Find research datasets worth reusing

opencc-zeroDec 2020View details →

Predicting how climate change threatens the prey base of Arctic marine predators

Arctic sea ice loss has direct consequences for predators. Climate-driven distribution shifts of native and invasive prey species may exacerbate these consequences. We assessed potential changes by modelling the prey base of a widely distributed Arctic predator (ringed seal; Pusa hispida) in a sentinel area for change (Hudson Bay) under high- and low-greenhouse gas emissions scenarios from 1950 to 2100. All changes were relatively negligible under the low-emission scenario, but under the high-emission scenario, we projected a 50% decline in the abundance of the well-distributed, ice-adapted, and energy-rich Arctic cod (Boreogadus saida) and an increase in the abundance of smaller temperate-associated fish in southern and coastal areas. Further, our model predicted that all fish species declined in mean body size, but a 29% increase in total prey biomass. Declines in energy-rich prey and restrictions in their spatial range are likely to have cascading effects on Arctic predators.

Temperature predicts the rate of molecular evolution in Australian Eugongylinae skinks

Temperature differences over time and space has been hypothesized to cause variation in the rate of molecular evolution of species, but empirical evidence is mixed. To further test this hypothesis, we utilized a large exon-capture sequence data of Australian Eugongylinae skinks, exemplifying a radiation of temperature-sensitive ectotherms spanning a large latitudinal gradient. The association between temperature (and other species traits) and long-term substitution rate was assessed based on 1268 sequenced exons of 44 species pairs from the Eugongylinae subfamily using regression analyses. Temperature is the strongest, positively-correlated predictor of variation in substitution rate across the Australian Eugongylinae. It explains 45% of variation in synonymous substitution rate, and 11% after controlling for all the other factors. Synonymous substitution rate is also negatively associated with body size, with 6% variation explained by body size after controlling for the effects of temperature. Other factors are not associated with synonymous substitution rate after controlling for temperature. Overall, this study points to temperature as a strong predictor of the molecular evolution rate in the Eugongylinae subfamily, and demonstrates the power of large-scale exonic data to identify correlates of the rate of molecular evolution.

Haplotype-based genome-wide association increases the predictability of leaf rust (Puccinia triticina) resistance in wheat

Resistance breeding is crucial for a sustainable control of wheat leaf rust and SNP-based genome-wide association studies (GWAS) are widely used to dissect leaf rust resistance. Unfortunately, GWAS based on SNPs explained often only a small proportion of the genetic variation. We compared SNP-based GWAS with a method based on functional haplotypes (FH) considering epistasis in a comprehensive hybrid wheat mapping population composed of 133 parents plus their 1,574 hybrids and characterized with 626,245 high-quality SNPs. In total, 2,408 and 1,139,828 significant associations were detected in the mapping population by using SNP-based and FH-GWAS, respectively. These associations mapped to 25 and 69 candidate regions, correspondingly. SNP-based GWAS highlighted two already-known resistance genes, i.e. Lr22a and Lr34-B, while FH-GWAS not only detected associations on these genes but also on two additional genes, i.e. Lr10 and Lr1. As revealed by a second hybrid wheat population for independent validation, using detected associations from SNP-based and FH-GWAS reached predictabilities of 11.72% and 22.86%, respectively. Therefore, FH-GWAS is not only more powerful to detect associations, but also improves the accuracy of marker-assisted selection as compared to the SNP-based approach.

Data from: Predicting range expansion of invasive species: pitfalls and best practices for obtaining biologically realistic projections

Species Distribution Models (SDM) for seven invasive plant species in North America. Species distribution models (SDMs) are widely used to forecast potential range expansion of invasive species. However, invasive species occurrence datasets often have spatial biases that may violate key SDM assumptions. We examined alternative methods of spatial bias correction and multiple methods for model evaluation for seven invasive plant species. Species SDMs include Common Tansy (Tanacetum vulgare), Wild Parsnip (Pastinaca sativa), Leafy Spurge (Euphorbia virgata), Common Teasel (Dipsacus fullonum), Brown Knapweed (Centaurea jacea), Black Swallowwort (Vincetoxicum nigrum), and Dalmatian Toadflax (Linaria dalmatica).

