally accepted notion that VFA are not converted to methane in the rumen. Or are they instead implying that electrons released during fermentative acetate production are preferentially directed toward methane production, and that butyrate production is also enhanced because acetate is the preferred substrate for butyrate production via acetate chain elongation? This latter scenario is more likely.

All figures, particularly Figure 4, would benefit from larger font sizes for axis labels.

Minor edits:

L2: Here and throughout the title, why is “Methane” capitalized? Unless leading off a sentence, it should be lower-case.

L16: Delete “the nutritional composition of their”.

L92: Correct misspelling (“combinig”).

L216: Correct misspelling (“propnally”).

Comments on the Quality of English Language

English is adequate, a few minor edits suggested.

Author Response

General Response:

Model Explanation:

Reviewer: The authors should provide an explanatory paragraph or two on the underlying principles of the models (KNN, ANN), and better explain how the KNN model could yield a perfect fit of the data.

Response: Thank you for your insightful comments. We agree that providing clear explanations of the underlying principles of the KNN and ANN models will be beneficial to the readers. We will include a dedicated section explaining the basic concepts and mechanisms of KNN and ANN, emphasizing how KNN achieves high accuracy through its non-parametric approach and discussing the potential reasons for its perfect fit in this context. We will also reference studies where the KNN model has shown similar superiority.

Specific Comments:

L41-42: Reviewer: Indicate whether these percentages are on a weight basis or a molar basis.

Response: Thank you for pointing this out. We will specify that these percentages are on a weight basis in the revised manuscript.

L44-45: Reviewer: More importantly, it also is the major contributor to gluconeogenesis.

Response: Thank you for the suggestion. We will revise the sentence to emphasize that propionate is the major contributor to gluconeogenesis.

L49-51: Reviewer: This statement is a little misleading, as there is virtually no aceticlastic methanogenesis in the rumen.

Response: Thank you for the correction. We will revise the statement to reflect that methane in the rumen is primarily produced by hydrogenotrophic methanogens, and that the relatively small proportion of methane not produced by hydrogenotrophic methanogenesis is more likely due to the conversion of methyl groups rather than acetate.

L110: Reviewer: Describe the filtration process (cheesecloth? type of filter?). Response: We will provide a detailed description of the filtration process, specifying that cheesecloth was used for filtration.

L119-121: Reviewer: Describe the types of syringes, needles, and gas packs used. Also, the method seems imprecise. With what precision could the gas values be read on the syringes?

Response: We have included detailed descriptions of the syringes, needles, and gas packs used in our study in the revised manuscript. Specifically, we used 100 milliliter (ml) calibrated glass syringes attached to a 20-gauge 30.5 cm needle (Popper®, Fisher Scientific) and aluminum gas packs with rubber inserts for gas collection. The gas values could be read on the syringes with a precision of 1 ml.

By utilizing syringes that are commonly employed for measuring total gas production in in vitro experiments, we aimed to ensure the accuracy and reliability of our measurements. This approach allows for a reasonable degree of accuracy in measuring the degree of fermentation.

Equations 1 and 3: Reviewer: What does xmin' represent?

Response: Thank you for pointing out the error. The correct term should be xmin, representing the minimum value in the dataset. We will correct this in the manuscript.

L190-195: Reviewer: It’s one thing for a model to “overfit” the data, it’s quite another thing to obtain a perfect relationship (k=1). What could account for such a radical extent of overfitting?

Response: Thank you for highlighting this critical point. We acknowledge the concern regarding the KNN model showing a perfect relationship (k=1), which is an indicator of extreme overfitting. There are several potential reasons for this:

1. Limited Dataset Size: The dataset used in our study might be relatively small, causing the KNN model to fit the data points too closely. In smaller datasets, KNN can exhibit high sensitivity to individual data points, leading to a perfect fit.
2. Data Homogeneity: The dataset might lack sufficient variability, making it easier for the KNN model to predict values with high accuracy. When data points are very similar to each other, the KNN algorithm can easily find the nearest neighbors with identical or very close values.
3. Parameter Selection: The choice of k (the number of neighbors) is crucial in KNN. A very low value of k can lead to overfitting, as the model will rely heavily on the closest data points without considering the overall distribution. We will review and potentially adjust the k parameter to mitigate this issue.
4. Feature Engineering: The features used in the model may contribute to overfitting if they capture too much specific information about the training data. We will revisit our feature selection process to ensure that the features used are generalizable and not overly tailored to the training dataset.
5. Model Validation: We will employ more robust cross-validation techniques to better assess the model's performance and prevent overfitting. By using k-fold cross-validation, we can ensure that the model is tested on multiple subsets of the data, providing a more accurate measure of its generalizability.

