Computational simulations of potential Pachycrocuta bite damage based on a ~1.2 Ma ravaged hippopotamus femur from Fuente Nueva 3 (Orce, Granada, Spain)

  1. Lloyd A. Courtenay
  2. Alexia Serrano-Ramos
  3. Juha Saarinen
  4. Suvi Viranta
  5. Deborah Barsky
  6. Juan-Manuel Jiménez-Arenas
  7. José Yravedra

Reviewed by

Read the full article See related articles

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

Log in to save this article

Abstract

Understanding the behaviour and interactions of extinct carnivoran species present a significant challenge in archaeological and palaeontological research, often limited by numerous constraints in the fossil record. Here we analyse a hippopotamus femur that presents extensive damage by carnivorans, recovered from the open-air site Fuente Nueva 3 ( Pachycrocuta brevirostris . Leveraging the use of advanced microscopic techniques to digitise the tooth marks observed on this specimen in three dimensions, the present study utilises artificially intelligent algorithms to then simulate the possible morphological variability of this carnivoran. This allows us to propose a characterisation of Pachycrocuta brevirostris  tooth pit morphology, so as to construct a proposal for a diagnostic reference sample of this species. If our findings are correct, they underscore the importance tooth mark size has on identifying the activity of Pachycrocuta , revealing the giant hyena to have produced remarkably large, deep, and circular tooth pits on dense cortical bone.

Article activity feed

  1. Version published to 10.24072/pcjournal.771
  2. Peer Community in Archaeology

    Dear Authors,

    Thank you for submitting your manuscript to PCI Archaeology. I appreciate your efforts in employing advanced computational simulations and geometric morphometrics to characterize fossilized bite marks. All reviewers agreed that your work is sound and well-written, providing sufficient information for readers to understand your research question, methods, and results.

    However, there are several aspects that could be addressed to enhance the clarity and rigor of your manuscript. As noted in the reviewers’ comments, a few common themes have emerged, which I outline below:

    While the large number of simulated pits is impressive, it raises concerns about potential bias due to the imbalance between the simulated and manually landmarked data. Using only four samples to generate this many simulations may lead to biases and underrepresentation of variability.
    The manuscript could be improved by including additional figures. In particular, a schematic figure illustrating the VAE process and how the latent space was used to generate simulated tooth mark samples would be beneficial.
    It would be helpful to provide more insight into the motivation for using simulations rather than relying solely on the original data.
    There are concerns regarding class imbalance and the mix of wild and captive specimens.
    One reviewer encouraged you to approach the current work more as a proof of concept, demonstrating that this approach can outline patterns in extinct fauna, provided that good contextual evidence and robust sample sizes are available for training the models, rather than presenting it as a readily available diagnostic tool.
    Finally, please address the issues related to the post-depositional state of the specimens and how this affects measurements (e.g., depth) and the interpretability of your results.
    All the suggestions outlined above are well-supported by the reviewers, and you can find the rest of the comments in their replies. Thank you for considering this feedback. I believe that addressing these points will significantly enhance the quality of your manuscript.

    Best regards,
    Anastasia Eleftheriadou

  3. Peer Community in Archaeology

    REVIEW : Computational simulations of Pachycrocuta bite damage based on a ∼1.2 Ma ravaged hippopotamus femur from Fuente Nueva 3 (Orce, Granada, Spain) Lloyd A. Courtenay, Alexia Serrano-Ramos, Deborah Barsky, Juan-Manuel Jiménez-Arenas, José Yravedra

    doi: https://doi.org/10.1101/2024.07.07.602373

     


    First I’d like to thank the opportunity to read and offer some insight into such an interesting article, I’m very interested in seeing these approaches being used to fill in the gaps for difficult species interactions and the potential 

     


    Computational simulations of Pachycrocuta bite damage based on a ∼1.2 Ma ravaged hippopotamus femur from Fuente Nueva 3 (Orce, Granada, Spain) offers an interesting, proven and compelling methodological approach to aiding the interpretation and reconstruction of large Hyaenids and potentially other carnivoran fauna via analysis of their ichnologic traces.

     


    First I’ll outline some brief comments and opinions related to the text referencing to their line number and then the questionnaire provided for the editorial team, all the comments provided I think could highlight the strongest suits of the manuscript by arranging the argumentation towards the achieved results and strengthen the aspects and conclusions that could be debatable or controversial.

