In sp2019-antibiotics/ECOFFBayes: Bayesian Finite Mixture BNN.. Model for EUCAST Zone Data

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

This documents is an introduction to the package ECOFFBayes. The aim of this package is to find cutoff points between resistant and non-resistant antibiotics.

Therefore, first load the package ECOFFBayes.

library(ECOFFBayes)

Simulated Data

The package provides simulated data to try out the different functions.

data("Antibiotics")

If you do not have any prior knowledge just apply this function to your data and it will come up with the number of components and the estimated parameter.

bayes.density <- TheBayesteApproach(Antibiotics)

If you have prior knowledge for your data, you can include this either through some prior parameter and/or through fixing the number of components k.

In this case, both information are included.

prior <- list(a.prior = 1, 
              b.prior = 1, 
              e0.prior = 1, 
              b0.prior = 5,
              B0.prior = 0.01, 
              c0.prior = 5, 
              C0.prior = 10)
k <- 3
bayes.density.with.prior <- TheBayesteApproach(data = Antibiotics,
                                               prior = prior, 
                                               k = k)

The output is of type TheBayesteApproach and includes the following components:

• number of components k

bayes.density$k
bayes.density.with.prior$k

• parameter of the normal components

bayes.density$normal.params
bayes.density.with.prior$normal.params

• parameter of binomial component

bayes.density$binom.res
bayes.density.with.prior$binom.res

As we are also interested in the cutoff point between wildtype and non-wildtype bacteria (ECOFF), we can also calculate this value. Therefore we

FindECOFF(bayes.density, 
          pi.level = 0.3, 
          quantile = 0.01)
FindECOFF(bayes.density.with.prior, 
          pi.level = 0.3, 
          quantile = 0.01)

For this object, also a plotting function is provided, which plots the histogram of the data and the estimated density.

A line that shows the ECOFF value can also be included in the graph. Therefore the parameter ECOFF, which is by default FALSE, needs to be set to TRUE. Also here the quantile and pi.levelparameter can be changed.

plot(bayes.density,
     ECOFF = TRUE,
     quantile = 0.01,
     pi.level = 0.3)

plot(bayes.density.with.prior,
     ECOFF = TRUE,
     quantile = 0.01,
     pi.level = 0.3)

Data: EUCASTData

Get the data of the zone diameter from the external package EUCASTData.

data("ZD", package = "EUCASTData")

The code for three examples is included. Due to the required run-times for larger samples only the results for the second example are included, while the other examples are not evaluated.

Example 1:

For one example dataset, first only selected the antimicrobial and bacteria you are interested in. Afterwards, you can run you analysis using the following functions:

data1 <- subset(ZD, 
                Antimicrobial == "Ampicillin" & 
                  Bacterium == "Escherichia coli", 
                grepl("^Z", colnames(ZD)))
res1 <- TheBayesteApproach(data = data1)
FindECOFF(res1
          pi.level = 0.3, 
          quantile = 0.01)
plot(res1,
     ECOFF = TRUE,
     quantile = 0.01,
     pi.level = 0.3)

Example 2:

data2 <- subset(ZD, 
                Antimicrobial == "Piperacillin" & 
                  Bacterium == "Escherichia coli", 
                grepl("^Z", colnames(ZD)))
res2 <- TheBayesteApproach(data = data2)
```r
FindECOFF(res2, 
          pi.level = 0.3, 
          quantile = 0.01)
```r
plot(res2,
     ECOFF = TRUE,
     quantile = 0.01,
     pi.level = 0.3)

Example 3:

data3 <- subset(ZD, 
                Antimicrobial == "Mecillinam" & 
                  Bacterium == "Escherichia coli", 
                grepl("^Z", colnames(ZD)))
res3 <- TheBayesteApproach(data = data3)
FindECOFF(res3, 
          pi.level = 0.3, 
          quantile = 0.01)
plot(res3,
     ECOFF = TRUE,
     quantile = 0.01,
     pi.level = 0.3)

sp2019-antibiotics/ECOFFBayes documentation built on Aug. 28, 2019, 6:06 p.m.