Utilization of FTIR and Machine Learning for Evaluating Gluten-Free Bread Contaminated with Wheat Flour
Round 1
Reviewer 1 Report
Manuscript ID: sustainability-2326721
Title: Utilization of FTIR and machine learning for evaluating gluten-free bread contaminated with wheat flour
Authors: Maryam Mottaghi, Theresa K. Meyer, Ross John Tieman, David Denkenberger and Joshua M. Pearce
Dear authors,
your manuscript Utilization of FTIR and machine learning for evaluating gluten-free bread contaminated with wheat flour is about promising and interesting analysis of hidden gluten and the use of this research could be very interesting. The method applied in the manuscript is well-written and gave novelty to gluten analysis. But the formulation of cornbread making is not quite clear and I suggest it be better explained.
It could be seen that you invested a lot of effort and time in making quality research.
The suggested manuscript is scientifically sound and well-written. The theme of the proposed paper is very interesting and I think would be very useful in everyday life, especially the mobile application. The manuscript is suitable to be published in Sustainability after some minor corrections.
In the pdf file, you will find suggestions for improvement of the written text.
Best regards!
Comments for author File: 
Comments.pdf
Author Response
Sustainability-2326721 Responses to Reviewer 1
Comment: The formulation of cornbread making is not clear explained and is unclear to repeat. Could you please explain it in more details to be easy to follow formulation.
Response: Please refer to the updated manuscript (Lines: 120-126) in which the section 2.1 (ingredients for breadmaking) has been updated as follows:
"Corn flour (CF) and wheat flour (WF) used in this study were purchased from Bob’s Red Mill Natural Foods (Milwaukie, OR, US). The formulation used with the basic bread ingredient was adopted from [31]. The cornbread formulations included corn flour (100 %) and the following ingredients were added based on the weight of the corn flour: water (70 %), dried yeast (2 %), salt (2 %), sugar (2 %), vegetable fat (3 %); and 0 – 10 % wheat flour was used for contamination at 0.5 % increment up to 1.5 % and then at 1 % increment afterwards until 10 % contamination level, which resulted in 13 different formulations. "
Comment: Dot instead
Response: Please refer to the updated manuscript (Line: 130) where this was corrected.
Comment: Could not find Figure 4.1. Could you please check it?
Response: Please refer to the updated manuscript (Line 237) where this was corrected to Figure 1.a
Comment: Could you please check the value. In Table 5 it is 0.34?
Response: Please refer to the updated manuscript (Line 307) where this was corrected to 0.34
Comment: Could you please check the value. In Table 5 it is 0.34?
Response: Please refer to the updated manuscript (Line 311) where this was corrected to 0.34
Comment: Please, use capital letters on figures instead small letters to be compatible with the description of Figure 3 and the discussion.
Answer: Please refer to the updated manuscript capital letters were used to refer to sub-figures in figure 1 and figure 3.
Reviewer 2 Report
How can an accident due technological process be determined by the method presented, when a single batch shows high amounts of gluten through wheat flour contamination while the rest of the products fall within the normal parameters of gluten-free? In the case of an mobile phone application that compares the products with a database, how app warn the consumer that the product is high in gluten, if all the other batches comply with the gluten free content?
Author Response
Sustainability-2326721 Responses to Reviewer 2
Comment: How can an accident due technological process be determined by the method presented, when a single batch shows high amounts of gluten through wheat flour contamination while the rest of the products fall within the normal
parameters of gluten-free?
Response: We refer the reviewer to Table 6 which was just added to show ELISA test results for gluten detection in the 13 formulations used for the study. It is clear that at least six of the formulations were gluten-free and four were gluten contaminated, contrary to the position of the reviewer that only ONE formulation is considered to be gluten contaminated. There is no other way we could have obtained gluten contamination and applied supervised machine learning in the regression analysis without formulating gluten-contaminated bread ourselves by the addition of a gluten-containing flour (wheat flour) to a non-gluten flour (cornmeal).
Thus, the methodology developed in the study can be applied as an off-line technique for quality control in factories that handle such types of powders (i.e. bakery factories), especially with the fact that FT-IR is already used as a rapid quality assurance method in several industries.
Comment: In the case of an mobile phone application that compares the products with a database, how app warn the consumer that the product is high in gluten, if all the other batches comply with the gluten free content?
