Welcome to the main repository to the BayClump Shiny Dashboard.
C. Román-Palacios, H. M. Carroll, A. J. Arnold, R. J. Flores, Q. Guan, S. Petersen, K. A. McKinnon, and A. Tripati. BayClump: Bayesian Calibration Models of the ‘Clumped Isotopes’ Paleothermometer. Under review in Geochemistry, Geophysics, Geosystems. Preprint.
Developers
Hannah M. Carroll hcarroll@epss.ucla.edu
Cristian Román-Palacios cromanpa@g.ucla.edu
Senior author
Aradhna Tripati atripati@g.ucla.edu
GitHub repo
https://github.com/Tripati-Lab/BayClump
Bug reports and issues
https://github.com/Tripati-Lab/BayClump/issues
BayClump version 0.9 was developed in R version 4.1.0 and RStudio version 1.4.1106. This app accompanies Román-Palacios et al. (xxxx): BayClump: Bayesian Calibration and Temperature Reconstructions for Clumped Isotope Thermometry
BayClump is separated into calibration and reconstruction workflows. Any calibration models selected on the Calibrations tab are automatically made available for use in the Reconstructions tab. At this time, only
Two calibration datasets are included in BayClump by default. Model 1 is based on the reprocessed datasets of Petersen et al. (2019). Model 2 is based on the reprocessed and original datasets of Anderson et al. (2021). The datasets from Petersen et al. (2019) have been reprojected into the I-CDES reference frame after Bernasconi et al. (2021) at a growth temperature of 90°C, so as to be compatible with the datasets of Anderson et al. (2021). Throughout, we refer to this as
The default calibration datasets may be used individually or in combination to create calibration models, and may also be combined with the user's own calibration data in the I-CDES 90 reference frame. The user may also choose to work exclusively with their own calibration data in
BayClump provides a template for users to upload calibration data. It can be downloaded from the Calibrations tab. Sample data are shown in the first ten rows of the template and should be overwritten. The template must be left in .csv format, although the user is free to change the name of the spreadsheet as needed. Do not change column headings or orders. Do not enter extra columns - these will not be used by the app and may create problems.
Sample Name: This should be a unique identifier for each sample. Combinations of letters, numbers, spaces, and special characters may be used. It is best to avoid number signs (#
) as this may cause the code behind BayClump to malfunction.
Mineralogy: This should be the full, capitalized name, i.e., Aragonite, Calcite, Dolomite, Mixed, Unknown, etc.
Material: This must be entered as a number, where 1 = Natural, 2 = Synthetic, and 3 = Biogenic. Additional numbers may be used as necessary. All preloaded data in the app follow this convention.
N: The number of replicates measured to produce the final value. This must be numeric - no letters or special characters are accepted.
D47: The final
D47error: The absolute error around
Temperature: Growth temperature in
TempError: Absolute error of the growth temperature in
Note: You must provide EITHER the number of replicates OR
If calibration data values are produced from a very wide range of growth temperatures, it may be necessary to scale your data prior to analyses. Select this checkbox to automatically scale your data. We call the generic function 'scale' included in base R. Data are scaled by dividing the raw values by their root mean square.
BayClump offers five calibration model options. Bootstrapping is used to produce robust estimates of regression coefficients. The user is able to select 50, 100, 500, or 1,000 bootstrap replicates. We recommend that the number of bootstrap replicates chosen be at least 10 times the number of unique samples, with a minimum of 50 and a maximum of 1,000 offered. For example, if you have ten unique samples you wish to use in calibration models, you should choose at least 100 bootstrap replicates (10 x 10). If you have 3 or fewer samples, do not exceed 50 bootstrap replicates.
Warning: Selecting all available calibration models and a large number of bootstraps will require significant processing time. This may exceed 30-40 minutes if server traffic is heavy. The app may appear to hang while calculations are being performed, but should complete normally.
Important: Calibration data should be entered into the template in
Note: The descriptions provided below are taken from the publication this app accompanies. For additional information, please see Román-Palacios et al. (xxxx).
This regression model is the simplest model used in this study. The simple linear regression does not explicitly account for measurement error in
Simple linear regression fit under least squares, with observations being weighted based on the inverse of their squared uncertainty in the measured
The Deming model also fits under least squares, and the optimization steps while fitting the straight line follow the algorithm of Deming (1943). This model explicitly accounts for measurement error in both
The York model is closely related to the Deming regression model. However, under the York model, the ratio of the weights (related to their uncertainty) in
This model is the simplest Bayesian model currently implemented in the app. This Bayesian model is equivalent to the simple linear regression model (Model 1). However, instead of parameter estimates based on least squares, optimization of regression parameters is conducted under a Bayesian framework.
This model explicitly accounts for measurement error in both
In some situations, it is desirable to account for potential differences in the relationship between
All plots are created in plotly version 4.9.3 (Sievert 2020). They are fully interactive and downloadable. Plots download at web viewing resolution (72 dpi), 1244x400 pixels, and in .png format by default, but may be scaled and resized in external software if desired.
![Default (Model 1) calibration data plot](caldatexample.png)The Calibration Plots tab will display raw calibration data based on user input. The default display is of calibration data for Model 1. It will automatically update to reflect the choice of calibration data made on the Calibrations tab.
Once calibration model runs are complete, the Calibration Plots tab will update to show plots of raw data on the
BayClump provides a template for users to upload reconstruction data. It can be downloaded from the Reconstructions tab. Sample data are shown in the first ten rows of the template and should be overwritten. The template must be left in .csv format, although the user is free to change the name of the spreadsheet as needed. Do not change column headings or orders. Do not enter extra columns - these will not be used by the app and may create problems.
Models selected in the Calibrations tab are automatically transferred to the Reconstructions Tab. You are asked to confirm that your calibration and reconstruction data are in the same reference frame before the reconstructions will be performed.
Temperature predictions for specific
Temperature predictions are explicitly based on parameter estimates for Bayesian and non-Bayesian models. Parameter estimate and uncertainties follow the median and 95% CI across a certain number of replicates. Therefore, non-Bayesian predictions do not simultaneously propagate uncertainties in parameter estimates and target
By default, up to six rows of data for each reconstruction model are displayed in the output. This may be copied and pasted directly into external software if desired. The download button will provide a multi-tabbed spreadsheet in .xlsx format containing the full output from all selected models.