benjaminhlina / ecotox

Analysis of Ecotoxicology

ecotox

Overview

{ecotox} was created as a simple approach to using either a probit or logit analysis to calculate lethal concentration (LC) or time (LT) and the appropriate fiducial confidence limits desired for the selected LC or LT for ecotoxicology studies (Finney 1971; Wheeler et al. 2006; Robertson et al. 2007). The simplicity of {ecotox} comes from the syntax it implies within its functions which are similar to functions like glm() and lm(). In addition to the simplicity of the syntax, a comprehensive tibble is produced which gives the user a predicted LC or LT value for the desired level and a suite of parameters such as fiducial confidence limits, z-value, and slope. {ecotox} was built for and is published in Hlina et al. 2021.

Installation

You can install the CRAN released version of {ecotox} from CRAN with:

install.packages("ecotox")

You can install the developer version of {ecotox} from github with:

install.packages("devtools")
devtools::install_github("benjaminhlina/ecotox")

Example

This is an example which uses the LC_probit function to calculate a LC₅₀ and LC₉₉ for a probit analysis:

## Calculate LC50 and LC99
head(lamprey_tox)

## within the dataframe used, control dose, unless produced a value
## during experimentation, are removed from the dataframe,
## as glm cannot handle values of infinite. Other statistical programs
## make note of the control dose but do not include within analysis. 

## calculate LC50 and LC99 for May

m <- LC_probit((response / total) ~ log10(dose),
               p = c(50, 99),
               weights = total,
               data = lamprey_tox[lamprey_tox$nominal_dose != 0, ],
               subset = c(month == "May"))

## view calculated LC50 and LC99 for seasonal toxicity of a pisicide,
## 3-trifluoromethyl-4-nitrophenol (TFM) to lamprey in 2011

m

## several new features include 1) being able to change the output length 
## 2) you can indicate whether the x variable has been log10 transformed or 
## not if it has the output will take that into consideration

m_2 <- LC_probit((response / total) ~ dose,
                 p = c(50, 99),
                 weights = total,
                 data = lamprey_tox,
                 subset = c(month == "May"), 
                 log_x = FALSE, 
                 long_output = FALSE)
                  
## view calculated LC50 and LC99 for seasonal toxicity of a pisicide,
## 3-trifluoromethyl-4-nitrophenol (TFM) to lamprey in 2011.

m_2

See StackExchange post about differences in using cbind() vs. response / total cbind() function in R for a logistic regression.

## Additionally changes have been made to allow for the user 
## to use `cbind()` method when specificying the response variable  

m_3 <- LC_probit(cbind(response, survive) ~ log10(dose),
                 p = c(50, 99),
                 data = lamprey_tox[lamprey_tox$nominal_dose != 0, ],
                 subset = c(month == "May"))
                  

m_3 

# notice that m and m_3 produce the same results, however m_3 will produce 
# a warning to ensure you have not weighted the model as it is not necessary 

Example of using ratio_test from Wheeler et al. 2006 to determine differences in LC values:

## A new function `ratio_test` has been added 

# view lamprey_tox data

head(lamprey_tox)

# using glm() to detemine LC values using probit model for May and June

m <- glm((response / total) ~ log10(dose),
         data = lamprey_tox[lamprey_tox$nominal_dose != 0, ],
         subset = c(month == "May"),
         weights = total,
         family = binomial(link = "probit"))


j <- glm((response / total) ~ log10(dose),
         data = lamprey_tox[lamprey_tox$nominal_dose != 0, ],
         subset = c(month == "June"),
         weights = total,
         family = binomial(link = "probit"))

# now that both May and June models have been made. use ratio_test to
# compare LC50 values or whatever LC values of interest.

ratios <- ratio_test(model_1 = m, model_2 = j, 
                     percentage = 50, 
                     compare = "May - June")

# view ratio test results

ratios

# you can also use LC_* or LT_* objects to create the models and use ratio test:

m_1 <- LC_probit((response / total) ~ log10(dose), p = c(50, 99),
                 weights = total,
                 data = lamprey_tox[lamprey_tox$nominal_dose != 0, ],
                 subset = c(month == "May"))



j_1 <- LC_probit((response / total) ~ log10(dose), p = c(50, 99),
                 weights = total,
                 data = lamprey_tox[lamprey_tox$nominal_dose != 0, ],
                 subset = c(month == "June"))



ratios_2 <- ratio_test(model_1 = m_1, model_2 = j_1, 
                       percentage = 50,
                       compare = "May - June", 
                       obj_type = "df")

ratios_2

References

• Finney, D.J., 1971. Probit analysis. Cambridge University Press, Cambridge, England. ISBN: 052108041X

• Wheeler, M.W., Park, R.M., and Bailey, A.J. 2006. Comparing median lethal concentration values using confidence interval overlap or ratio tests. Environ. Toxic. Chem. 25(5), 1441-1444. 10.1897/05-320R.1

• Robertson, J.L., Savin, N.E., Russell, R.M. and Preisler, H.K., 2007. Bioassays with arthropods. CRC press. ISBN: 0849323312

• Version 1.4.4 written by Benjamin L. Hlina, Carleton University, Ottawa, Ontario, Canada. Written in 'Programming Language R', R version 4.1.1 (2021-08-10) -- "Kick Things". Source code is available at {ecotox} or by contacting Benjamin L. Hlina at [email protected]

ecotox's Issues

SE calculation

Add in the ability for it LCx and LTx to calculate asymptomatic SE, similar to what predict() calculates.

Using output dataframe from LCx or LTx for ratio_test

Hi, is it possible to have an option to pipe that output directly into ratio_test instead of calling the glm model again.

Different Confidence Limit Methods

Add in the ability to change the method for calculating confidence limits and add in new methods e.g. delta method.

Add in feature similar to cf argument in dose.p()

cf argument in dose.p allows the user to select the slope and intercepts from the coefficient outputs. Currently this is quite hard coded into both LCx and LTx. If this could be adjusted more complex models could be created, e.g. synergistic dose response curves, where two compounds are used.

ratio_test() fails to properly calculate test statistic and p value when log_x = FALSE

When ratio_test()'s argument is set to log_x = FALSE, it returns incorrect p_value of zero. The mistake comes from when log_x = TRUE based on Wheeler et al. 2006, when subtracting the doses from each other you do not take the log of the doses as they have been already transformed. This was missed when creating the argument log_x.

Suggested Features

Suggested Features

• ecotox has now been up in operation for about 3 months. I've made some recent improvements, e.g. subset feature, and am still working on a spearman_karber function which I hope to implement within the next month or so. I have three other functions I want to implement:

• 1. Ratio comparison test for comparing LCs and LTs following Wheeler et al. 2006
• 2. A parallelism test following Robertson et al. 2007
• 3. A equality test following Robertson et al. 2007.

What features or functions do people want to see? What do people think of current functions and those that are currently under development?