Using niche centrality within the scope of the nearly neutral theory of evolution to predict genetic diversity in a tropical conifer species-pair

Aim: Estimating genetic diversity is key for understanging biogeographic and evolutionary processes. However, gathering genetic information is not feasible for all taxa or populations, particularly in the tropical regions. Identifying proxies for inferring such values has thus become essential. Here, we built on the niche centrality hypothesis (NCH; or central-abundance hypothesis) and the nearly neutral theory of evolution (NNT) to identify some of such proxies using a montane tropical conifer species-pair as model. The NCH predicts more genetic diversity under optimal ecological conditions, which should also allow for more efficient purifying selection, according to the NNT. Location: The Transmexican Volcanic Belt, central Mexico. Taxa: A fir species-pair endemic to central Mexico, Abies flinckii and A. religiosa. Methods: We estimated patterns of genetic diversity from nuclear SSRs (A, HE), and gene-coding sequences (πS, πN), together with the efficacy of purifying selection, measured as πN/πS. After testing for niche overlap, we used several geographic and ecological proxies (i.e. longitude, latitude, elevation, estimated area, and distance to the niche centroid in the present and in the LGM) to predict genetic diversity and πN/πS using general linear models. Results: Populations at the west of the Trans Mexican Volcanic Belt (TVB) had lower genetic diversity than populations in the east of this mountain chain. Both species had significant niche overlap. The principal predictors for neutral genetic diversity (HE, A and πS) were longitude and latitude, followed by the current distance to the niche centroid; the efficiency of purifying selection was mostly accounted for by the current distance to the niche centroid (which was also correlated to elevation). No correlation was observed between genetic diversity or πN/πS and current population area. Main conclusions: Historical and ecological factors have to be taken into account for explaining the amounts of genetic diversity in mountain tropical species. Following the NTT, populations closer to the niche centroid are more efficient at eliminating slightly deleterious mutations than marginal stands, independently of their size or geographical location (longitude). Expanding the central-abundance theory within the scope of the NTT might help reconciling conflicting views concerning the extent of its empirical support.

opencc-by-nc-nd-4.0Aug 2021View details →

Predicting readers' prototypical eye-movement behavior using MASC, a model of Attention in the Superior Colliculus: Stimulus materials, model code, data, and statistical analyses.