     


    1 - Id advise to shift the title tone into a more speculative one, as it is now it implies many assumptions as facts as I’ll try to outline in the next comments, I’d recommend something along the lines of “Comp. simul. of potential Pachycrocuta bite damage based on a -1.2Ma Artiodactyl femur …”

     


    Abstract - Same as with the title, perhaps a less conclusive tone is needed to address what was actually done in the methodological exercise, I’d advise the following revisions:

    “Here we analyse a potential/hypothesized hippopotamus femur, potentially modified by an extinct species of giant hyena…”
    “This together with contextual stratigraphic information allows us to hypothesize Pachycrocuta b. tooth pit morph…”
    “,so as to construct an ad-hoc/heuristic reference model of this species”
    “…tooth mark size has on hypothesize extinct carnivores BSM patterns, heavily suggesting giant hyenas to have produced…”

     

    Keywords - I’m not entirely sure if the Numerical simulation and MCMC keywords would really help the discoverability of this paper by the right audience, I’d add more keywords related to the geological timeframe, location and both extinct species targeted

     


    47 - 58 - This whole paragraph is crucial to the argument of the whole paper, as it implies that FN3-11-T39-Sup-5-1 could have been modified by other/s multiple agents in different reiterative instances. The admission of the impossibility of establishing causal agent->trace relationships is key to establish the workflow as every statement is an alternative based on cumulative multivariate patterning recognition. See it as a derivative of the notion of zooarchaeology tallies as ordinal at-best, here trace and therefore modification process recognition also possess the same flaws related to sampling. The admission of unlikliness of diagnosis is paradoxically the strongest reason to propose such an excersize as you do in this paper. The paper does not reference these key aspects as frequently as it should, and therefore produces some claims that are in direct contradiction to these proven aspects of BSM diagnostics

     


    60 - 65 - I’d rephrase some of these aspects as to stablish that the framework allows for distinction of a large hyaenid, without implying too strongly that is P. brevirostris at the species level the one possible to infer, even without any other species that clear-cut identification is not possible yet, and certainly not with just one case-scenario.

     


    68 - 69 - It’s a well rounder introduction into the route of diagnostics of carnivores BSM, but you’re not identifing extinct carnivorans, you are documenting their presence via ichnology, therefore perhaps i’d be better to try and address thee traces as new ichnotaxa rather than directly referring them as P. brevirostris bite-marks

     


    72 - 73 - I’d revise as “… we employed the dataset of modern carnivoran tooth pits from Courtenay et al. (2021a,b)

     


    73 - 76 - This phrase is a little confusing to decipher, is it completely neccesary to add? I think it doesn’t add much to what is already stated and some could even misread as that not only carnivores were sampled for comparison.

     


    83 - 85 - Despite what is stated here the tooth pit characteristics and digitization are described in extensive detail further in the paper, delete this phrase. If you refer to the particular details of the tooth pit morphology for the mentioned taxa I’d argue that this phrase is also unnecessary.

     


    209 - 257 - It’s really great that you provided in-deph and in-detail your protocols of surveying and mapping the tooth marks, but i fail to see the need to delineate these in such detail when what was actually compared was GMM to previous tooth pits from the 2021 papers, I’d add this to an appendix or make another methodological/data paper entirely only related to this methodological segment.

     


    285 - 286 - The cylindrical morphology of the shaft and size is insufficient as a criteria to determine the taxonomy of this specimen, the diagnosis of H. antiquus from dental evidence at FN3 layer is also very weak, I’d revise these statements and just classify this specimen as a large Artiodactyl femur unless more compelling diagnostic paleontological criteria is made explicit, I don’t see the issue with naming it an undescript Artiodactyl as the focus of this paper are the BSM by large carnivores

     


    292 - 300 - I think that the fact that the FN3-11-T39-Sup-5-1 specimen presents poor preservation, abrasion and chemical modification observable at surface level might have modified intensely the original patterning, depth and shape of the bite marks, I fail to see these extremely relevant variables mentioned again in the discussions and conclusions, specially in regard to morphology and depth of the gnawing traces. “Opitimal” survey conditions for tooth pit mark morphology are not met, regardless of the absence of trephic modifications, these issues could be partially tackled further down by providing 3d models of P. brevirostris molars and correlate their overall morphology/size with the observed pits, but the case as presented is not very strong considering this key paragraph of other taphonomic modifications.

     


    301 - typo in furrowing “furrowiong”

     


    309 - 314 - This statement is very troublesome in the sense that it feels that the argument for these patterns being attributed to P. brevisrostris appear to be derived in absentia of other large furrowing agents, as stated, in line 42 the Ocre site is described as having evidence of a diverse Carnivoran guild therefore multiple potential agents of gnawing BSM, including the other possible agent Machairodontidae…, the bone assemblage is large and full of an eclectic taxonomic variability, multiple agents could’ve modified these bones in different points in time, assuming that the tooth marks on bone are synchronic and produced by the same agent due to the specimen having furrowing is an incorrect assumption.