Response: We think the reviewer misunderstood how a phone APP will be used in the context of the approach presented in the manuscript. A phone or the APP to be built is never meant to collect FTIR data, it is meant for quickly analyzing data collected with an FTIR instrument and to provide an output for the amount of gluten present, and tell if it crosses the threshold for gluten contamination. In this case, the mobile phone application can be used for analyzing the spectra obtained from the FT-IR system and then applying the model trained before and obtaining a decision about the contamination possibility of wheat flour in the product (i.e. cornbread). Thus, applying the application here is restricted to the availability of the FT-IR sensor and the app. will not require any machine learning skills to obtain a fast and relatively accurate decision about the contamination of cornbread by gluten.
An APP was not part of the objective of this paper, it was mentioned as a future work, which we have actually developed and it worked as described - we have no intention of including the APP in this paper because it is outside of the scope of what we want presented.
Reviewer 3 Report
Gluten-free products have a vast number of customers and some of whom cannot tolerate the presence of gluten in foods. The authors showed an approach using Fourier Transform Infrared (FTIR) spectroscopy coupled with machine learning (ML) to detect and quantify wheat flour (WF) contamination in gluten-free cornbread. The results will be contributed to the rapid offline evaluation of wheat flour contamination in gluten-free products. However, some important results and conclusions should be clarified clearly before considering for publication.
How effective is this method for evaluating gluten-free bread contaminated with wheat flour in actual samples?
Whether the influence of spectrometer sensitivity on this method is considered.
Line 157: The machine learning module from the open source in this study, so whether it is suitable for the detection of other types of contamination in food.
Author Response
Sustainability-2326721 Responses to Reviewer 3
Comment: How effective is this method for evaluating gluten-free bread contaminated with wheat flour in actual samples?
Response: The prediction results provided in the study were developed for the test set data, that is, a separate or independent data set that was not used in building the classification or regression models. Thus, the prediction models showed robust performance when applied to a different data set. However, the performance of the training models can be enhanced by adding data for various diverse sources of corn and/or wheat flour which in the end brings the models to be generalised for real applications.
Comment: Whether the influence of spectrometer sensitivity on this method is considered.
Response: While the sensitivity is an important factor of measurement systems, the sensitivity of the FTIR system used in this study was not considered as an interfering factor of the acquired spectra, especially with the fact that canning of all samples was conducted at the same operating conditions which lead to training models capable of detecting contaminated samples with 100% of accuracy.
Comment: Line 157: The machine learning module from the open source in this study, so whether it is suitable for the detection of other types of contamination in food.
Response: The authors would like to state that this open source provides a wide range of machine learning algorithms that can be programmed for whichever the problem is. In our study, we focused on developing classification and regression models for detecting and quantifying gluten contaminations from wheat flour. Thus, the models are specific to gluten from wheat. However, it is possible to conduct transfer learning by which gluten from other grains like barley and rye that have slight differences in terms of their spectral signature can be detected. In our first study in this project, we did look at the detection of gluten from the main gluten-containing grains, and we found that a model for one cannot be used for detecting the other.
Reviewer 4 Report
The authors explored the possibilities of using the application of FTIR to predict the gliadin contamination in corn-based gluten-free breads.
The study is interesting and can be an alternative to ELISA-based detection methods, which require skilled manpower to evaluate the gluten contamination in gluten-free breads.
Some clarifications are required before going further.
CD-related peptides can be found in oat, barley, wheat, durum, etc.,
FTIR detects alpha helix, beta sheets, anti-parallel beta sheets, random coils, etc., at very fine levels, which can also be analyzed by Raman; however, the detection of specific kinds of secondary structure by FTIR and Raman vary at some levels.
For example
as mentioned by authors
Lines -312: "Cui et al. [44] followed a similar approach for optimizing the hyperparameters (i.e. gagging fraction and frequency) for the gradient boosting decision tree ensemble learning to enhance regression models based on FTIR data to predict the decontamination rate of cold plasma-treated Chitosan-DNA films. 315
Here DNA and chitin can be detect easily because of distinct structural differences.
Wheat flour was blended with corn flour in the range of 0.5% and 10%, and baked products were subjected to FTIR analysis after baking.
