Skip to contents

Model and source

  • Citation: Chae SU, Min JS, Jo SJ, Lee CB, Park J, Bae SH, Bae SK. Prediction of first-in-human dose for new composition bee venom based on allometric scaling and pharmacokinetic modeling approach. Transl Clin Pharmacol. 2025;33(1):27-39. doi:10.12793/tcp.2025.33.e4
  • Article: https://doi.org/10.12793/tcp.2025.33.e4 (open access, PMC11976151)

Chae 2025 developed a new composition bee venom (NCBV), enriched in bee venom phospholipase A2 (bvPLA2), for Alzheimer’s disease and predicted its first-in-human (FIH) subcutaneous dose. Plasma bvPLA2 after a single subcutaneous (SC) NCBV injection was fitted separately in three species, each with a two-compartment model with first-order absorption (Figure 1, Table 2), and the three parameter sets were scaled to a 70 kg human by three allometric methods (Table 3). The paper therefore contributes six model files:

Model Species Basis
Chae_2025_beeVenomPLA2_mouse ICR mouse population PK fit, Table 2
Chae_2025_beeVenomPLA2_rat Sprague-Dawley rat population PK fit, Table 2
Chae_2025_beeVenomPLA2_dog beagle dog population PK fit, Table 2
Chae_2025_beeVenomPLA2_human_simple human (projected) simple allometry, Table 3
Chae_2025_beeVenomPLA2_human_brw human (projected) brain-weight corrected allometry, Table 3 (the paper’s recommended method)
Chae_2025_beeVenomPLA2_human_mlp human (projected) maximum-lifespan-potential corrected allometry, Table 3

All parameters are apparent (divided by the unknown SC bioavailability F). The animal models are per kg body weight, so their doses are given in mg/kg; the human projections take doses in mg. Concentrations are in ng/mL.

Population

  • Mice: male and female ICR mice (7-8 weeks, 20-35 g), a single 0.5 mg/kg SC dose, terminal cardiac-puncture sampling with 4 animals per time point (48 per sex) from 0.05 to 24 h.
  • Rats: male and female Sprague-Dawley rats (7-9 weeks, 200-330 g), 0.2, 0.5, 1 and 2 mg/kg SC, 4 per sex per dose, serial sampling to 24 h. An IV arm gave SC bioavailability of 11.9-18.4% (Table 1); the model itself is apparent.
  • Beagle dogs: 0.3, 0.6 and 0.9 mg/kg SC, 4 per sex per dose, sampling at 0.25-24 h.
  • Humans: no human data; the three human files are forward projections for a 70 kg adult.

bvPLA2 was measured by ELISA (calibration range 0.78-100 ng/mL). The same information is available programmatically, for example readModelDb("Chae_2025_beeVenomPLA2_rat")()$population.

Source trace

Equation / parameter Mouse Rat Dog Source location
Two-compartment model, first-order absorption and elimination Methods “PK model development”; Figure 1
lka (ka, 1/h) 9.08 0.803 0.544 Table 2
lcl (CL/F, L/h/kg) 0.243 1.83 0.150 Table 2
lvc (Vc/F, L/kg) 0.912 0.712 0.385 Table 2
lvp (Vp/F, L/kg) 5.89 6.89 1.47 Table 2
lq (Q/F, L/h/kg) 0.121 0.862 0.141 Table 2
etalcl (variance; CV%) 0.01806 (13.5%) 0.4186 (72.1%) 0.07393 (27.7%) Table 2 BSV
etalq - 0.3479 (64.5%) - Table 2 BSV
etalka - 0.1181 (35.4%) - Table 2 BSV
etalvc - - 0.1547 (40.9%) Table 2 BSV
propSd 0.327 0.322 0.166 Table 2 sigma prop
Parameter Simple BrW MLP Source location
lcl (CL/F, L/h) 12.9 3.25 0.612 Table 3
lvc (Vc/F, L) 19.5 4.92 10.0 Table 3
lvp (Vp/F, L) 67.3 17.0 34.7 Table 3
lq (Q/F, L/h) 14.7 3.72 7.59 Table 3
lka (1/h) 0.803 0.803 0.803 rat Table 2 value; back-solved against Table 4 (below)
e_wt_cl, e_wt_vc, e_wt_vp, e_wt_q 0.847, 0.842, 0.729, 0.971 - - Results, “Allometric scaling for the prediction of human PK parameters”
propSd 0 (fixed) 0 (fixed) 0 (fixed) not reported for the projections

BSV is reported as CV%; the log-normal variances are log(CV^2 + 1).