     


    Figure 2 - As presented and described the tooth pits do note appear to be 4 distinct marks as mentioned in text, it is understood that for methodological reasons composite pits might not have a desirable shape for comparison via GMM but these traces look like two composite pits, and one isolated notch, not 4 distinct bite events as the text implies.

     


    315 - 324 - Certainly via this method you are in a very compelling manner modeling for these carnivore pits, expanding and delimiting the potential variability of their morphology as these 4 samples provide the limits, but you are not demonstrating P. brevirostris tooth pit morphology, again, the fact that the bone was furrowed (indicative of large hyaenas yes) does not imply that these pits are related to the event, you are here trying to outline an ichnospecies rather than describe the feeding ethology of an extinct species. Again, the negative space of actual P. brevisrostris teeth as shape comparison might be a nice counterpoint to assess empirically how do these pits would look like, same for every other potential agent. While felids are not typically prone to produce tooth marks the do ocasionally leave shallow traces, the FN3-11-T39-Sup-5-1 certainly could be seen as shallow imprints too. You need to revise these claims carefully.

     


    Figure 3 - You should not really name the produced sample as P. brevirostris as it implies that you’ve decided beforehand that the samples were produced by that particular species, the fact is that it is the sample produced amplifying the FN3-11-T39-Sup-5-1 sampled pits, I’d advise to rename in tables and graphs every instance that affirms that the pits were produced by that species.

     


    334 - 338 - I think that considering all the situations mentioned above on other biostratinomic surface modifications, other potential diachronic carnivore gnawing, low sample size, one case scenario and most of the variation. The claim that the potential Pachycrocuta bite marks appear distinct is not sufficiently sustained.

     


    338 - 341 - Depth being a diagnostic criteria is interesting and i think one of the strongest suits of the approach, certainly abrasion and weathering did influence on the final aspect of the traces therefore the “real” depth and shape of the pits was modified implying even deeper traces. Could be interesting to discuss this further.

     


    343 - 348 - This particular phrase is by far the strongest and more compelling contextual argument in favor of the interpretation for these pits as produced by P. brevirostris, as by mere zooarchaeological “ordinal at best” approach, this phrase should not be hidden here in the beggining of your discussion, it would have been perfect to weight these facts with what is stated in lines 309 to 314 about Machairodontinae as a way to present P. brevirostris as way more probable of an agent by virtue of the sheer presence of these remains.

     


    348 - 350 - It sadly is not a clear example of a bone consumed exclusively by hyaenas, and P. brevirostris was not proven to be the agent, it is implied, it’s contextually assumed and compellingly explored, but it is not a clear example, certainly not with just one specimen

     


    351 - 352 - The bone is assumed to be a hippopotamus, being a large Artiodactyl is merit enough to be addressed as that

     


    353 - 355 - But the gracility of the dental structures does not deny the possibility, again a simple model of these species teeth negatives could dispel some doubts when put against the observed traces, it is not a certain method but it allows to illustrate how these shapes and sizes can correlate with the actual tooth pits. Felids do not tend to produce pits as frequently but medium sized relatively gracile large cats can produce a wide array of modifications (see Kaufmann et al. 2018 https://doi.org/10.1016/j.quaint.2016.03.003)

     


    Table 1 - See comments to Figure 3

     


    357 - These claims would be more compelling with teeth models to compare sizes and shapes

     


    360 - 361 - This phrase should guide your abstract and introduction as it presents the case in a parsimonious and fair light

     


    366 - “… all potential P. brevirostris ...“

     


    369 - I think that while it is nice that the small pit provided a more nuanced model, i think that the lack of consideration of the pit being produced by other agent could be a point of debate, did the previous model classified these pits as any of the other species provided? That could be an interesting excercise. Are there really not any instance of “grey” identification areas for these pits?

     


    370 - 371 - Are these samples even possible? What constitutes an undeniable modification by P. brevirostris, I’d figure that some specimen contained within a bromalite could be a nice starting point to explore

     


    371 - 374 - I’d argue that this is a gross understatement, certainly having little to no cortical surface without subtractive modification can skew at least the data regarding depth, this could also modify width and should be more carefully explored as an alternative explanation

     


    375 - 378 - See comments above

     


    380 - 388 - Then probably these samples and datasets should be incorporated to this initial effort to encompass potential variability as it would provide the sample size robustness that this study might lack

     


    388 - 391 - I find this claim a bit misleading, as this model is not predicting the most likely carnivoran, it produced many iterations of the evidence FN3-11-T39-Sup-5-1 and labeled them as P brevirostris and then compared means from other , how these traces were interpreted by the automated sorting before labeling? If this information was provided in the supplementary addendum and is further discussed here then I think it is necessary to be included in text.