The author has not used any other protein source, like egg white, pulse protein isolates, meat protein isolates, pulse flour, pseudocereal flour, etc., for the validation of changes in the secondary structure of proteins after baking.
The changes reported herein in MS cannot be claimed to be because of gluten contamination. So many protein-protein, protein-starch and protein-starch-lipids (amylose-lipid) interactions occurred during the melting of starches. So the changes reported in MS in Fig 1 and fig 3 may be related to some other structural events.
Entire data must be validated by ELISA, NIR, and ESI-qToF. ESI-QToF can easily detect the presence of CD peptides on food material. Therefore, I am not satisfied with the methodology used to validate the hypothesis.
Control is missing in Figure 1:
Figure 1. (a) FTIR-spectra of the corn flour contaminated with 0.5% wheat flour (raw sample); (b) 227 FTIR-spectra of baked cornbread contaminated with 0.5 % wheat flour; (c) FTIR-spectra of baked 228 cornbread mixed with different levels of wheat flour (0.5 – 10 % at 0.5 % increment).
Author Response
Sustainability-2326721 Responses to Reviewer 4
Comment: The author has not used any other protein source, like egg white, pulse protein isolates, meat protein isolates, pulse flour, pseudocereal flour, etc., for the validation of changes in the secondary structure of proteins after baking.
The changes reported herein in MS cannot be claimed to be because of gluten contamination. So many protein-protein, protein-starch, and protein-starch-lipids (amylose-lipid) interactions occurred during the melting of starches. So the changes reported in MS in Fig 1 and fig 3 may be related to some other structural events.
Entire data must be validated by ELISA, NIR, and ESI-qToF. ESI-QToF can easily detect the presence of CD peptides on food material. Therefore, I am not satisfied with the methodology used to validate the hypothesis.
Response: We respectfully disagree with the reviewer that what was detected could not have been gluten in the baked bread because we did not use other protein ingredients like egg, pea protein isolate, etc., as the ingredients used in the experiments was referenced for bread making and not other baked products. Corn flour was intentionally contaminated with gluten at pre-defined levels, and corn does not contain gluten naturally. So, wheat-contaminated corn flour will only show a response to FTIR if the secondary structure of proteins in it is gluten, not zein fraction which is the predominant protein in corn. We did not have to use other types of gluten sources for validation because they will not make any difference as the main target was to detect the gluten protein and not other sources of proteins. We did conduct an ELISA test on all the treatments but did not think it was necessary to provide the result in the manuscript. We have added the results of the ELISA test to indicate that what was detected was indeed gluten not protein-starch complex or others as alluded to by the reviewer.
Comment: Control is missing in Figure 1:
Figure 1. (A) FTIR-spectra of the corn flour contaminated with 0.5% wheat flour (raw sample); (B) FTIR-spectra of baked cornbread contaminated with 0.5 % wheat flour; (C) FTIR-spectra of baked cornbread mixed with different levels of wheat flour (0.5 – 10 % at 0.5 % increment).
Response: Please refer to the updated manuscript where a sub-figure (Figure 1A) was added to show an example of the FTIR spectrum of a pure corn flour. Please note that Figure 1 is just a sample of the FTIR spectra pure and one contamination level (0.5 % wheat flour contamination level) of the several treatments in the study. The actual data analyzed included data for the control and the wheat-contaminated corn flours.
Round 2
Reviewer 4 Report
Sustainability-2326721 Responses to Reviewer 4
Authors response to my questions is not satisfactory.
The supplemented figure (Examples of (A) FTIR-spectra of pure corn flour;) is not matching with other figures.
Data presented in table 6 need clarifications.
Please follow the link
https://elibrary.asabe.org/abstract.asp?aid=51366
title of table 6 is from table 5.2 of a master thesis. while data presented in table 6 is fetched from table 5.2 (Table 5.3: Quantification of the amount of gluten in ppm for the processed flour (bread) samples contaminated with wheat flour (WF))
Original thesis is attached herewith report for the kind perusal of EIC.
https://elibrary.asabe.org/abstract.asp?aid=51366
We predict secondary structure by FTIR by calculating the Second derivative of the processed FTIR spectra, which gave us information about the proportion of alpha helix, beta plated and anti-parallel beta plated sheets, random coils etc.,
Authors should do the deconvolution of FT-IR spectra and check effect of baking on the gluten structure.