To address these issues, we have:

Increased the dataset size where possible, to provide the model with more diverse data points.

Re-evaluated the choice of k in the KNN model and experimented with different values to find a more balanced approach.

Implemented cross-validation to validate the model's performance more effectively.

Conducted a thorough review of the features used to ensure they are appropriate and not leading to overfitting.

We will continue to conduct experiments to further expand the dataset and refine our approach. We hope these steps will address the overfitting concern and improve the robustness and generalizability of our model.

Figures 1, 2, 3, 5: Reviewer: Typically, MLR analysis also yields a P-value, useful in assessing the probability that the relationship is statistically significant. What was the P-value in this MLR analysis? Are P-values also obtainable from the KNN and ANN analyses?

Response: Thank you for your question. In traditional multiple linear regression (MLR) analysis, P-values are indeed used to assess the statistical significance of the relationships between the predictors and the outcome variable. In our MLR analysis, the P-values were calculated and reported to indicate the significance of the predictors.

However, for non-parametric models like K-nearest neighbors (KNN) and artificial neural networks (ANN), the concept of P-values does not directly apply. These models do not make assumptions about the distribution of the predictors and outcome variables, and they do not provide P-values for significance testing. Instead, the performance of KNN and ANN models is typically assessed using metrics such as accuracy, precision, recall, F1 score, and cross-validation results.

While P-values are a useful measure in parametric models like MLR, their utility in non-parametric models like KNN and ANN is limited. Instead, we focus on evaluating the model's overall performance and predictive accuracy through cross-validation and other validation techniques.

To maintain consistency in the evaluation of all three models (MLR, KNN, and ANN), we did not report P-values. Instead, we relied on performance metrics that are applicable across all models to provide a fair comparison.

We hope this clarifies the distinction between the statistical significance testing in MLR and the performance evaluation in KNN and ANN models. If you have any further questions or require additional details, please let us know.

L210: Reviewer: Why are the units “%” for predicted nutrient intake? Percent of what?

Response: Thank you for your question. The units “%” for predicted nutrient intake refer to the percentage of the actual nutrient intake relative to the predicted values. Specifically, it indicates the deviation or error percentage between the predicted nutrient intake values generated by the models and the actual observed intake values.

For example, if a model predicts a nutrient intake value, the percentage indicates how close or far this prediction is from the actual intake, expressed as a percentage of the actual intake value. This helps in understanding the accuracy and reliability of the predictions made by different algorithms.

We hope this clarifies the use of percentage units in the context of predicted nutrient intake values. If you have any further questions or require additional details, please let us know.

L216-218: Reviewer: From Table 3, it appears that the regression coefficient of 1.57 in MC4 is for DMI, not NDFI.

Response: Thank you for the correction.

L235-236: Reviewer: Should the units for TVFA be in mmol/L, as opposed to mmol/100 mmol?

Response: We will correct the units for TVFA to mmol/L.

L237: Reviewer: Why are the units for DMI and CPI d/kg, rather than kg/d?

Response: We will correct the units for DMI and CPI to kg/d.

L278-280: Reviewer: This sentence needs rewriting, as its intent is unclear.  In its current form, it essentially states that “Methanogenesis…...requires a methane prediction model”, which is certainly not true.

Response: Thank you for pointing out the ambiguity in the sentence. We appreciate your feedback and will revise the sentence to clarify its intent. The goal was to highlight the relationship between methanogenesis and the production of ruminal volatile fatty acids, and to explain the need for developing a model to predict methane production under varying rumen conditions.