The goal of the present research was to determine the role of rudimentary visuo-motor pathways, from the retina and the primary visual cortex to the superior colliculus (SC), in the guidance of human eye movement during reading. To this end, we used MASC, our model of Attention in the Superior Colliculus (Adeli et al., Journal of Neuroscience 2017), a model that relies on well-established saccade-programming principles in the SC. MASC predicts sequences of fixations over an input image by spatially integrating incoming signals in the space of the SC. Here, MASC computed the distribution of luminance contrast over sentences' images (visual-saliency map), after blurring it proportional to retinal eccentricity (retina transformation). It then projected the visual-saliency map into SC space, using a logarithmic afferent-mapping function (magnification factor). Input signals were averaged over retinotopically organized populations of neurons (point images) of constant size, first in the visual map and then in a spatially-registered motor map. The most active population was identified through a winner-take-all process. After jitter applied to the winning population, the next fixation location was determined using inverse efferent mapping. This sequence of events was then repeated to predict following fixation locations, but inserting after each saccade an inhibitory spatial tag (Inhibition of Saccade Return; ISR -referred to as IOR in the uploaded files). All MASC's parameters, but one, were biologically determined, using electrophysiological data in macaque; the ISR window was the one fit parameter. MASC was tested by comparing its predicted sequences of fixations over sentences from the French-Sentence Corpus (FSC) to the eye-movement behavior of 40 French-native speakers reading the same sentences for comprehension (Albrengues et al., Plos One 2019). Then, MASC was dissected to determine the crucial processing steps enabling prediction of human behavior (10 comparison models -see the general README file). Finally, to address crucial issues in the reading literature, i.e., the role of inter-word spacing and character print size in eye-movement guidance, MASC was additionally tested in four additional display conditions: the same sentences from the FSC, but with blank spaces between words being either filled or removed, or with the screen width angle being multiplied by 2 or 4, such that characters were larger in angular size (0.5° and 1°) than in the original experiment (0.25°). MASC's predicted effects of inter-word spacing and print size were compared to previously published data. All material relevant to the project is reported here, including the FSC materials (bitmap and information text files), the Matlab code for our MASC model, raw simulation data for MASC and all our comparison models, as well as MASC's simulations in the different display conditions, the scripts we developed in R to transform raw simulation data into data matrices for statistical analyses of (word-based) eye-movement behavior, the resulting data matrices for all models as well as the data matrix for FSC readers, the R-scripts for statistical comparison of oculomotor behavior between data sets and conditions, literature-review tables of previously published data (for comparison with MASC's predictions), and the R-scripts generating the figures summarizing our results. Further information can be found in the general README file as well as in the README files attached to each folder. The authors' respective contributions to the project, the licence attached to the included materials and their condition of use are listed in the general README file. A manuscript reporting and discussing these modeling data is in preparation (Vitu, F., Adeli, H. & Zelinsky, G. J.); A reference will be provided here when the manuscript appears in a journal. Other references to be cited: - For the model code: Adeli, H., Vitu, F., & Zelinsky, G. J. (2017). A model of the superior colliculus predicts fixation locations during scene viewing and visual search. Journal of Neuroscience, 37(6), 1453-1467. http://www.jneurosci.org/content/37/6/1453 - For FSC materials and data: Albrengues, C., Lavigne, F., Aguilar, C., Castet, E., & Vitu, F. (2019). Linguistic processes do not beat visuo-motor constraints, but they modulate where the eyes move regardless of word boundaries: Evidence against top-down word-based eye-movement control during reading. PLoS ONE 14(7): e0219666. https://doi.org/10.1371/journal.pone.0219666

opennotspecifiedDec 2016View details →

Figure 11. Predicted suitable habitat for the Eirenis persicus species group. A in Alpine-Himalayan orogeny drove correlated morphological, molecular, and ecological diversification in the Persian dwarf snake (Squamata: Serpentes: Eirenis persicus)

Figure 11. Predicted suitable habitat for the Eirenis persicus species group. A, western operational taxonomic unit (OTU) specimens; B, eastern and nigrofasciatus OTUs. The model was reclassified into ten equal probability classes. Only classes with probabilities greater than 60% are presented here. Maximum training sensitivity plus specificity logistic threshold (dark grey) is equal to 15.7% in (A) and 23% in (B). In (A), circles indicate the south-western Iran sub-OTU, ◆ indicate the south-eastern Turkey and western Iran sub-OTU, and plus symbols indicate the northern Iran specimens. In (B), circles indicate specimens of the nigrofasciatus OTU, triangles indicate the eastern Iran and Turkmenistan sub-OTU, plus symbols indicate the north-eastern Pakistan sub-OTU, and stars indicate the localities of the specimens referred to Eirenis mcmahoni.

opencc-by-4.0Sep 2021View details →

The predictive role of executive functions and psychological factors on chronic pain after orthopaedic surgery A longitudinal cohort study

This is a dataset from a longitudinal cohort study. Patients listed for orthopaedic surgery were enrolled. Variables measured before surgery were pain intensity, the sensory, affective, cognitive and mixed components of pain, state and trait variables associated with the psychological status of the patient, fear of movement, pain catastrophizing, visual attention and cognitive flexibility. Pain intensity and the components of pain were re-evaluated after surgery and after three months. 167 patients were enrolled.

opencc-by-4.0Sep 2021View details →

Subnetwork representation learning for discovering network biomarkers in predicting lymph node metastasis in early oral cancer

Dataset for the specified study.