Thanks
I strongly recommend use of other protein sources also.
Comments for author File: Comments.pdf
Author Response
Comment: Authors response to my questions is not satisfactory.
The supplemented figure (Examples of (A) FTIR-spectra of pure corn flour;) is not matching with other figures.
Response: We have tried to address the reviewer's comments as best as possible. We have provided further clarification based on the reviewer's new comments.
The authors would like to state that the purpose of different line formats was to distinguish the spectrum of pure corn flour from other formulations with wheat flour. In fact, the spectrum for corn flour is expected to be slightly different from the rest. The newly added Figure 1A (as requested by the reviewer) was made on Excel with the extracted numerical data from the FTIR scan - we do not have access to the FTIR-generated interface to recapture the cornflour data in the original format as shown for the other others, hence the slight difference in formation. An obvious difference is around 2,300 cm-1 in Fig 1A, there is an absorbance in corn not seen in others with wheat flour. Unfortunately, this is the most we can provide in terms of spectrum for the corn formulation.
Comment: Data presented in table 6 need clarifications.
Please follow the link: https://elibrary.asabe.org/abstract.asp?aid=51366
title of table 6 is from table 5.2 of a master thesis. while data presented in table 6 is fetched from table 5.2 (Table 5.3: Quantification of the amount of gluten in ppm for the processed flour (bread) samples contaminated with wheat flour (WF))
Original thesis is attached herewith report for the kind perusal of EIC.
https://elibrary.asabe.org/abstract.asp?aid=51366
Response: We believe the reviewer is mixing a lot of unrelated information from the thesis of the student that worked on this study, the proceeding paper published on some aspect of his thesis that is different from this manuscript, and this manuscript.
The reviewer incorrectly stated that Table 6 in this manuscript is the same as Table 5.2 from the student's thesis. Second, the reviewer incorrectly referred to a proceeding paper (https://elibrary.asabe.org/abstract.asp?aid=51366) that was extracted from one objective in the thesis.
If the reviewer and the Managing Editor are interested in the student's thesis, here is the link to it: https://uknowledge.uky.edu/cgi/viewcontent.cgi?article=1076&context=bae_etds
Table 5.2 from the thesis is for RAW formulations (flours/dough yet baked). Table 5.3 (Table 6 - that was just added because the reviewer asked for the data on the ELISA test to prove what was detected by FTIR was actually gluten) from the thesis is where the reviewer should please focus attention because this manuscript is about predicting gluten in baked bread and not raw flours/dough. The thesis and this manuscript are not different entities but the correct portion of the thesis should be compared with this manuscript. Many of the questions and comparisons the reviewer makes are quite clear and differentiable with closer attention.
Comments: Authors should do the deconvolution of FT-IR spectra and check effect of baking on the gluten structure.
Response: It is not clear what the reviewer meant by the deconvolution of the FTIR spectrum. If the reviewer was referring to the data he saw in the proceeding paper or the student thesis, that is for raw flour/dough not baked bread. And the spectra transformation was a preprocessing step. The authors would like to state that using 2nd derivative was meant as a preprocessing method for eliminating noise related to the FT-IR system, sampling, and environmental conditions that interfere the developed models. The study conducted by Chompoorat et al. (2022), stated that heat-treated gluten showed similarity of the 2nd derivative of FT-IR spectra compared with control gluten that was not subjected to heat and additives which leads to a conclusion that the secondary structure of gluten reflected similar FT-IR behavior as the gluten before baking. Moreover, the study also concluded that no new protein was confirmed after baking.
Reference: Chompoorat, P., Fasasi, A., Lavine, B. K., & Rayas-Duarte, P. (2022). Gluten Conformation at Different Temperatures and Additive Treatments. Foods, 11(3), 430.
Comment: I strongly recommend use of other protein sources also.
Response: The authors agree with the reviewer regarding using other protein sources (milk, nuts, eggs, fish, and soy) which has been suggested as future work in the updated manuscript (Lines: 372-374) : ” Moreover, future research should consider testing other glycoproteins (milk, nuts, egg, fish, and soy). Our focus in this study was to test the possibility of using FTIR to detect and quantify gluten from wheat flour in gluten-free and gluten-contaminated bread.