Animal models: visual predictive checks

Each species is simulated at its studied doses, 100 virtual animals per dose group, to mirror the VPCs of Figures 2-4.

make_animal <- function(model, species, doses, n = 100L) {
  obs_times <- sort(unique(c(0, 0.05, 0.08, 0.25, 0.5, 1, 1.5, 2, 3, 4, 6, 8, 10, 24, seq(0, 24, by = 0.25))))
  ev <- lapply(seq_along(doses), function(i) {
    ids <- (i - 1L) * n + seq_len(n)
    dose_rows <- data.frame(id = ids, time = 0, evid = 1L, amt = doses[i], cmt = "depot")
    obs_rows <- expand.grid(id = ids, time = obs_times)
    obs_rows$evid <- 0L
    obs_rows$amt <- 0
    obs_rows$cmt <- "central"
    out <- rbind(dose_rows, obs_rows[, names(dose_rows)])
    out$dose_group <- paste(doses[i], "mg/kg")
    out
  })
  ev <- dplyr::bind_rows(ev) |> dplyr::arrange(id, time, dplyr::desc(evid))
  rxode2::rxSetSeed(2025)
  sim <- rxode2::rxSolve(readModelDb(model), events = ev, keep = "dose_group")
  sim <- as.data.frame(sim)
  sim$species <- species
  sim$dose_group <- factor(sim$dose_group, levels = paste(doses, "mg/kg"))
  list(events = ev, sim = sim)
}

mouse <- make_animal("Chae_2025_beeVenomPLA2_mouse", "Mouse", 0.5)
#> ℹ parameter labels from comments will be replaced by 'label()'
rat <- make_animal("Chae_2025_beeVenomPLA2_rat", "Rat", c(0.2, 0.5, 1, 2))
#> ℹ parameter labels from comments will be replaced by 'label()'
dog <- make_animal("Chae_2025_beeVenomPLA2_dog", "Beagle dog", c(0.3, 0.6, 0.9))
#> ℹ parameter labels from comments will be replaced by 'label()'
vpc_summary <- function(sim) {
  sim |>
    dplyr::group_by(species, dose_group, time) |>
    dplyr::summarise(
      Q05 = quantile(Cc, 0.05, na.rm = TRUE),
      Q50 = quantile(Cc, 0.50, na.rm = TRUE),
      Q95 = quantile(Cc, 0.95, na.rm = TRUE),
      .groups = "drop"
    )
}
vpc <- dplyr::bind_rows(vpc_summary(mouse$sim), vpc_summary(rat$sim), vpc_summary(dog$sim)) |>
  dplyr::filter(time > 0) |>
  dplyr::mutate(panel = paste(species, dose_group))

ggplot(vpc, aes(time, Q50)) +
  geom_ribbon(aes(ymin = Q05, ymax = Q95), alpha = 0.25, fill = "steelblue") +
  geom_line(colour = "steelblue4") +
  facet_wrap(~panel, ncol = 4) +
  scale_y_log10() +
  labs(
    x = "Time after SC dose (h)", y = "Plasma bvPLA2 (ng/mL)",
    title = "Simulated median and 90% prediction interval",
    caption = "Replicates the VPCs of Figures 2 (mouse), 3 (rat) and 4 (dog) of Chae 2025."
  )

PKNCA against the observed NCA (Table 1)

Table 1 reports non-compartmental parameters of the observed data (AUC to the last sample at 24 h, terminal half-life, Cmax, Tmax). The simulation is sampled at the study schedule (mice and rats 0.05-24 h, dogs 0.25-24 h) and analysed with PKNCA.