Supplementary data: "Secondary control activation analysed and predicted with explainable AI"

This repository contains processed data and result files for the paper  <a href="https://arxiv.org/abs/2109.04802">Secondary control activation analysed and predicted with explainable AI</a> .   The code for producing the processed data and the results is <a href="https://github.com/johkruse/XAI-for-aFRR-activation">available at github</a>. Data The data folder contains the feature and target data used to train the ML model. The data for Germany comprises the following folders and files: <ul> <li>raw_input_data.h5 : The aggregated external features without additional engineered features.</li> <li>inputs_<model_type>.h5 : The input features for the different model types used in the paper including the engineered features. Depending on the model type, the input files also contain the IGCC features.</li> <li>outputs.h5 :  The activated aFRR volumes in Germany.</li> <li>version_2021-08-20: Folder containing the training and test sets used for the results.</li> <li>documentation_of_data_download: Information files concerning the ENTSO-E raw data and its aggregation.</li> </ul> In addition to the German time series, the data folder contains the raw input data for the remaining IGCC states. Note that the results contain more model types as actually discussed in the paper. Data sources The data for input features (raw_input_data.h5 and input_<model_type>.h5) is derived from ENTSO-E Transparency Platform data [1]. The target data (outputs.h5) is based on publicly available data from the German Transmission System Operators (TSOs) [2]. Results The result folder comprises the results of hyper-parameter optimization, model prediction and interpretation via SHAP. The model type, the loss function to train the model and the data set for prediction/interpretation were varied. <ul> <li>cv_results_<model_type>_<loss_function>.csv : Performance results for each combination in the hyper-parameter grid search.</li> <li>cv_best_params_<model_type>_<loss_function>.csv : Hyper-parameters used in the final (optimized) model.</li> <li>shap_values_<data_set>_<model_type>_<loss_function>.npy : First-order SHAP values calculated on different data sets: The train set, the randomized test set and the continuous test set.</li> <li>y_pred_<data_set>.h5 : Predictions of daily profile predictor and Machine Learning models.</li> </ul> Disclaimer The data might be subject to copyright or related rights. Please consult the primary data owner.

opencc-zeroDec 2021View details →

Lineage-level distribution models lead to more realistic climate change predictions for a threatened crayfish

Aim: As climate change presents a major threat to biodiversity in the next decades, it is critical to assess its impact on species habitat suitability to inform biodiversity conservation. Species distribution models (SDMs) are a widely used tool to assess climate change impacts on species' geographical distributions. As the term suggests, the species-level is the most commonly used taxonomic unit in SDMs. However, recently it has been demonstrated that SDMs considering taxonomic resolution below (or above) the species-level can make more reliable predictions of biodiversity change when different populations exhibit local adaptation. Here, we tested this idea using the Japanese crayfish (Cambaroides japonicus), a threatened species encompassing two geographically structured and phylogenetically distinct genetic lineages. Location: Northern Japan. Methods: We first estimated niche differentiation between the two lineages of C. japonicus using n-dimensional hypervolumes, then made climate change predictions of habitat suitability using SDMs constructed at two phylogenetic levels: species and intraspecific lineage. Results: Our results showed only intermediate niche overlap, demonstrating measurable niche differences between the two lineages. The species-level SDM made future predictions that predicted much broader and severe impacts of climate change. However, the lineage-level SDMs led to reduced climate change impacts overall, and also suggested that the eastern lineage may be more resilient to climate change than the western one. Main conclusions: The two lineages of C. japonicus occupy different niche spaces. Compared with lineage-level models, species-level models can overestimate climate change impacts. These results not only have important implications for designing future conservation strategies for this threatened species, but also highlight the need for incorporating genetic information into SDMs to obtain realistic predictions of biodiversity change.

opencc-by-4.0Jul 2021View details →

A unified DTI prediction framework based on knowledge graph and recommendation system