     


    391 - 392 - I propose a more parsimonious edit to this phrase “…provides potential taphonomic evidence that suggests that the likely carnivores to have produced these pits was P. brevirostris.”

     


    393 - I don’t think that this statement adds to the argument.

     


    396 - I propose a more parsimonious edit to this phrase “… therefore provides a potential way of understanding P. brevirostris gnawing behavior…”

     


    399 - perhaps a citation would be helpful here as to illustrate examples of ML aiding in elucidating aspects related to tackling equifinality

     


    Title and abstract
    Does the title clearly reflect the content of the article? [X] Yes, [ ] No (please explain), [ ] I don't know
    Does the abstract present the main findings of the study? [X] Yes, [ ] No (please explain), [ ] I don’t know
    Introduction
    Are the research questions/hypotheses/predictions clearly presented? [ ] Yes, [X] No (please explain), [ ] I don’t know
    The introduction does address relevant questions and hypotheses, but their argumentation needs a bit more work in regards of the parsimony of some of the statements presented
    Does the introduction build on relevant research in the field? [X] Yes, [ ] No (please explain), [ ] I don’t know
    Materials and methods
    Are the methods and analyses sufficiently detailed to allow replication by other researchers? [ ] Yes, [X] No (please explain), [ ] I don’t know
    The methods presented are really detailed specifically in regards to describe the sampling, visualization and virtual modeling of the BSM, also the ML method is reiterated sufficiently from Courtenay et al. 2022 to be replicated, all necessary code and data needed to test the method is also provided, the only aspect i think could be omitted is the in length explanation of virtualization of the bite marks, since the final data being analyzed of the sample are the GMM point-data i don’t see the need to explain this aspect of the study in such detail, at least not in the main text (It’s always better to have this though, I’m not saying to just go and delete it).
    Are the methods and statistical analyses appropriate and well described? [X] Yes, [ ] No (please explain), [ ] I don’t know
    Results
    In the case of negative results, is there a statistical power analysis (or an adequate Bayesian analysis or equivalence testing)? [X] Yes, [ ] No (please explain), [ ] I don’t know
    Are the results described and interpreted correctly? [ ] Yes, [X] No (please explain), [ ] I don’t know
    See in depth manuscript review above
    Discussion
    Have the authors appropriately emphasized the strengths and limitations of their study/theory/methods/argument? [ ] Yes, [X] No (please explain), [ ] I don’t know
    See in depth manuscript review above
    Are the conclusions adequately supported by the results (without overstating the implications of the findings)? [ ] Yes, [X] No (please explain), [ ] I don’t know
    I think that some of the claims in the paper are overstating the implications of the results presented regarding the method being a viable and readily available diagnostic tool, certainly the method is very compelling and the implications for diagnostic based on pit morphology have been proven before, but I feel this article should be focused as a proof of concept that this approach could also be useful to try and outline patterns by extinct fauna (provided that good contextual evidence and robust sample sizes are achieved to train the models) rather than presenting a readily available diagnostic tool. Unfortunately considering the small sample size, contextual relationships, other taphonomic processes that could’ve skewed their results (specially abrasion and weathering and undisclosed diagnostic criteria i don’t think that is the case for this article.

     

    I thank again the editors for this opportunity and apologize in advance to the authors if any aspect of this review could be read as rude or accusatory, I blame my inadequacy in english together with my passion for taphonomy in archaeological interpretation for any unintended unpleasant wording.

  4. Peer Community in Archaeology

    The manuscript details a new methodology based on the use of Variational auto-encoder and MCMC to sample the variability present in extinct Pachycrocuta tooth marks. The paper first focus on a description of the novel method where a VAE is trained on a dataset of modern tooth bites and simulations used to generate new theoretical bite marks samples via MCMC. Then in a second part the algorithm is used to simulate the potential variability of Pachycrocuta tooth marks, using forward modelling of the VAE-MCMC algorithm on four newly-sampled Pachycrocuta tooth bites.

    Overall, the manuscript is in a good state and well written. It provides a good description of VAE and MCMC in the method section. However, this section could have been enchanced and easier to understand for the reader if it had been supported by a ‘method’ figure with schematics of the VAE and how the latent space was used to generate simulated tooth mark samples. Similarly, the set of landmarks used for GM analyses doesn’t appear in the current form of the manuscript, it would be clearer if it was present here and the reader was not redirected towards another paper.

    In terms of discussion and conceptually I would have liked the authors to expand on their  interpretations of the forward modeling - what happens when the landmark coordinates of Pachycrocuta samples are put in the algorithm ? Is the MCMC sampling the region of likelihood surrounding the ancient Pachycrocuta toothmarks or converging towards the probability distributions of an existing species present on the training dataset ? While providing a potential framework for simulation and data augmentation I am not sure how much interpretability can be attributed to the algorithm yet.