study_times <- list(
  Mouse = c(0, 0.05, 0.08, 0.25, 0.5, 1, 1.5, 2, 3, 4, 6, 8, 10, 24),
  Rat = c(0, 0.05, 0.08, 0.25, 0.5, 1, 1.5, 2, 3, 4, 6, 8, 10, 24),
  "Beagle dog" = c(0, 0.25, 0.5, 1, 2, 4, 8, 24)
)
animal_sim <- dplyr::bind_rows(mouse$sim, rat$sim, dog$sim) |>
  dplyr::mutate(treatment = paste(species, dose_group), dose_group = as.character(dose_group))
animal_nca_conc <- animal_sim |>
  dplyr::filter(!is.na(Cc)) |>
  dplyr::filter(
    (species == "Beagle dog" & time %in% study_times[["Beagle dog"]]) |
      (species != "Beagle dog" & time %in% study_times[["Mouse"]])
  ) |>
  dplyr::select(id, time, Cc, treatment)
animal_dose <- animal_sim |>
  dplyr::distinct(id, treatment, dose_group) |>
  dplyr::mutate(time = 0, amt = as.numeric(sub(" mg/kg", "", dose_group)))

conc_obj <- PKNCA::PKNCAconc(animal_nca_conc, Cc ~ time | treatment + id)
dose_obj <- PKNCA::PKNCAdose(animal_dose, amt ~ time | treatment + id)
intervals <- data.frame(start = 0, end = 24, cmax = TRUE, tmax = TRUE, auclast = TRUE, half.life = TRUE)
animal_nca <- PKNCA::pk.nca(PKNCA::PKNCAdata(conc_obj, dose_obj, intervals = intervals))

# Table 1 of Chae 2025; AUC converted from ug*h/mL to ng*h/mL.
published_t1 <- tibble::tribble(
  ~treatment, ~cmax, ~tmax, ~auclast, ~half.life,
  "Mouse 0.5 mg/kg", 474, 0.5, 1280, 2.24,
  "Rat 0.2 mg/kg", 68.2, 0.5, 101, 1.05,
  "Rat 0.5 mg/kg", 113, 0.25, 209, 1.89,
  "Rat 1 mg/kg", 231, 0.25, 586, 3.13,
  "Rat 2 mg/kg", 439, 0.5, 1220, 4.17,
  "Beagle dog 0.3 mg/kg", 354, 1, 1350, 3.29,
  "Beagle dog 0.6 mg/kg", 477.4, 1, 2290, 3.11,
  "Beagle dog 0.9 mg/kg", 680, 2, 3150, 2.93
)
cmp_t1 <- nlmixr2lib::ncaComparisonTable(
  simulated = animal_nca,
  reference = published_t1,
  by = "treatment",
  units = c(cmax = "ng/mL", tmax = "h", auclast = "ng*h/mL", half.life = "h"),
  tolerance_pct = 20
)
knitr::kable(cmp_t1, caption = "Simulated (median of 100 animals per group) vs. Table 1 observed NCA. * differs by more than 20%.")
Simulated (median of 100 animals per group) vs. Table 1 observed NCA. * differs by more than 20%.
NCA parameter treatment Reference Simulated % diff
Cmax (ng/mL) Mouse 0.5 mg/kg 474 464 -2.1%
Cmax (ng/mL) Rat 0.2 mg/kg 68.2 38.8 -43.1%*
Cmax (ng/mL) Rat 0.5 mg/kg 113 107 -5.5%
Cmax (ng/mL) Rat 1 mg/kg 231 167 -27.7%*
Cmax (ng/mL) Rat 2 mg/kg 439 352 -19.8%
Cmax (ng/mL) Beagle dog 0.3 mg/kg 354 232 -34.5%*
Cmax (ng/mL) Beagle dog 0.6 mg/kg 477 485 +1.6%
Cmax (ng/mL) Beagle dog 0.9 mg/kg 680 715 +5.2%
Tmax (h) Mouse 0.5 mg/kg 0.5 0.5 +0.0%
Tmax (h) Rat 0.2 mg/kg 0.5 0.5 +0.0%
Tmax (h) Rat 0.5 mg/kg 0.25 0.5 +100.0%*
Tmax (h) Rat 1 mg/kg 0.25 0.5 +100.0%*
Tmax (h) Rat 2 mg/kg 0.5 0.5 +0.0%
Tmax (h) Beagle dog 0.3 mg/kg 1 2 +100.0%*
Tmax (h) Beagle dog 0.6 mg/kg 1 2 +100.0%*
Tmax (h) Beagle dog 0.9 mg/kg 2 2 +0.0%
AUClast (ng*h/mL) Mouse 0.5 mg/kg 1280 1560 +22.1%*
AUClast (ng*h/mL) Rat 0.2 mg/kg 101 108 +7.2%
AUClast (ng*h/mL) Rat 0.5 mg/kg 209 308 +47.3%*
AUClast (ng*h/mL) Rat 1 mg/kg 586 484 -17.4%
AUClast (ng*h/mL) Rat 2 mg/kg 1220 951 -22.0%*
AUClast (ng*h/mL) Beagle dog 0.3 mg/kg 1350 1670 +23.7%*
AUClast (ng*h/mL) Beagle dog 0.6 mg/kg 2290 3480 +52.0%*
AUClast (ng*h/mL) Beagle dog 0.9 mg/kg 3150 4830 +53.2%*
t½ (h) Mouse 0.5 mg/kg 2.24 8.43 +276.1%*
t½ (h) Rat 0.2 mg/kg 1.05 7.99 +661.1%*
t½ (h) Rat 0.5 mg/kg 1.89 8.73 +361.8%*
t½ (h) Rat 1 mg/kg 3.13 6.96 +122.3%*
t½ (h) Rat 2 mg/kg 4.17 7.52 +80.3%*
t½ (h) Beagle dog 0.3 mg/kg 3.29 7.23 +119.9%*
t½ (h) Beagle dog 0.6 mg/kg 3.11 7.46 +140.0%*
t½ (h) Beagle dog 0.9 mg/kg 2.93 6.75 +130.5%*

