Worked examples II

library(anovir)

Extending nll_proportional_virulence to multiple treatments

This example shows nll_proportional_virulence extended to comparing multiple treatments against a reference treatment for virulence.

Data from [1,2]


data01 <- data_lorenz

# create column 'ref_vir' in dataframe coded 0, 1 and 2
# for control, reference dose, and other dose treatments, respectively

data01$ref_vir <- ifelse(data01$g == 0, 0,
  ifelse(data01$g == 1, ifelse(data01$Infectious.dose == 5000, 1, 2), 0)
  )

# copy and modify 'nll_proportional_virulence' function
# where the virulence in the reference treatment is 'theta'
# which is multiplied by 'th10', 'th20', etc... in the other dose treatements

nll_proportional_virulence2 <- nll_proportional_virulence
  body(nll_proportional_virulence2)[[6]] <- substitute(pfth <- theta *
  ifelse(data01$Infectious.dose == 10000, th10,
  ifelse(data01$Infectious.dose == 20000, th20,
  ifelse(data01$Infectious.dose == 40000, th40,
  ifelse(data01$Infectious.dose == 80000, th80,
  ifelse(data01$Infectious.dose == 160000, th160,
  exp(0)
  )))))
  )

# update formals 

formals(nll_proportional_virulence2) <- alist(
  a1 = a1,  b1 = b1,  a2 = a2,  b2 = b2,
  theta = theta,
  th10 = th10, th20 = th20, th40 = th40, th80 = th80, th160 = th160,
  data = data01,
  time = t,
  censor = censored,
  infected_treatment = g,
  reference_virulence = ref_vir,
  d1 = 'Gumbel', d2 = 'Weibull')

# write 'prep function'

m01_prep_function <- function(
  a1, b1, a2, b2, theta, th10, th20, th40, th80, th160){
  nll_proportional_virulence2(
  a1, b1, a2, b2, theta, th10, th20, th40, th80, th160)
  }

# send 'prep function' to mle2
# NB fixing 'theta = 1' scales virulence in other treatments relative to
# that in the reference treatment

m01 <- mle2(m01_prep_function,
  start = list(
    a1 = 24, b1 = 5, a2 = 4, b2 = 0.2,
    theta = 1,
    th10 = 1, th20 = 1, th40 = 1, th80 = 1, th160 = 1),
    fixed = list(theta = 1)
    )

summary(m01)
#> Maximum likelihood estimation
#> 
#> Call:
#> mle2(minuslogl = m01_prep_function, start = list(a1 = 24, b1 = 5, 
#>     a2 = 4, b2 = 0.2, theta = 1, th10 = 1, th20 = 1, th40 = 1, 
#>     th80 = 1, th160 = 1), fixed = list(theta = 1))
#> 
#> Coefficients:
#>        Estimate Std. Error z value   Pr(z)    
#> a1    23.165445   0.652943 35.4785 < 2e-16 ***
#> b1     4.660055   0.384772 12.1112 < 2e-16 ***
#> a2     3.205421   0.084396 37.9807 < 2e-16 ***
#> b2     0.188362   0.020123  9.3607 < 2e-16 ***
#> th10   2.280418   1.142473  1.9960 0.04593 *  
#> th20   2.303080   1.144282  2.0127 0.04415 *  
#> th40   3.649006   1.798433  2.0290 0.04246 *  
#> th80   4.590818   2.277040  2.0161 0.04379 *  
#> th160  5.513346   2.706341  2.0372 0.04163 *  
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#> 
#> -2 log L: 1504.575

The values of a2 and b2 define virulence at time t in the reference dose treatment of 5 (x103) spores larva-1 as,

\[\begin{equation} h_{5000}(t) = \frac{1}{0.188 t} \exp \left[ \frac{\log t - 3.205}{0.188} \right] \end{equation}\]

which was estimated as being multiplied by 2.280, 2.303, 3.649, 4.591 and 5.513 times in the dose treatments of 10, 20, 40, 80 and 160 (x103) spores larva-1, respectively, assuming virulence is proportional among dose treatments.

Modifying nll_basic_logscale

This example shows how to manipulate the function nll_basic_logscale, which assumes location and scale parameters are on a logscale.

This function can avoid the generation of warning messages associated with parameters with negative values being given to logarithmic functions.

Data from [1,2]

data01 <- data_lorenz

head(data01)
#>    Infectious.dose Food Sex Spore.Count    t censored d g
#> 1            10000  100   F      425000 13.0        0 1 1
#> 2            10000   50   F       22000  3.5        0 1 1
#> 3            10000  100   F      465000 20.5        0 1 1
#> 8                0   50   F          NA 17.0        0 1 0
#> 10          160000   50   F      295000 17.5        0 1 1
#> 11          160000  100   F           0  4.0        0 1 1


nll_basic_logscale2 <- nll_basic_logscale 

body(nll_basic_logscale2)[[4]]
#> pfa2 <- exp(a2)

Here the location parameter a2 is to be made a linear function of log(dose),

a2 = a2i - a2ii*log(data01$Infectious.dose)

where a2i and a2ii are parameters to be estimated.