    Overall, the paper is clear, not overreaching in any of its claims and at the same time provides an innovative simulation framework for geometric morphometric frameworks where the sample data is scarce. I therefore recommend it for publication.

    Title and abstract

    • Does the title clearly reflect the content of the article? Yes
    • Does the abstract present the main findings of the study? Yes

    Introduction

    • Are the research questions/hypotheses/predictions clearly presented? Yes
    • Does the introduction build on relevant research in the field? Yes

    Materials and methods

    • Are the methods and analyses sufficiently detailed to allow replication by other
      researchers? Yes
    • Are the methods and statistical analyses appropriate and well described? Yes

    Materials and methods

    • In the case of negative results, is there a statistical power analysis (or an adequate
      Bayesian analysis or equivalence testing)? Not applicable
    • Are the results described and interpreted correctly? Yes

    Discussion

    • Have the authors appropriately emphasized the strengths and limitations of their
      study/theory/methods/argument? Overall, yes, although as stated before I would have
      like a deeper discussion surrounding the use of VAE simulation in ancient dataset and
      its interpretability.
    • Are the conclusions adequately supported by the results (without overstating the
      implications of the findings)? Yes

     

  5. Peer Community in Archaeology

    Firstly, I believe that the title accurately represents the paper, however, I think it could perhaps mention how the simulations are used to distinguish the bite marks of Pachycrocuta from other taxa. The abstract does a good job of summarising the paper.

    The introduction is clear, but I would like more about the motivation to use a simulation-based approach compared to a standard classification approach (e.g. using CNNs) or a purely morphological-based approach (e.g. using standard GPA). I think the motivation to identify fossil taxa responsible for leaving bit marks in the fossil record is well justified. The introduction is also well-cited, and it is clear where the samples are obtained and how the methodology builds on from previous studies performed by the author (e.g. Courtenay 2021a, 2020). 

    The methods are relatively easy to follow, but I think they need more details throughout. For instance, the software used for landmarking needs to be mentioned and also the R packages used and their versions need to be explicitly stated as they have done for the Python. The use of error measurement is sensible and works well. I would, however, if possible like some figures (probably in the supplementary) that show the latent space or how it works. 

    The results are generally clear and the figures help with understanding the paper, but I think the text needs to be made bigger on the scales to make them more readable. The differences in the estimated lengths, widths and depths are shown as box plots (where there is lots of overlap between species), might an ordination plot show the differences between taxa more clearly (e.g. 3D plot of all three traits for each individual). This feeds into Figure 4, where the eigenvalue for PC1 is almost 100% meaning very little variation is actually explained on the second axis. I, therefore, wonder if there is a better way of distinguishing between taxa, as PC2 is not that informative. 

    I think the authors have identified the benefits and weaknesses of their study. The approach does allow for a quantitative comparison of bite marks, and this is a cool idea. But, I think given the previous research performed on the extant data, they do need more extinct taxa included in the analysis/ training set to improve the paper. They note the limitation, but for a sufficiently novel paper, I think they need to include more taxa. Using only 4 bite marks to simulate variability across 4000 projections, is likely to lead to biases and an underrepresentation of variability. 

    More general comments are found below:

    The study by Courtenay et al. (2024) presents a novel approach to characterising fossilised bite marks using advanced computational simulations and geometric morphometrics. The authors focus on bite marks found on a ~1.2 Ma hippopotamus femur from the Fuente Nueva 3 site in Spain, attributed to Pachycrocuta brevirostris, a giant hyena species. Through their simulations, they aim to enhance our understanding of carnivore-induced damage in the fossil record, with implications for early hominid interactions with predators, migration, and competition.

    This work builds upon previous studies by employing landmark and semilandmark-based geometric morphometrics, combined with advanced machine learning techniques such as Variational Autoencoders (VAE) and probabilistic algorithms with Markov Chain Monte Carlo (MCMC) algorithms. Specifically, the authors simulate tooth mark morphology and compare it to a reference dataset of 823 modern carnivores. Here, they find that P. brevirostris left distinct, large, and deep tooth pits on the fossil, which differ in size and shape from those made by modern hyenas and other carnivores.

    One of the central arguments of the paper is the importance of accurately identifying the species responsible for fossil bite marks, particularly in contexts where early human populations and large carnivores coexisted. The Fuente Nueva 3 site is notable in this regard, as it contains remains of P. brevirostris alongside early hominids and large mammals, many of which show evidence of carnivore activity in the form of bite marks. Figure 1 effectively contextualizes the stratigraphic and geographical location of the site, though the authors should clarify whether their analysis is based on a single specimen or multiple samples (I believe the former).