The revised sentence aims to convey that understanding the dynamics of methanogenesis, influenced by volatile fatty acids and the rumen environment, motivated the creation of a model to predict methane production. This model helps in managing and mitigating methane emissions in ruminant systems.

We hope this revision addresses the concern and clarifies the intent of the original statement. If you have any further questions or require additional adjustments, please let us know.

L307-309: Reviewer: This appears to be generally true, but why does the KM1 equation show a strong negative coefficient (-0.94) for methane production for MC2 (L321)?

Response: Thank you for your insightful question regarding the strong negative coefficient observed in the KM1 equation for methane production in MC2. We appreciate the opportunity to clarify this point.

The strong negative coefficient for methane production in MC2 can be attributed to the dietary intake patterns observed in fattening cattle, which typically consume more concentrate feeds as opposed to forage. As you are aware, the rumen environment and the substrates available for methanogens can vary significantly based on diet. Concentrate feeds tend to produce different fermentation end products compared to forage, influencing the methanogenesis process.

In our study, we accounted for these variations by clustering the rumen environments before making predictions. By doing so, we aimed to tailor the predictions based on the specific characteristics of each cluster. This approach acknowledges that the fuel sources utilized by methanogens can change depending on the rumen conditions, thereby affecting methane production differently across clusters.

L339-342 and L347-349: Reviewer: Are they implying that methane is produced from acetate and butyrate? This would go against the generally accepted notion that VFA are not converted to methane in the rumen.

Response: Thank you for your question. We would like to clarify that methane in ruminants is produced by methanogens, which utilize substrates influenced by volatile fatty acids (VFAs) such as acetate and butyrate. Although VFAs themselves are not directly converted to methane, their presence and concentration in the rumen significantly impact the substrates available for methanogens, thereby influencing methane production. We hope this clarifies the role of VFAs in influencing methane production in the rumen. If you have any further questions or require additional details, please let us know.

All figures: Reviewer: Axis labels would benefit from larger font sizes.

Response: We will increase the font sizes of the axis labels in all figures to improve readability.

Minor Edits:

L2: Reviewer: Why is “Methane” capitalized? Unless leading off a sentence, it should be lower-case.

Response: We will change "Methane" to lowercase throughout the title and text (except at the beginning of sentences).

L16: Reviewer: Delete “the nutritional composition of their”.

Response: We will delete this phrase for clarity.

L92: Reviewer: Correct misspelling (“combinig”).

Response: We will correct the misspelling to “combining”.

L216: Reviewer: Correct misspelling (“propnally”).

Response: We will correct the misspelling to “proportionally”.

Reviewer 4 Report (New Reviewer)

Comments and Suggestions for Authors

Dear authors, 

Please find below some minor suggestions for improving your manuscript. Thanks.

Abstract: I would suggest to mention which one of the three methods performed better in terms of statistical fitting

Line 34: keywords in alphabetical order.

Line 40: The VFAs

Line 55 and onwards: Why Methane with capital M?

Line 64: R² instead of R2 (superscript)

Line 87: 31 papers according to Table 1. Please change this in the abstract section (30 articles)

Line 102 – Table 2: Please include % of concentrate in the diet within the diet composition section

Line 124: The VFA

Line 156: When referring to the k value, please use k (in italics). Revise this throughout the text.

Line 258: R² instead of R2 (superscript)

Line 283 – 284: Please include the ‘elbow method’ in the material and methods section for determining the number of clusters.

Author Response

Dear Reviewer,

Thank you very much for your valuable feedback and suggestions for improving our manuscript. We appreciate the time and effort you have taken to review our work and provide us with these constructive comments. Below are our detailed responses to your suggestions:

Abstract:

Suggestion: Mention which one of the three methods performed better in terms of statistical fitting.
Response: We agree with your suggestion and have revised the abstract to include a statement about which method showed superior statistical fitting.

Line 34:

Suggestion: Keywords in alphabetical order.
Response: We have reordered the keywords alphabetically as recommended.

Line 40:

Suggestion: The VFAs
Response: We have corrected this to "The VFAs."

Line 55 and onwards:

Suggestion: Why Methane with capital M?
Response: This was an oversight, and we have corrected all instances of "Methane" to "methane" in lowercase.