## A unified DTI prediction framework based on knowledge graph and recommendation system   # Code and data description ## Scripts - `kge_nfm.py`: the complement of the KGE_NFM & NFM methods. - `kge_rf.py`: the complement of the KGE_RF & RF methods. - `deepdit.py`: the complement of the MPNN_CNN & DeepDTI methods. - the complement of DTINet and DTiGEMS is tested based on their source packages (more in Prerequisites)   ## `data/` directory #### `yamanishi_08/` directory - `data_folds/`: 10 folds training set and test set in the three scenarios - `warm_start_1_1/` - `warm_start_1_10/` - `drug_coldstart/` - `protein_coldstart/` - `kg_data/`: supporting knowledge graph data - `dt_all_08.csv`: whole DTI dataset - `791drug_struc.csv`: drugbank id and smiles of drugs - `989proseq.csv`: kegg id and sequences of proteins - `morganfp.txt`: list of drug morgan fingerprints - `pro_ctd.txt`: list of protein descriptors   #### `BioKG/` directory - `data_folds/`: 10 folds training set and test set in the three scenarios - `warm_start_1_10/` - `drug_coldstart/` - `protein_coldstart/` - `kg.csv`: supporting knowledge graph data - `dti.csv`: whole DTI dataset - `comp_struc.csv`: drugbank id and smiles of drugs - `pro_seq.csv`: sequences of proteins - `fp_df.csv`: list of drug morgan fingerprints - `prodes_df.csv`: list of protein descriptors   #### `hetionet/` directory - `data_folds/`: 10 folds training set and test set in the three scenarios - `warm_start_1_10/` - `drug_coldstart/` - `protein_coldstart/` - `kg.csv`: supporting knowledge graph data - `dti.csv`: whole DTI dataset - `map_drugs_df`: drugbank id and smiles of drugs - `pro_seq.csv`: sequences of proteins - `fp_df.csv`: list of drug morgan fingerprints - `prodes_df.csv`: list of protein descriptors   #### `luo's_dataset/` directory - `data_folds/`: 10 folds training set and test set in the three scenarios - `warm_start_1_1/` - `warm_start_1_10/` - `drug_coldstart/` - `protein_coldstart/` - `mapping/`: related mappings and similarity matrix (https://github.com/luoyunan/DTINet) - `protein.txt`: list of protein names - `disease.txt`: list of disease names - `se.txt`: list of side effect names - `drug_dict_map`: a complete ID mapping between drug names and DrugBank ID - `protein_dict_map`: a complete ID mapping between protein names and UniProt ID - `Similarity_Matrix_Drugs.txt` : Drug similarity scores based on chemical structures of drugs - `Similarity_Matrix_Proteins.txt` : Protein similarity scores based on primary sequences of proteins - `feature/`: related features used in methods - `drug_smiles.csv`: drugbank id and smiles - `seq.txt`: list of protein sequences - `morganfp.txt`: list of drug morgan fingerprints - `pro_ctd.txt`: list of protein descriptors   #### `eg_model/` directory We provided a pre-trained kge model for example. - `dismult_400_warm_1_10.pkl`   # Prerequisites #### Operating system: Linux #### Programing language: python #### KGE_NFM & NFM dependencies ``` - python 3.6 - pandas '1.1.5' - numpy '1.18.4' - scikit-learn '0.24.1' - tensorflow '1.15.0' - ampligraph '1.3.2' - deepctr '0.8.4' ``` #### baseline dependencies - RF & KGE_RF (included in KGE_NFM&NFM dependencies) - MPNN_CNN & DeepDTI: - source: https://github.com/kexinhuang12345/DeepPurpose ``` - deeppurpose '0.0.9' - torch '1.6.0+cu101' ``` - DTINet: - source: https://github.com/luoyunan/DTINet - note: in this work, we run the DTINet in a python environment, which need Linux system and python2. Importantly, this method requires the [Inductive Matrix Completion](http://bigdata.ices.utexas.edu/software/inductive-matrix-completion/) (IMC) library. More detailed information about the installation of this method could be found in the source code of the DTINet. - DTiGEMS: - source: https://github.com/MahaThafar/DTiGEMSplus - TriModel: - source: http://drugtargets.insight-centre.org/   # Example (kge_nfm.py)   #### A brief presentation of the results: - return average loss when training kge model ``` Average Loss: 0.475181: 2%|###3 | 1/50 [01:10<57:31, 70.44s/epoch] ``` - return performance(mrr) on training set of DTI for early stopping (kge_model in `eg_model/`) ``` In [35]: roc = roc_auc(test_label,test_score) ...: pr = pr_auc(test_label,test_score) ...: print(roc) ...: print(pr) 0.8731770833333332 0.44079654835037246 ```   - nfm training process (`patience=10`)   ``` In [45]: roc_nfm,pr_nfm,pred_y = train_nfm(feature_columns,train_model_input,train_label,test_model_input,test_label,patience) Train on 44851 samples Epoch 1/2000 44851/44851 - 2s - loss: 0.5332 - precision: 0.0976 Epoch 2/2000 44851/44851 - 1s - loss: 0.4143 - precision: 0.0000e+00 Epoch 3/2000 44851/44851 - 1s - loss: 0.3456 - precision: 0.0000e+00 Epoch 4/2000 44851/44851 - 1s - loss: 0.3443 - precision: 0.0000e+00 Epoch 5/2000 44851/44851 - 1s - loss: 0.3470 - precision: 0.0000e+00 Epoch 6/2000 44851/44851 - 1s - loss: 0.3382 - precision: 0.0000e+00 ...... Epoch 279/2000 44851/44851 - 1s - loss: 0.0758 - precision: 0.9248 Epoch 280/2000 44851/44851 - 1s - loss: 0.0753 - precision: 0.9327 Epoch 281/2000 44851/44851 - 1s - loss: 0.0796 - precision: 0.9155 Epoch 282/2000 44851/44851 - 1s - loss: 0.0764 - precision: 0.9276 Epoch 283/2000 44851/44851 - 1s - loss: 0.0739 - precision: 0.9127 ```   - reutrn results as type of roc_auc & pr_auc ``` 0.9812476679104477 0.8803416284646345 ```

