A step towards improved replicability and accessibility of 3D microwear analyses

Dear authors,

After reading your responses and the new version of the manuscript, I want to thank you for all the work you have undertaken. Congratulations for this new software and R package. I am sure it will be useful for many researchers.

I just have one last recommendation: please provide specific links or path to get the raw data on indores. I could not find them, even being myself a French speaker.

Looking forward seeing your manuscript in press,

Best regard,

​​Review comments to PCIPaleo #256 "Introducing ‘trident’: a graphical interface for discriminating groups using dental microwear texture analysis"​​​​​

​Thank you very much for the opportunity to review this very interesting paper. This research is closely related to my interests and the research that I am currently working on.​

In this manuscript, a program package called "trident" developed by the authors for statistical comparison of dental microwear is introduced and its usefulness is demonstrated using three examples. As a researcher deeply involved in dental microwear research, I would like to express my sincere respect to the authors for developing such a wonderful program. This is because recent 3D dental microwear texture analysis (DMTA) involves the calculation of parameters that characterize microwear properties from the surface, but this computation is highly dependent on paid software (MountainsMap, sold by the French company DigitalSurf), which is a potential barrier to entry in this field (because of its high functionality, MountainsMap is very expensive).​

While the paper is clearly written and has no major problems, I believe it is important to further clarify the reliability and usefulness of this program in order for it to be widely used, and some suggestions for manuscript revision are provided below. Some of these require additional calculations, graphing, figures, etc., and I encourage the authors to include them in the revised manuscript. Other minor comments are noted in the PDF.​

I hope to see the revision published in PCIPaleo.​

1) Prerequisite for trident​

Authors explained that trident can read the SUR file, but on p. 9 L. 183-184 the authors write "All surfaces were preprocessed following Mercecron et al, (2016 )". It is necessary to clearly state whether software other than trident is required to prepare the SUR files for analysis and what further processing was done, rather than just showing a citation. This is important to indicate to the reader whether the entire analysis can be completed using only trident, or whether the data can only be used after it has been acquired with profilometers and then pre-processed with software other than trident.​

2) Comparison of calculated values with the standard analysis software MountainsMap​

As mentioned earlier, MountainsMap is widely used as the standard analysis software for DMTA. Therefore, if the values calculated with trident are shown to be consistent with those shown in previous studies, users who have already used MountainsMap can use trident with confidence. Specifically, for basic parameters such as Sq, it would be good to show that the values calculated by MountainsMap and those calculated by trident match on some (e.g. N=10) surface data (without subsampling of the surface).​

3) Advantages of trident​

The superiority of trident's new method (subsampling of the surface and then obtaining the dispersion indices of the parameters, which are then used as new parameters) over the traditional univariate comparison of DMTA parameters, or PCA with multiple parameters without ranking the variables should be shown. For example, Fig. 4C shows a box plot of PC1 using the trident method, but next to it is a box plot using univariate parameter (e.g. Sq), the variable with the largest significant difference between groups using the conventional method, or the conventional PCs, showing that trident's PC1 better captures the trend of the feeding groups. Similarly in Fig. 5, a comparison with the conventional method would be more convincing.​

4) Screenshot of trident in use​

The Supplement includes a manual for trident, which describes in detail the interface and usage of trident, but it would be easier for readers to understand what the program is like if the main manuscript also includes screenshots of trident in use.​

None of the above comments require a great deal of effort on the part of the authors. Again, it is hoped that this software will lead to more DMTA research cases. The search for parameters that better illustrate differences in food properties (including derived parameters obtained by subsampling surfaces) will also provide a better understanding of how differences in food properties produce microwear features, i.e., the etiology of microwear. These future prospects should be discussed as a perspective at the end of the Discussion.​​

General notes. The manuscript presents a new R package, and associated shiny application, called “trident”. The aim of this package is to analyse dental microwears using various measurements to finally detect which measurements are the best to discriminate the dataset into categories based on the diet, taxonomy etc. To illustrate the different applications of this package, the authors analysed three datasets. They detailed the analytical protocol and the results for each case (A, B and C) as well as how the results highlight the power of discrimination of trident.

What makes the trident R package of high interest is that it gathers various dental microwear texture analyses (DMTA) in one place. It considerably eases the investigations and the shiny application helps the users that are not used with R language, providing .SUR files as inputs. Even if I have a very positive opinion on this manuscript, I still have multiple questions and comments that are detailed below and I wish the authors answer to.

Abstract. It synthesises well the aim of the study in overall, but there are few information that are too detailed and few other information that are missing in my opinion. I would advise to remove the sentence about the remove of polynomial surfaces (l. 25-26). I don’t think specifying the number of variables and parameters (l. 26-27) is useful but it would be good to detail the “five different methods” (l. 28; isn’t it four instead, as said at l. 207?). Finally, I would advise the authors to add a first sentence that reminds the reader about the interest of dental microwear in ecological investigations.