Line 64:

Suggestion: R2 instead of R2 (superscript).
Response: We have corrected "R2" to "R²" using superscript formatting.

Line 87:

Suggestion: 31 papers according to Table 1. Please change this in the abstract section (30 articles).
Response: We have updated the abstract to reflect the correct number of papers, changing "30 articles" to "31 articles."

Line 102 – Table 2:

Suggestion: Please include % of concentrate in the diet within the diet composition section.
Response: Thank you for your valuable feedback regarding Table 2. We appreciate your suggestion to include the percentage of concentrate in the diet within the diet composition section.

In response, we would like to clarify that due to the feeding methods used in our study, such as Total Mixed Ration (TMR) or the separate feeding of forage and concentrate, it is more appropriate to view the nutrient intake as a whole rather than focusing solely on the percentage of concentrate. This approach allows for a more accurate reflection of the nutritional intake and its effects on the study outcomes.

We hope this explanation addresses your concern. Thank you once again for your insightful suggestions.

Line 124:

Suggestion: The VFA
Response: We have corrected this to "The VFA."

Line 156:

Suggestion: When referring to the k value, please use k (in italics). Revise this throughout the text.
Response: We have revised the text to use "k" in italics consistently throughout the manuscript.

Line 258:

Suggestion: R2 instead of R2 (superscript).
Response: We have corrected "R2" to "R²" using superscript formatting.

Line 283 – 284:

Suggestion: Please include the ‘elbow method’ in the material and methods section for determining the number of clusters.
Response: We have added a description of the ‘elbow method’ in the materials and methods section for determining the number of clusters.

Thank you once again for your insightful comments. We believe these changes have significantly improved the clarity and quality of our manuscript. We hope that our revisions meet with your approval.

Best regards,

Myungsun Park.

Round 2

Reviewer 2 Report (New Reviewer)

Comments and Suggestions for Authors

The authors responded to the main criticisms of the work:

 They discussed the risk of overfitting in KNN models due to the limited amount of data, (Line 241-line 244.

Emphasised in the conclusion (Line 353-line 355) the need for external validation with field data under different conditions to confirm the conditions to confirm the practical applicability of predictive models. 

I still think it's a limited study because their database consists of 31 articles (I don't even know if these cover all the necessary conditions) but they say so in the paper.

Author Response

Dear Reviewer,

Thank you for your valuable feedback on our manuscript. We appreciate your acknowledgment that we have addressed the primary concerns regarding the risk of overfitting in KNN models (Lines 241-244) and emphasized the need for external validation with field data in our conclusion (Lines 353-355).

We understand your concern regarding the limited scope of our database, which consists of 31 articles. While this sample size may appear limited, it is important to note that the articles selected encompass a range of conditions and contexts pertinent to our study. The diversity within these studies allows us to draw preliminary conclusions and develop predictive models that, while not exhaustive, provide a meaningful contribution to the field.

We acknowledge that our current dataset may not cover all possible conditions, but it represents a comprehensive subset that provides valuable insights into volatile fatty acid and methane production prediction. We have clearly stated the limitations of our study in the manuscript and emphasized the necessity for future research to expand on this work by including more diverse and extensive datasets.

We hope this addresses your concerns and demonstrates our commitment to advancing this field of research responsibly.

Thank you once again for your insightful comments.

Sincerely,
Dr. Park.

Reviewer 3 Report (New Reviewer)

Comments and Suggestions for Authors

The authors have diligently considered all of the reviewer’s comments, and have satisfactorily revised the manuscript, except in one regard: While the authors have, as requested, identified several sources of overfitting for the KNN model that may have led to the perfect fit (R² =1), they explained these only in their Author Comment letter, and did not incorporate them into the manuscript itself. Additionally, the authors indicated that they would stress-test the KNN model by enlarging the data set, or by tweaking the k value, to see if it altered the fit, but it is not clear to the reviewer if they actually did this. Could the authors further address this concern?

Author Response

Dear Reviewer,

Thank you for your thorough review and for recognizing our efforts in addressing the comments provided. We appreciate your feedback and the opportunity to clarify further and enhance our manuscript.