opencc-zeroSep 2021View details →

Comparison of atmospheric refractivity estimation methods and their influence on radar wave propagation predictions

Environmental predictions in the marine atmospheric surface layer (MASL) are imperative to optimize X-band radar system performance in marine environments. Evaporation ducts (ED) lead to anomalous propagation where characterization of EDs in the MASL occurs primarily through two methods: in-situ measurements and numerical modeling. This study investigates differences in co-located and synchronous refractivity estimations from the CASPER-East campaign. Propagation predictions are generated for refractive profiles from in-situ measurements, Monin-Obukov boundary layer similarity theory, and numerical weather prediction forecasts. Variations in evaporation duct height (EDH) are found to be a primary driver of differences in propagation between the estimated refractivity profiles, where location of the EDH relative to the transmitter changes the sensitivity of propagation predictions to EDH estimates. Differences in propagation are large when EDH estimates span the transmitter height and the lowest EDH across the methods is small, regardless of how much variation there is in EDH estimates. When the lowest EDH is small and EDH estimates span the transmitter height there are differences in physical regimes causing large propagation discrepancies – e.g., leakage into versus trapping within the duct. Variation in EDH between the methods is greatest in stable environments. M-deficit and curvature of the refractive profiles also influence propagation specifically in scenarios when EDH spans the transmitter. When all EDHs are below the transmitter, EDH variance is the primary contributor to propagation variance, but M-deficit and profile curvature variance play a secondary role. M-deficits and curvature between the methods agree most often during periods of atmospheric stability.

opencc-zeroSep 2021View details →

Data from: Multi-generation genomic prediction of maize yield using parametric and non-parametric sparse selection indices