    Despite their findings, the authors acknowledge limitations in definitively attributing all bite marks to P. brevirostris, as some could have been made by other carnivores. This is a well-balanced discussion, and they reference their earlier work on methods for identifying bite marks from a broader range of species (Courtenay 2021a, 2020). The study benefits from a strong methodological foundation, with the use of landmark-based morphometrics being well-established in prior research. However, the paper could be improved by providing additional figures that illustrate the landmark configuration and by stating the software and packages used in the analyses, such as those for collecting landmarks and performing Generalized Procrustes Analysis (GPA).

    The authors' use of modern carnivores as a reference dataset is reasonable, though there are concerns regarding the class imbalance, particularly the overrepresentation of wolves. Addressing this by simulating data for extant species or by evening out the sample sizes might improve the robustness of the model, but I would like to see even classes for the modern taxa. Furthermore, the mix of wild and captive specimens should be explored in greater depth, as plasticity in captivity could influence bite mark morphology. It would be beneficial to investigate any differences between the two groups to ensure consistent results. I think also given the notes describing how extant taxa cannot be used as analogues for their ancestors, I would like to see more extinct carnivorans in the training set (even if taken from different fossil sites). The authors acknowledge this problem and the lack of other extant taxa (striped, brown hyenas etc.), but I think it is fundamental they are used. 

    The description of the computational techniques, particularly the VAE and MCMC algorithms, is clear and accessible. The VAE is appropriately designed to capture the diversity of tooth pit morphology, with 15 latent dimensions deemed optimal. The authors' use of loss functions, including the square root of the mean squared error and the Kullback-Leibler divergence, is sound and justified. They report no evidence of overfitting, though an early stopping mechanism might have been useful to prevent potential overfitting during training. Further clarification on the software used for these analyses would enhance the reproducibility of the study, particularly the R and Python packages employed (though I see some of the Python packages are listed and the GitHub page is easily navigated).

    One of the most significant contributions of this paper is the generation of 3D models based on silicon moulds of the fossil tooth pits. These were digitised using confocal microscopy, providing high-resolution point cloud data, which were then used to produce detailed models of the pits. The landmark coordinates were superimposed with the reference samples, and the VAE and MCMC algorithms were employed to simulate 4,000 new tooth pits based on the Pachycrocuta specimen. While this large number of simulated pits is impressive, it raises concerns about potential bias due to the imbalance between the simulated and manually landmarked data. Using only 4 samples to generate this many simulations is likely to lead to some biases and underrepresentation of variability. 

    The results reveal that P. brevirostris likely produced larger and deeper tooth pits than modern hyenas and other carnivores. The principal component analysis (PCA) of these pits shows that most of the variation is explained by size (PC1 accounts for 98.31% of the variability), with the length-to-width ratio contributing to PC2. While this is informative, the overwhelming influence of size on the variation raises questions about whether other aspects of tooth pit morphology could be explored further, for instance may the centroid size provide the same answer. It also means in differences along PC2 have significantly reduced meaning, so using this axis to find differences between taxa is less meaningful.  

    Figures 2 and 3 are clear and informative, though the text in Figure 3 could be enlarged for readability. The PCA results indicate that the tooth pits left by P. brevirostris are distinct from those made by modern carnivores, with relatively circular and deep pits being a key characteristic. However, the authors note that these pits are somewhat similar to those made by modern lions, particularly in terms of size variation. This suggests that while P. brevirostris can be distinguished from other carnivores, further refinement of the method may be necessary to account for overlap with larger modern species.

    Overall, I like the idea but I think it needs further refinement by bringing in a wider range of taxa (if possible) and fixing the issues of class balance. The idea is also very complex, and thus I have concerns about the widespread use, which is why I think further justification in the introduction and an explicit statement of the software and R/ python packages used is needed. I am also curious to know if just plotting the landmark data using normal Procrustes transformations and generating PCAs may yield the same results without the added complications (and by extension I was wondering if the authors have tested this?). I understand why the authors use the simulations (to allow for more variations), but I think it may overcomplicate a more simple problem (unless there is a paper that I am unaware of that discusses why this approach is needed). I would, therefore, also like to hear more about the motivation for using simulations rather than the original data alone, and why it may be the best approach for identifying the taxa responsible for leaving bite marks in the fossil record. 