We acknowledge that, while we identified sources of overfitting for the KNN model in our Author Comment letter, these explanations were not fully integrated into the manuscript. To rectify this, we have now incorporated a detailed discussion of the potential sources of overfitting into the revised manuscript. This includes an analysis of the risk factors that could lead to a perfect fit (R² = 1) and the steps we have taken to mitigate these risks.

Furthermore, regarding your concern about stress-testing the KNN model, we confirm that we have indeed performed additional tests. We enlarged the dataset and adjusted the k value to observe the impact on the model's fit. Specifically, we utilized the elbow method for K-means clustering to determine the optimal k value. By plotting the within-cluster sum of squares against the number of clusters, we identified the point where the rate of decrease sharply slows (the "elbow"), indicating the optimal number of clusters. This method ensured that the chosen k value was robust and minimized the risk of overfitting.

The results of these tests are now included in the manuscript, providing clear evidence of the model's performance under varied conditions. This additional analysis confirms the reliability of our KNN model and demonstrates our commitment to rigorous and comprehensive validation.

We hope that these revisions and the detailed explanations provided will address your concerns and enhance the overall quality and clarity of our manuscript.

Thank you once again for your valuable feedback.

Sincerely,
Dr. Park.

Round 3

Reviewer 2 Report (New Reviewer)

Comments and Suggestions for Authors

I have nothing to add to the comments made earlier.

Reviewer 3 Report (New Reviewer)

Comments and Suggestions for Authors

The reviewer thanks the authors for their timely revision of the manuscript. The revised manuscript has addressed all of the reviewer's concerns.

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The logic of this article is quite confusing, and the methods are not adequately described, such as:

The abstract mentions data from in vitro experiments, but the focus of the results is on DMI.

It is difficult to understand the role of in vitro data in this article.

L17-18, why not predicted methane production based on nutrient intake direactly.

L22-24, how to  optimize， reducing neutral detergent fiber intake?

L70-71, Why choose this 30 articles, as more  than 30 articles is avilable.

Table 1, Hanwoo is not belong to Dariry of Beef?

L69-88, which animal, cattle?

Comments on the Quality of English Language

 Extensive editing of English language required

Author Response

Reviewer 1

1. L17-18, why not predicted methane production based on nutrient intake direactly.

• Response: Thank you for your insightful question. Our approach starts with predicting VFA concentrations from nutrient intake because VFAs are precursors to methane production in ruminants. By first establishing a model that accurately predicts VFA concentrations, we can then convert these values into molar ratios, providing a more robust foundation for the subsequent methane production estimates. This method enhances the precision of our methane predictions by accounting for the intermediate metabolic processes.

2. L22-24, how to  optimize， reducing neutral detergent fiber intake?

• Response: We appreciate your inquiry into the optimization process. Reducing neutral detergent fiber (NDF) intake is crucial because high NDF levels typically result in increased methane production due to the fermentation characteristics of fibrous feeds. By determining the optimal level of NDF intake, we aim to balance nutritional needs with methane emission reduction, thus minimizing energy loss and environmental impact.

3. L70-71, Why choose this 30 articles, as more than 30 articles is avilable.
   Descriptive statistics for the data collected from 30 articles are presented in Table 1

• Response: Thank you for your observation. The selection of these 31 articles was based on specific criteria that included the relevance to our study’s focus, the quality of the data, and the methodologies used. These articles provided comprehensive and comparable data essential for developing our predictive models, ensuring consistency and accuracy in our analysis.

4. Table 1, Hanwoo is not belong to Dariry of Beef?

• Response: Thank you for your query regarding the classification of Hanwoo in Table 1. We initially differentiated Hanwoo due to their unique genetic makeup and the typical purposes for which they are raised, which differ from general beef cattle. However, based on your valuable feedback, we have updated the classification to align Hanwoo under the same category as other beef cattle. We appreciate your attention to detail and thank you for helping us improve the clarity of our manuscript

5. L69-88, which animal, cattle?

• Response: Thank you for your question regarding the specific animal focus. In our study, we used cattle comprehensively, encompassing various breeds and types. This approach allowed us to generalize our findings across the broader category of cattle, enhancing the applicability of our research. We appreciate your interest in the specifics of our study subjects.