Genomic prediction models are often calibrated using multi-generation data. Over time, as data accumulates, training data sets become increasingly heterogeneous. Differences in allele frequency and linkage disequilibrium patterns between the training and prediction genotypes may limit prediction accuracy. This leads to the question of whether all available data or a subset of it should be used to calibrate genomic prediction models. Previous research on training set optimization has focused on identifying a subset of the available data that is optimal for a given prediction set. However, this approach does not contemplate the possibility that different training sets may be optimal for different prediction genotypes. To address this problem, we recently introduced a sparse selection index (SSI) that identifies an optimal training set for each individual in a prediction set. Using additive genomic relationships, the SSI can provide increased accuracy relative to genomic-BLUP (GBLUP). Non-parametric genomic models using Gaussian kernels (KBLUP) have, in some cases, yielded higher prediction accuracies than standard additive models. Therefore, here we studied whether combining SSIs and kernel methods could further improve prediction accuracy when training genomic models using multi-generation data. Using four years of doubled haploid maize data from the International Maize and Wheat Improvement Center (CIMMYT), we found that when predicting grain yield the KBLUP outperformed the GBLUP, and that using SSI with additive relationships (GSSI) lead to 5-17% increases in accuracy, relative to the GBLUP. However, differences in prediction accuracy between the KBLUP and the kernel-based SSI were smaller and not always significant.

opencc-by-4.0Aug 2021View details →

The impact of the cross-docked poses on the performance of machine learning classifier for protein-ligand binding pose prediction

Datasets, features, and some representative scripts utilized in the paper "The impact of the cross-docked poses on the performance of machine learning classifier for protein-ligand binding pose prediction".

Figure 12 in Reconstruction of the adductor chamber and predicted bite force in pterodactyloids (Pterosauria)

Figure 12. Right mandibular ramus of Pteranodon sp. AMNH 153 in medial view, (A) whole view of posterior region of the ramus, (B) and (C) close-ups corresponding to the rectangles shown in (A). Scale bar = 10 mm. Arrowheads indicate muscle scars. Abbreviations: glfo, glenoid fossa; san, surangular; rapr, retroarticular process.

Figure 15 in Reconstruction of the adductor chamber and predicted bite force in pterodactyloids (Pterosauria)

Figure 15. Right mandibular ramus of Pteranodon sp. (A) YPM 1175, (B) YPM 1175 (detail of scars) and (C) YPM 2738 in medial view. Arrowheads indicate muscle scars. Scale bars = 10 mm. Abbreviations: adfo, adductor fossa; glfo, glenoid fossa; rapr, retroarticular process.

Figure 18 in Reconstruction of the adductor chamber and predicted bite force in pterodactyloids (Pterosauria)

Figure 18. Mandibular rami of Pteranodon sp. in lateral view. A, YPM 2487. B, YPM 2476. C, YPM 1175. D, YPM 2738. Arrowheads indicate muscle scars. Scale bars = 10 mm. Abbreviations: glfo, glenoid fossa; rapr, retroarticular process.

Figure 8 in Reconstruction of the adductor chamber and predicted bite force in pterodactyloids (Pterosauria)

Figure 8. Palatal bones in occlusal view of Caupedactylus ybaka holotype (MN 4726-V). Scale bar = 10 mm. Arrowhead indicates pterygoid fossa. Abbreviations: ec, ectopterygoid; m, maxilla; pplf, postapalatinal fenestra; pty, pterygoid; stf, subtemporal fenestra.

Figure 4 in Reconstruction of the adductor chamber and predicted bite force in pterodactyloids (Pterosauria)

Figure 4. Upper temporal regions of (A) Tropeognathus mesembrinus holotype (BSPG 1987 46) in left lateral view, (B) Anhanguera blittersdorffi (MN 4805-V) in right lateral view, (C) Pteranodon longiceps (YPM 2594) in left lateral view, and (D) Ludodactylus sibbicki (SMNK PAL 3828) in right lateral view. The arrowheads mark the adductor ridges. Scale bars = 20 mm. Abbreviations: f, frontal; fpcr, frontoparietal crest; ltf, lateral temporal fenestra; op, opisthotic; or, orbit; r, right; sq, squamosal; utf, upper temporal fenestra.