The comparison is against the observed NCA, so departures reflect both the model fit and the sparse design (mouse NCA is on pooled terminal samples). Cmax is within 20% for five of the eight groups; the paper itself notes that the rat model under-predicts Cmax at 0.2 mg/kg (Discussion, Figure 3). AUC to 24 h is within about 25% for five groups; the rat 0.5 mg/kg group and the dog 0.6 and 0.9 mg/kg groups are over-predicted by about 50%. For the dogs, the observed AUC is less than dose-proportional (dose-normalised AUC falls from 4.5 to 3.5 ug.h/mL per mg/kg across 0.3-0.9 mg/kg), which a single linear model fitted to all three doses cannot follow. The half-lives differ most. The fitted models have a slow disposition phase (mouse and rat Vp/F are 6-10 times Vc/F) that keeps the simulated profiles quantifiable to 24 h and puts their log-linear tail at 7-9 h, whereas the observed NCA half-lives (1-4 h) were fitted to the quantifiable samples above the 0.78 ng/mL calibration limit.

# Centre-of-distribution check on the clearance-driven endpoint: a
# mis-transcribed CL/F or V/F moves AUC by tens of percent.
auc_cmp <- as.data.frame(animal_nca$result) |>
  dplyr::filter(PPTESTCD == "auclast") |>
  dplyr::group_by(treatment) |>
  dplyr::summarise(sim = median(PPORRES), .groups = "drop") |>
  dplyr::inner_join(published_t1, by = "treatment") |>
  dplyr::mutate(ratio = sim / auclast)
stopifnot(abs(log(median(auc_cmp$ratio))) < log(1.5))

Mouse efficacious exposure

The FIH dose was back-calculated from the exposure of the efficacious dose in MPTP-model mice, 0.1 mg/kg, for which Chae 2025 reports an AUC of 0.397 ug.h/mL from the mouse model.

mouse_typ <- readModelDb("Chae_2025_beeVenomPLA2_mouse") |> rxode2::zeroRe()
#> ℹ parameter labels from comments will be replaced by 'label()'
ev_m <- rxode2::et(amt = 0.1, cmt = "depot") |> rxode2::et(seq(0, 240, by = 0.05), cmt = "central")
s_m <- as.data.frame(rxode2::rxSolve(mouse_typ, events = ev_m))
#> ℹ omega/sigma items treated as zero: 'etalcl'
trap <- function(t, c) sum(diff(t) * (head(c, -1) + tail(c, -1)) / 2)
mouse_auc <- c(
  "AUC0-24" = trap(s_m$time[s_m$time <= 24], s_m$Cc[s_m$time <= 24]) / 1000,
  "AUC0-240" = trap(s_m$time, s_m$Cc) / 1000,
  "AUC0-inf = dose / (CL/F)" = 0.1 / 0.243
)
knitr::kable(
  data.frame(Quantity = names(mouse_auc), "AUC (ug*h/mL)" = unname(signif(mouse_auc, 3)), check.names = FALSE),
  row.names = FALSE
)
Quantity AUC (ug*h/mL)
AUC0-24 0.305
AUC0-240 0.406
AUC0-inf = dose / (CL/F) 0.412

The reported 0.397 ug.h/mL lies between the 0-24 h AUC and the AUC to infinity of the typical mouse; the paper does not state the integration window. The mouse slow phase (Vp/F 5.89 L/kg, Q/F 0.121 L/h/kg) keeps a large fraction of the exposure beyond 24 h.