Reviewer 2 Report

Comments and Suggestions for Authors

Respected authors

All opinion was emitted with all respect to the efforts of the authors for the preparation of the experiment and its report

The purpose of this study was a development of volatile fatty acid and methane production prediction model using ruminant nutrition comparison of algorithms. Justification for perform the analysis is clear and mathematical and statistical procedure were appropriate. However, the manuscript has serious flaws that impede, in its present form, that this manuscript be considered for publishing in the high quality journal as is Fermentation.

The main concern is the that main focus (and objective) is very confusing, but reading the document I perceive that the purpose of this study was to test various prediction models generated from 31 scientific papers on the results obtained in an In vitro assay in order to prove the its accuracy. However, this is not what is stated in the manuscript. Authors must adequately specify and clarify their objective and put it in line in the description of their Materials and Methods.

Specific

Abstract

Where are the result of comparative observed (in vitro assay) vs predicted??

L19-24: The results shown from lines 19 to 24 do not reveal anything new! What is the contribution of this study? If there is no knowledge novelty, the manuscript not have the quality to be considered by a scientific journal. Please express the novelty findings of the study.

Introduction

L30-31: Definition of ruminant feed is not precise. Please define as (Hint): Ruminant feed is categorized into coarse forages, characterized by contain high concentration of structural carbohydrates (fiber) and low concentration of soluble carbohydrates, and concentrates which contains high soluble carbohydrates (i.e. starch) and low fiber content.

L36-37: Levels of all VFA are increased by increased feed intake, because the total OM fermented in rumen is greater, what changes is the proportion in which AGVs are formed. Therefore, this statement must be change as (Hint): Propionate proportion, which rise with increased soluble carbohydrates in diet, play a pivotal energy supply in tissue formation (muscle and fat).

L41: Methane formation it is important as H2-sink in rumen, but not all of hydrogen formed in rumen is converted in methane (i.e. some is utilized to hydrogenated unsaturated lipids, some pass to the postruminal tract). Thus, the correct statement must be: The most of C2 and hydrogen generated in this process are eventually convert into me-41 thane by methanogens.

L52: to predict methane production

L65: This is the objective?? Or generate a prediction model and validate it based on in vitro assay? Please clarify.

Mat & Methods

Mat & methods description must be improved greatly.

If the purpose of this study was to test various prediction models generated from 31 scientific papers on the results obtained in an In vitro assay in order to prove the its accuracy. Then, the Mat & Methods description must be in line with this.

First describe how the models were developed (through the data analysis for 31 scientific papers), then describes how the laboratory data were obtained to contrast observed to predicted by models. And at last, describe how the predicted models were contrasted with observed result in the in vitro assay.

NOTE for authors (this only a personal view): There are several studies (>1000) in which data of DMI, breed, diet composition, and ruminal VFA and methane production are specified. Therefore, I’ convinced that a specific model of each category must be developed, i.e., from beef cattle studies (in vivo), from Dairy cattle studies (In vivo), and from in vitro studies (since it is very easy to find 30 studies of each one). Finally, all data is used to develop a "universal model" and then evaluate the accuracy of the models in each category, as well as evaluate the predictability of the universal model. This would be an important contribution.

Others (Mat & Method)

L68: I count 31 articles, not 30! Please describe separately “in vivo” and In vitro (Hint): Descriptive statistics for the data collected from 31 articles are presented in Table 1. Data obtained where from 17 in vivo studies and 16 in vitro assays  

L68: Is Table 1, correct. Hanwoo is a different of “beef”. If you are going to differentiate breed, then do it also in the all others (For example, Beauchemin et al. (31) used Angus, Bougouin, et al (37) used Holsteins), if not, then you must categorize Hanwoo as "beef"

L75-77: ???  made a “pool” of data including your results to construct models??? This is not proper to validate a model. Please explain

L83: Table 2. Please check the data presented in this Table. For example, 98 kg DM/day?? or 71 kg of CP?? Even When the nutrient intake is expressed as g/kg this is not possible.  Could the dot be the mistake? 9.8 instead of 98?  Correct this.