However instead of directly replacing a2 with the right hand side of the expression above, i.e.,

pfa2 <- exp(a2i - a2ii*log(data01$Infectious.dose))

both parameters need exponentiating, i.e.,

pfa2 <- exp(a2i) - exp(a2ii)*log(data01$Infectious.dose)


body(nll_basic_logscale2)[[4]] <- substitute(
  pfa2 <- exp(a2i) - exp(a2ii)*log(data01$Infectious.dose)
  )

body(nll_basic_logscale2)[[4]]
#> pfa2 <- exp(a2i) - exp(a2ii) * log(data01$Infectious.dose)

formals(nll_basic_logscale2) <- alist(
  a1 = a1, b1 = b1, a2i = a2i, a2ii = a2ii, b2 = b2,
  data = data01,
  time = t,
  censor = censored,
  infected_treatment = g,
  d1 = 'Gumbel', d2 = 'Weibull'
)

m01_prep_function <- function(a1, b1, a2i, a2ii, b2){
  nll_basic_logscale2(a1, b1, a2i, a2ii, b2)
  }

m01 <- mle2(m01_prep_function,
  start = list(a1 = 3, b1 = 1.5, a2i = 1.4, a2ii = -3, b2 = -1.5)
  )

summary(m01)
#> Maximum likelihood estimation
#> 
#> Call:
#> mle2(minuslogl = m01_prep_function, start = list(a1 = 3, b1 = 1.5, 
#>     a2i = 1.4, a2ii = -3, b2 = -1.5))
#> 
#> Coefficients:
#>       Estimate Std. Error z value     Pr(z)    
#> a1    3.143306   0.028334 110.937 < 2.2e-16 ***
#> b1    1.550561   0.081129  19.112 < 2.2e-16 ***
#> a2i   1.346133   0.049796  27.033 < 2.2e-16 ***
#> a2ii -2.511533   0.217948 -11.524 < 2.2e-16 ***
#> b2   -1.684226   0.104964 -16.046 < 2.2e-16 ***
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#> 
#> -2 log L: 1506.184

exp(coef(m01))
#>          a1          b1         a2i        a2ii          b2 
#> 23.18036585  4.71411429  3.84253844  0.08114371  0.18558808

Estimates made using nll_basic are similar, but generate Warning messages


nll_basic2 <- nll_basic 

body(nll_basic2)[[4]]
#> pfa2 <- a2

body(nll_basic2)[[4]] <- substitute(
  pfa2 <- a2i + a2ii*log(data01$Infectious.dose)
  )

body(nll_basic2)[[4]]
#> pfa2 <- a2i + a2ii * log(data01$Infectious.dose)

formals(nll_basic2) <- alist(
  a1 = a1, b1 = b1, a2i = a2i, a2ii = a2ii, b2 = b2,
  data = data01,
  time = t,
  censor = censored,
  infected_treatment = g,
  d1 = 'Gumbel', d2 = 'Weibull'
)


m02_prep_function <- function(a1, b1, a2i, a2ii, b2){
  nll_basic2(a1, b1, a2i, a2ii, b2)
  }

m02 <- mle2(m02_prep_function,
  start = list(a1 = 23, b1 = 4.5, a2i = 3.8, a2ii = -0.1, b2 = 0.4)
  )
#> Warning in log(h1 + g * h2): NaNs produced

#> Warning in log(h1 + g * h2): NaNs produced

#> Warning in log(h1 + g * h2): NaNs produced

#> Warning in log(h1 + g * h2): NaNs produced

#> Warning in log(h1 + g * h2): NaNs produced

#> Warning in log(h1 + g * h2): NaNs produced

#> Warning in log(h1 + g * h2): NaNs produced

#> Warning in log(h1 + g * h2): NaNs produced

#> Warning in log(h1 + g * h2): NaNs produced

#> Warning in log(h1 + g * h2): NaNs produced

#> Warning in log(h1 + g * h2): NaNs produced

summary(m02)
#> Maximum likelihood estimation
#> 
#> Call:
#> mle2(minuslogl = m02_prep_function, start = list(a1 = 23, b1 = 4.5, 
#>     a2i = 3.8, a2ii = -0.1, b2 = 0.4))
#> 
#> Coefficients:
#>       Estimate Std. Error z value     Pr(z)    
#> a1   23.189717   0.657339 35.2782 < 2.2e-16 ***
#> b1    4.720309   0.383171 12.3191 < 2.2e-16 ***
#> a2i   3.833275   0.190004 20.1747 < 2.2e-16 ***
#> a2ii -0.080280   0.017580 -4.5666 4.956e-06 ***
#> b2    0.185412   0.019389  9.5627 < 2.2e-16 ***
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#> 
#> -2 log L: 1506.186

References

1. Lorenz LM, Koella JC. 2011 The microsporidian parasite vavraia culicis as a potential late life-acting control agent of malaria. Evolutionary Applications 4, 783–790. (doi:10.1111/j.1752-4571.2011.00199.x)

2. Lorenz LM, Koella JC. 2011 Data from: The microsporidian parasite vavraia culicis as a potential late life-acting control agent of malaria. Dryad Digital Repository (doi:10.5061/dryad.2s231)