Vignette: MOSAIC_bioacc prediction tool (Delta version)

Introduction

The purpose of this document is to introduce you how to use the prediction tool of MOSAIC application. This application is based on the R software¹ and especially the rbioacc library (version 0.0.5), to predict internal concentration of a hemical in organism over time by toxicokinetic (TK) models under a Bayesian framework. MOSAIC is developed as an R-Shiny interface (version 1.6.0)².
If you want to be kept informed, please email us: sandrine.charles@univ-lyon1.fr.

Short presentation

The prediction tool is presented as three different tabs:

• Model definition
• Prediction
• Validation

Model definition is the section where the user can indicate all the inputs required to performed predictions and/or validations.

Prediction is the section where the user can view the output of predictions and download the results.

Validation is the section where the user can upload data for validation and to view the output of validation, as well as download the results.

1 Model Definition

Several cases can be encounter using the MOSAIC prediction tool:

• case 1: use example files;
• case 2: use distributed parameters from previous fit directly done within MOSAIC;
• case 3: use distributed parameters from older version MOSAIC application;
• case 4: use no distributed parameters with single values

All panels for each scenario are illustrated in Fig. 1.1 to 1.5.

1.1 Case 1: example file

Figure 1.1: Case 1: with an example file selected (directly available in MOSAIC_bioacc

Loading an example file will automatically fill in the field corresponding to the selected data. The user only need to click on the ‘calculate and display’ button to obtain the corresponding predictions (Fig. 1.2.A).

When using example files, the user has also the possibility to change the exposure concentration, as illustrated in Fig. 1.2.B.

Figure 1.2: Internal concentration predictions for an exposure at (A) 0.0004 and (B) 0.1 µg.mL^{-1}.

1.2 Case 2: distributed parameters from previous fit

Figure 1.3: Case 2: with distributed parameters from previous fit directly done within MOSAIC_bioacc

Loading a previous fit will automatically fill in the field corresponding to the previous data loaded in the application. The user only need to click on the ‘calculate and display’ button to obtain the corresponding predictions.

When using a previous fit, the user has also the possibility to change the exposure concentration.

1.3 Case 3: distributed parameters from an older version

Figure 1.4: Case 3: with distributed parameters from an older version MOSAIC_bioacc} application.

Once selecting “yes” on the right panel about distributed parameters, the user can upload its own marginal posterior distribution for each TK parameter by click on “Browse.” The MOSAIC prediction tool expects to receive data as a .txt file or a .csv file (comma, semicolon or tabular separator). Each line of the table corresponds to a parameters combination estimate. The table must contains at least two columns (at least one deterministic and one stochastic parameters), with exact header names (Table ??):

• deterministic parameter: “kuw,” “kus,”“kuf,”“kupw,”“kee,”“keg,”“kml” (with l = 1…L), “keml” (with l = 1…L);
• stochastic: “sigmaConc,”“sigmaCmetl” (with l = 1…L), “sigmaGrowth.”

According to your data, further columns can be added in the file:

Table 1.1: Example of a marginal posterior distribution data set ready to be uploaded.

kee	kuw	sigmaConc
0.0407066	653.2712	0.0140530
0.0371152	645.9469	0.0144206
0.0396529	616.6217	0.0143693
0.0379054	611.8630	0.0152456
0.0366674	639.0979	0.0138721
0.0382296	641.9268	0.0162840

Then, do not forget to click on the ‘refresh’ button to load correctly the data in selecting the good separator. Then fill in the fields corresponding to exposure and time before to start calculations.

1.4 Case 4: no distributed parameters

Figure 1.5: Case 4: with no distributed parameters.

The user has to manually check the TK parameter to build the corresponding TK model.

For the accumulation phase, four exposure routes can be considered:

• \(k_{u_w}\), water exposure;
  
• \(k_{u_s}\), sediment exposure;
  
• \(k_{u_f}\), food exposure;
  
• \(k_{u_{pw}}\), pore water exposure.
  

For the depuration phase, three elimination processes can be considered:

• \(k_{e_e}\), excretion;
  
• \(k_{e_g}\), growth dilution;
  
• \(k_{m_\ell}\), biotransformation, where \(\ell\) is the metabolite number.
  

Then the user needs to fill in the field of the duration of the accumulation phase, the appropriate time unit and the associate exposure concentration(s).

2 Predictions

Similar outputs are provided by the prediction tool whatever the case considered, which are the plot(s) of internal concentrations predictions against time (Fig. 2.1).

Figure 2.1: Example of output delivered by the prediction tool.

In this tab, the user has the possibility to add dashed green lines for each time point indicated in the corresponding field separated by a semi-colon (e.g.: 0; 1; 4; 5; 7 days), and for a threshold value of an internal concentration (e.g., 0.2 \(\mu g.g^{-1}\)), as illustrated in Fig. 2.2.

Figure 2.2: Example of output delivered by the prediction tool adding time points and threshold value for internal concentration.

3 Validation

In the tab ‘Validation,’ the user can upload a data frame which contains at least two columns, one for the time points (named ‘time’) and one for internal concentration (named ‘conc’ and ‘concml’ for metabolite l, from 1 … to L). This tab can be useful when the user has data for several exposure concentrations, but he wants to test other for which he has no data. Thus the user can first validated his data before to predict for an other exposure concentration. For example, a calibration step (i.e., parameters estimation) is performed with MOSAIC_bioacc for a given concentration. Then the prediction analysis is done for an other exposure concentration for which the user has data. Then for the validation process, the corresponding experimental data for this predicted exposure profile are plotted over predictions. This concept is in full compliance with the recent EFSA scientific opinion.³

Figure 3.1: Example of output delivered by the validation tool adding data uploaded over predictions.

4 Downloads

4.1 Plots

You can download all plots as displayed by the prediction tool in several formats (.png, .jpg, .pdf, .svg, .tiff and .eps).

4.2 Tables

You can also download table results in .txt or .csv.

References

(1) R Core Team. (2020) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria.

(2) Chang, W., Cheng, J., Allaire, J. J., Xie, Y., and McPherson, J. (2020) Shiny: Web Application Framework for R.

(3) Ockleford, C., Adriaanse, P., Berny, P., Brock, T., Duquesne, S., Grilli, S., Hernandez-Jerez, A. F., Bennekou, S. H., Klein, M., Kuhl, T., Laskowski, R., Machera, K., Pelkonen, O., Pieper, S., Smith, R. H., Stemmer, M., Sundh, I., Tiktak, A., Topping, C. J., Wolterink, G., Cedergreen, N., Charles, S., Focks, A., Reed, M., Arena, M., Ippolito, A., Byers, H., and Teodorovic, I. (2018) Scientific Opinion on the state of the art of Toxicokinetic/Toxicodynamic (TKTD) effect models for regulatory risk assessment of pesticides for aquatic organisms.