Table 2: Again diet composition based on DM can’t be greater than 100! For example, 176% of OM or 160% of EE is not possible. Correct this.

L89-99:  What type of feed (diet ingredient and nutrient composition), how quantity was offered to cows (total and by offered interval) and during what period (days, weeks) was the diet offered before taking the samples? Please include this information and a Table of diet characteristics. If the one of the purpose of the model is prediction of DMI through VFA production in vitro, why did not measure DMI from donors?

Discussion

L279: Affirm? Or Confirm…

L310-311: Diet carbohydrate composition of ruminants basically is composed by NDF and non-NDF components, when NDF is decreased into diet, the non-NDF components (such solubles sugars, starch and other non-fiber CHO) increased, thus methane production decreased.

Author Response

Reviewer 2

1. L19-24: The results shown from lines 19 to 24 do not reveal anything new! What is the contribution of this study? If there is no knowledge novelty, the manuscript not have the quality to be considered by a scientific journal. Please express the novelty findings of the study.

• Response: We appreciate your critical view. This study contributes new insights by quantifying the relationship between nutrient intake and VFA concentrations, particularly how changes in NDF and DMI affect methane production. These findings are crucial as they provide a detailed understanding that aids in the optimization of feed strategies to reduce methane emissions while maintaining animal health and productivity.

2. L30-31: Definition of ruminant feed is not precise. Please define as (Hint): Ruminant feed is categorized into coarse forages, characterized by contain high concentration of structural carbohydrates (fiber) and low concentration of soluble carbohydrates, and concentrates which contains high soluble carbohydrates (i.e. starch) and low fiber content.

• Response: Thank you for your valuable suggestion. We have updated the definition of ruminant feed in our manuscript to provide a more detailed distinction between coarse forages and concentrates, highlighting how the differences in carbohydrate content directly influence ruminal fermentation patterns and subsequent methane production. Additionally, it is important to note that ruminants are often fed a mix of forages and concentrates, sometimes combined into a fibrous mixed ration. Typically, the structure and form of these feeds vary widely, thus the quality of the feed is generally assessed through a basic compositional analysis. This comprehensive approach allows us to better understand the impact of diet on fermentation and methane emissions in ruminants.

3. L36-37: Levels of all VFA are increased by increased feed intake, because the total OM fermented in rumen is greater, what changes is the proportion in which AGVs are formed. Therefore, this statement must be change as (Hint): Propionate proportion, which rise with increased soluble carbohydrates in diet, play a pivotal energy supply in tissue formation (muscle and fat).

• Response: Thank you for the suggestion to refine our statement. We have adjusted the language to emphasize that while total volatile fatty acid (VFA) levels increase with feed intake, it is the proportion of specific VFAs like propionate that is influenced by the type of carbohydrates, particularly soluble carbohydrates in the diet. This distinction is crucial for understanding the nuanced role of diet composition in ruminal fermentation dynamics.

4. L41: Methane formation it is important as H2-sink in rumen, but not all of hydrogen formed in rumen is converted in methane (i.e. some is utilized to hydrogenated unsaturated lipids, some pass to the postruminal tract). Thus, the correct statement must be: The most of C2 and hydrogen generated in this process are eventually convert into me-41 thane by methanogens.

• Response: Your input is valuable, and we have revised this statement to clarify that while a significant portion of hydrogen produced in the rumen is converted into methane, not all hydrogen follows this pathway. Some hydrogen is indeed used for hydrogenating unsaturated lipids or escapes absorption, moving into the post-ruminal tract. This clarification better reflects the complexity of rumen hydrogen dynamics.

5. L52: to predict methane production

• Response: Thank you for your comment on L52 regarding methane production prediction. We have updated the content accordingly. We appreciate your valuable feedback.

6. L65: This is the objective?? Or generate a prediction model and validate it based on in vitro assay? Please clarify.

• Response: Thank you for prompting this clarification. The primary objective is indeed to generate and validate a predictive model for methane production based on VFA concentrations and nutrient intake, using both in vitro and in vivo data. This dual approach enhances the model’s applicability and accuracy in real-world settings.