Introduction. This section is well written in my opinion. I don’t think Figure 1 is necessary in the main text. Could it be replaced by a reference? Or put in supplementary information? More details are required to better understand this figure. Indeed, there is no information related to the studied taxa and their ecology. The units of each variable (epLsar and asfc) should also be explained.

trident methodology. I think the flowchart in Figure 2 would be more useful if it depicted the whole methodology of trident package, especially when the variables are ranked depending their ability to discriminate categories. Each case A, B and C could then be applied on this general flowchart, using coloured arrows for example. 

I am starting to get lost from lines 280 to 292. Why proposing all these protocols that have only subtle differences?

l. 280-283: I don’t understand the two sentences and so the protocol. What type of arrangement is it at line 280; what is this calculation of mean p-value at lines 281 and 282; what is the post-hoc tests at line 282; what p values are mentioned at line 283?

l. 284-285: what is “a given pair”? What is a “post-hoc p value”?

l.286-292: what data are used to perform Tukey’s HSD?

Case-specific analyses. To be honest, I have not fully understood the applied protocol for each case. A justification of the methodological choices is also missing: why choosing different protocols for these cases? 

The authors keep the variables that are best to discriminate the categories in the different datasets. It is interesting to know which variables differentiate the most the categories, but what is finally the biological meaning of the variables? For me it is still a “jungle of parameters” if we cannot connect the variables to specific type of dental microwears and so to a specific diet. It would have been interested to compare the retained variables between the three cases to see the similarities/differences that could be imputed to the diet but also to the way the animals process the food. Perhaps the correlation circles in Figures 3-5 could be simplified in replacing the name of the variables by the type of microwear they refer to.

Case A. 

l. 147 “… was fed 30% of barley seeds”: does it mean the diet was 70% base + 30% barley, or 100% base + 30% barley? I have the same question for cases B and C.

l. 307 “Variables that passed the multi check were classified”: what are the other variables and are they removed from the further classification? Wouldn’t it be better to favour a Kruskal-Wallis test to keep them in the analyses? Same question for cases B (l. 317) and C (l. 330-331).

l. 356 “the other groups are distinctly separated”: this can be tested using a MANOVA but it depends what the authors want to test, either a significant difference on the first PCA axes (even if it gathers only 60% of the variance), or a significant different based on the input variables that were used to build the PCA.

I cannot read Tables 3 and 4. What do the columns “p value ANOVA” and “Post-hoc p values” refer to? I don’t understand the ranking of the variables since they should be arranged based on their p-value: for example in Table 3, why Sk2 (p-value=0.04) is above Ssk (p-value=0.02) for the pair Co-CK?

l. 459: the separation between the groups has to be tested (see above).

Case B.

l. 323-324: why removing correlated parameters in this case (and in case C, l. 335-336) and not in case A? And why not integrating this step to the R script/trident protocol since it is used in 2/3 of the cases?

l. 385-394: since the diet of the taxa is one of the main focus of the results, I think it would be useful to add this information in Figure 4.

l. 483-484: if there is a continuum, then there is a significant difference between the two extremes that can be statistically tested. This continuum is observed only on PC1 which accounts for ~43% of the overall variance: the variables that contribute the most to this PC should be used as input to test this between-group difference.

l. 503: I don’t see any biomechanics in the discussion.

Case C. The choice of the different taxa is not justified in the main text: do they share phylogenetic relationships? Ecological similarities?

l. 337-338: “The remaining variables were used for a PCA. At this point, the surfaces of Gazellospira torticornis were added as supplementary individuals to the PCA”: so the retained variables for the extinct species are the same than the retained variables for the extant species?

l. 420-427: the authors mention significant differences between the extinct species and some of the extant taxa. These differences can be statistically tested with a MANOVA, based on the input variables that build the PCA. 

Reproducibility. I was not able to use the shiny app as no .SUR file was available as supplementary material. I also tried to use trident package using the .TXT files given as supplementary material. The categories for each case is not easily available as I had to build data frames based on the information from Tables S1-S3. I found the function “trident.arrange” a bit cryptic since there is no detail about the available parameters for the argument “by”.

It would be great if the authors provided the script they used to analyse the data on R and/or the .SUR files so the reader could reproduce the results with the shiny app. The functions could be mentioned in the main text of the manuscript so the reader would know exactly what function to use for each methodological step.

Finally, as the retained variables are specific for each data set, how the analyses can be reproduced within a given taxonomic group? For example, if the data set of the domestic pigs is increased with new data (new specimens or even new diets), the retained variables might change. Then, it would be impossible to compare the new results with the old ones.

Additional remarks.

l. 114: “… and measure 16 variables”. I see only 15 variables in Table 2.

l. 230 and 231: I don’t understand this sentence. What are the 360 computed variables?

l.232-238: I don’t think Box.1 is useful since it is the only box in the whole manuscript. The text could be integrated in the main text.

l. 494-496: I don’t think this information is useful here.