  6. Peer Community in Archaeology

    A key challenge in taphonomic research is identifying the agents affecting the formation of archaeological and palaeontological sites (e.g. Binford et al., 1985; Blumenschine et al., 1987; Dominguez-Rodrigo, 1997; Pante et al., 2012; Selvaggio, 1994). In the field of bone surface modifications (BSMs), recent studies are using a range of computational methods, including geometric morphometrics and machine learning, to analyse tooth marks (e.g. Abellán et al., 2021; Cifuentes-Alcobendas & Domínguez-Rodrigo, 2021; Courtenay et al., 2019, 2024; Courtenay, Herranz-Rodrigo, et al., 2020; Courtenay, Huguet, et al., 2020; Jiménez-García et al., 2020; Moclán et al., 2024; Pizarro‐Monzo et al., 2022; Yravedra et al., 2021). However, the presence of extinct taxa among potential agents influencing site formation raises issues, as modern taxa may not accurately reflect their ancestral counterparts, and obtaining more examples of such marks for analysis is often difficult or sometimes impossible. 

    To bridge the gap, the paper ‘Computational simulations of potential Pachycrocuta bite damage based on a ∼1.2 Ma ravaged hippopotamus femur from Fuente Nueva 3 (Orce, Granada, Spain)’ by Courtenay et al. (2025), presents an innovative approach for the analysis of tooth marks produced by extinct carnivoran taxa using artificially intelligent (AI) algorithms. 

    More specifically, the authors use Variational Autoencoder (VAE) and Markov Chain Monte Carlo (MCMC) to simulate the morphological and morphometric variability of tooth marks identified on a bone specimen from the open-air site of Fuente Nueva 3 (∼1.2 Ma) in Spain (ID: FN3-11-T93-5-1). The specimen is a hippopotamoid femur with seven tooth marks, which, based on archaeological and paleontological records, are most likely attributed to the giant hyena Pachycrocuta brevirostris. 

    Utilizing VAEs and MCMC algorithms, the authors modelled the potential variability of Pachycrocuta tooth pits based on four tooth marks from FN3-11-T93-5-1, with a 0.074 mm margin of error in predictions. After comparing them with the morphometric parameters of a reference collection of different agents, Pachycrocuta was shown as the most probable agent responsible for the tooth marks of FN3-11-T93-5-1. The attribution was not based solely on model outputs and morphological metric comparisons but also integrated the animals' dietary profiles and behaviours (i.e. durophagy), along with other bone surface modifications (i.e. furrowing) and additional archaeological evidence from the site (i.e. coprolites), providing a more holistic approach. 

    The authors acknowledge the limitations of their work, including taphonomic palimpsests, potential variability in Pachycrocuta tooth marks not captured on the specimen, and the effects of preservation on their results. On this end, the authors agree that their study could be strengthened further by including more samples of confirmed Pachycrocuta activity and by comparing it with other hyaenid species. 

    A key point of this study is that it does not aim to replace the original observations or determine the definitive agent responsible for the tooth marks on the FN3-11-T93-5-1 specimen. In contrast, the authors propose a practical approach to estimate the expected variation in the tooth mark morphology of extinct taxa, thereby providing a statistical framework for their quantitative analysis. 

    To conclude, this work introduces a novel computational approach to a classic zooarchaeological and palaeontological question. By combining AI algorithms with statistics, the authors have created a heuristic tool that can help researchers analyse tooth marks of extinct taxa through simulations and thus test hypotheses about site formation through a different methodological lens.

    References

    Abellán, N., Jiménez-García, B., Aznarte, J., Baquedano, E., & Domínguez-Rodrigo, M. (2021). Deep learning classification of tooth scores made by different carnivores: Achieving high accuracy when comparing African carnivore taxa and testing the hominin shift in the balance of power. Archaeological and Anthropological Sciences, 13(2). https://doi.org/10.1007/s12520-021-01273-9

    Binford, L. R., Ho, C. K., Aigner, J. S., Alimen, M.-H., Borrero, L. A., Te-K’un, C., Chung, T., Goldberg, P., Ikawa-Smith, F., Lanata, J. L., Zune, L., Luchterhand, K., Lyman, R. L., Goñalons, G. M., Pei, G., Straus, L. G., Yacobaccio, H. D., & Yi, S. (1985). Taphonomy at a distance: Zhoukoudian, “the cave home of beijing man”? [And comments and reply]. Current Anthropology, 26(4), 413–442. https://doi.org/10.1086/203303

    Blumenschine, R. J., Bunn, H. T., Geist, V., Ikawa-Smith, F., Marean, C. W., Payne, A. G., Tooby, J., & van der Merwe, N. J. (1987). Characteristics of an early hominid scavenging niche [and comments and reply]. Current Anthropology, 28(4), 383–407. https://doi.org/10.1086/203544