7. L68: I count 31 articles, not 30! Please describe separately “in vivo” and In vitro(Hint): Descriptive statistics for the data collected from 31 articles are presented in Table 1. Data obtained where from 17 in vivo studies and 16 in vitro assays  

• Response: We apologize for the oversight in the initial count. The manuscript has been corrected to accurately reflect the data from 31 articles, detailing the separation between in vivo and in vitro studies. This provides a clearer understanding of the data sources and their respective contributions to our analysis.

8. L78: Is Table 1, correct. Hanwoo is a different of “beef”. If you are going to differentiate breed, then do it also in the all others (For example, Beauchemin et al. (31) used Angus, Bougouin, et al (37) used Holsteins), if not, then you must categorize Hanwoo as "beef"

• Response: Thank you for your query regarding the classification of Hanwoo in Table 1. We initially differentiated Hanwoo due to their unique genetic makeup and the typical purposes for which they are raised, which differ from general beef cattle. However, based on your valuable feedback, we have updated the classification to align Hanwoo under the same category as other beef cattle. We appreciate your attention to detail and thank you for helping us improve the clarity of our manuscript

9. L75-77: ???  made a “pool” of data including your results to construct models??? This is not proper to validate a model. Please explain

• Response: We have meticulously addressed each point you raised by clarifying our objectives, correcting data errors, and expanding on methodological details to enhance the manuscript's accuracy and comprehensiveness.

10. L83: Table 2. Please check the data presented in this Table. For example, 98 kg DM/day?? or 71 kg of CP?? Even When the nutrient intake is expressed as g/kg this is not possible.  Could the dot be the mistake? 9.8 instead of 98?  Correct this.
11. Table 2: Again diet composition based on DM can’t be greater than 100! For example, 176% of OM or 160% of EE is not possible. Correct this.

• Response: Thank you for pointing out the values of 176% and 160% in Table 2. These figures represent the number of data points, not percentages. This clarification is noted in the footnotes associated with the table. We appreciate your attention to detail and hope this explanation resolves any confusion regarding the data presentation.

12. L89-99:  What type of feed (diet ingredient and nutrient composition), how quantity was offered to cows (total and by offered interval) and during what period (days, weeks) was the diet offered before taking the samples? Please include this information and a Table of diet characteristics. If the one of the purpose of the model is prediction of DMI through VFA production in vitro, why did not measure DMI from donors?

• Response: Thank you for your inquiry about the specifics of feed type, quantity offered, and the period over which the diet was administered before sampling. Our study did not measure Dry Matter Intake (DMI) directly from donors because the primary objective was to develop a predictive model for methane production using meta-analyzed data and in vitro ruminal fermentation tests. We utilized data from multiple studies to create a standardized model, ensuring robust and applicable results across various conditions and experiments. This approach allowed us to integrate a diverse range of experimental data, thereby enhancing the reliability and applicability of our predictive model. Further details on diet characteristics, including ingredients and nutrient composition, as well as feeding protocols, are summarized in a supplementary table provided in our documentation. This comprehensive approach helps in better understanding and predicting methane emissions from ruminants under varied dietary conditions. We appreciate your interest in our methodology and hope this clarification addresses your question.

13. L279: Affirm? Or Confirm…

• Response: Thank you for pointing out the nuance between "affirm" and "confirm." In this context, "confirm" is more appropriate as it more precisely indicates that the study's results definitively support the hypothesized relationships between the levels of acetate, butyrate, and methane production. We have updated the manuscript to reflect this clarification.

14. L310-311: Diet carbohydrate composition of ruminants basically is composed by NDF and non-NDF components, when NDF is decreased into diet, the non-NDF components (such solubles sugars, starch and other non-fiber CHO) increased, thus methane production decreased.

• Response: We appreciate your insight into the implications of dietary composition on methane production. The manuscript has been revised to include a more detailed explanation of how variations in NDF and non-NDF components affect methane production. We now specify that a decrease in NDF typically leads to an increase in rapidly fermentable carbohydrates, which can influence methane emissions differently depending on the overall diet composition. This addition helps to clarify the metabolic and environmental impacts of dietary adjustments in ruminants.

Author Response File: Author Response.pdf