    Cifuentes-Alcobendas, G., & Domínguez-Rodrigo, M. (2021). More than meets the eye: Use of computer vision algorithms to identify stone tool material through the analysis of cut mark micro-morphology. Archaeological and Anthropological Sciences, 13(10). https://doi.org/10.1007/s12520-021-01424-y

    Courtenay, L. A., Herranz-Rodrigo, D., Huguet, R., Maté-González, M. Á., González-Aguilera, D., & Yravedra, J. (2020). Obtaining new resolutions in carnivore tooth pit morphological analyses: A methodological update for digital taphonomy. PloS One, 15(10), e0240328–e0240328. https://doi.org/10.1371/journal.pone.0240328

    Courtenay, L. A., Huguet, R., González-Aguilera, D., & Yravedra, J. (2020). A hybrid geometric morphometric deep learning approach for cut and trampling mark classification. Applied Sciences, 10(1), 150. https://doi.org/10.3390/app10010150

    Courtenay, L. A., Serrano-Ramos, A., Saarinen, J., Viranta, S., Barsky, D.,  Jiménez-Arenas, J. M., Yravedra, J. (2025) Computational simulations of potential Pachycrocuta bite damage based on a ~1.2 Ma ravaged hippopotamus femur from Fuente Nueva 3 (Orce, Granada, Spain). bioRxiv, ver.2 peer-reviewed and recommended by PCI Archaeology https://doi.org/10.1101/2024.07.07.602373

    Courtenay, L. A., Vanderesse, N., Doyon, L., & Souron, A. (2024). Deep learning-based computer vision is not yet the answer to taphonomic equifinality in bone surface modifications. Journal of Computer Applications in Archaeology, 7(1), 388-388–411. https://doi.org/10.5334/jcaa.145

    Courtenay, L. A., Yravedra, J., Huguet, R., Aramendi, J., Maté-González, M. Á., González-Aguilera, D., & Arriaza, M. C. (2019). Combining machine learning algorithms and geometric morphometrics: A study of carnivore tooth marks. Palaeogeography, Palaeoclimatology, Palaeoecology, 522, 28–39. https://doi.org/10.1016/j.palaeo.2019.03.007

    Dominguez-Rodrigo, M. (1997). Meat-eating by early hominids at the FLK 22 Zinjanthropus site, Olduvai Gorge (Tanzania): An experimental approach using cut-mark data. Journal of Human Evolution, 33(6), 669–690. https://doi.org/10.1006/jhev.1997.0161

    Jiménez-García, B., Aznarte, J., Abellán, N., Baquedano, E., & Domínguez-Rodrigo, M. (2020). Deep learning improves taphonomic resolution: High accuracy in differentiating tooth marks made by lions and jaguars. Journal of the Royal Society Interface, 17(168), 20200446. https://doi.org/10.1098/rsif.2020.0446

    Moclán, A., Domínguez-Rodrigo, M., Huguet, R., Pizarro-Monzo, M., Arsuaga, J. L., Pérez-González, A., & Baquedano, E. (2024). Deep learning identification of anthropogenic modifications on a carnivore remain suggests use of hyena pelts by Neanderthals in the Navalmaíllo rock shelter (Pinilla del Valle, Spain). Quaternary Science Reviews, 329. https://doi.org/10.1016/j.quascirev.2024.108560

    Pante, M. C., Blumenschine, R. J., Capaldo, S. D., & Scott, R. S. (2012). Validation of bone surface modification models for inferring fossil hominin and carnivore feeding interactions, with reapplication to FLK 22, Olduvai Gorge, Tanzania. Journal of Human Evolution, 63(2), 395–407. https://doi.org/10.1016/j.jhevol.2011.09.002

    Pizarro‐Monzo, M., Organista, E., Cobo‐Sánchez, L., Baquedano, E., & Domínguez‐Rodrigo, M. (2022). Determining the diagenetic paths of archaeofaunal assemblages and their palaeoecology through artificial intelligence: An application to Oldowan sites from Olduvai Gorge (Tanzania). Journal of Quaternary Science, 37(3), 543–557. https://doi.org/10.1002/jqs.3385

    Selvaggio, M. M. (1994). Carnivore tooth marks and stone tool butchery marks on scavenged bones: Archaeological implications. Journal of Human Evolution, 27(1–3), 215–228. https://doi.org/10.1006/jhev.1994.1043

    Yravedra, J., Herranz-Rodrigo, D., Mendoza, C., Aragón-Poza, P., & Courtenay, L. A. (2021). The use of tooth marks for new research into identifying and understanding the first domestic dogs in Palaeolithic populations. Journal of Archaeological Science, Reports, 40, 103252. https://doi.org/10.1016/j.jasrep.2021.103252

  7. Version published to 10.1101/2024.07.07.602373 on bioRxiv