Human projections

Absorption rate constant

Table 3 gives CL/F, Vc/F, Vp/F and Q/F for each projection but no ka, although Methods states ka was scaled as well. Table 4 lists the simulated human Cmax, Tmax and half-life, so the ka used can be recovered. The sweep below solves the Simple and BrW projections at their Table 4 doses for several candidate ka values. The rat estimate, 0.803 1/h, reproduces Table 4 Cmax to 0.2% and Tmax to 0.01 h. The dog ka is too slow, and a power regression of the three species ka values on typical body weight (0.0275, 0.265 and 8-12 kg) gives 0.14-0.18 1/h at 70 kg, far too slow. All three human files therefore carry ka = 0.803 1/h, fixed.

two_cmt_sd <- function(ka, cl, vc, vp, q, dose, t = seq(0, 24, by = 0.01)) {
  k10 <- cl / vc
  k12 <- q / vc
  k21 <- q / vp
  s <- k10 + k12 + k21
  a <- (s + sqrt(s^2 - 4 * k10 * k21)) / 2
  b <- (s - sqrt(s^2 - 4 * k10 * k21)) / 2
  A <- ka / vc * (k21 - a) / ((ka - a) * (b - a))
  B <- ka / vc * (k21 - b) / ((ka - b) * (a - b))
  C <- ka / vc * (k21 - ka) / ((a - ka) * (b - ka))
  data.frame(t = t, c = 1000 * dose * (A * exp(-a * t) + B * exp(-b * t) + C * exp(-ka * t)))
}
sweep <- expand.grid(ka = c(0.17, 0.544, 0.803, 1, 2), method = c("Simple", "BrW"), stringsAsFactors = FALSE)
pars <- list(Simple = c(12.9, 19.5, 67.3, 14.7, 5.5), BrW = c(3.25, 4.92, 17.0, 3.72, 1.3))
sweep$Cmax <- NA_real_
sweep$Tmax <- NA_real_
for (i in seq_len(nrow(sweep))) {
  p <- pars[[sweep$method[i]]]
  d <- two_cmt_sd(sweep$ka[i], p[1], p[2], p[3], p[4], p[5])
  sweep$Cmax[i] <- max(d$c)
  sweep$Tmax[i] <- d$t[which.max(d$c)]
}
sweep |>
  dplyr::mutate(Cmax = round(Cmax, 1)) |>
  dplyr::rename("ka (1/h)" = ka, "Method" = method, "Cmax (ng/mL)" = Cmax, "Tmax (h)" = Tmax) |>
  knitr::kable(caption = "Table 4 targets: Simple 78.4 ng/mL at 1 h (5.5 mg); BrW 73.3 ng/mL at 1 h (1.3 mg).")
Table 4 targets: Simple 78.4 ng/mL at 1 h (5.5 mg); BrW 73.3 ng/mL at 1 h (1.3 mg).
ka (1/h) Method Cmax (ng/mL) Tmax (h)
0.170 Simple 27.7 2.25
0.544 Simple 62.1 1.20
0.803 Simple 78.3 0.99
1.000 Simple 88.5 0.88
2.000 Simple 123.7 0.61
0.170 BrW 26.0 2.25
0.544 BrW 58.1 1.20
0.803 BrW 73.4 0.99
1.000 BrW 82.8 0.88
2.000 BrW 115.9 0.61

Replicating Figure 5 and Table 4

human_models <- c(
  Simple = "Chae_2025_beeVenomPLA2_human_simple",
  BrW = "Chae_2025_beeVenomPLA2_human_brw",
  MLP = "Chae_2025_beeVenomPLA2_human_mlp"
)
human_doses <- c(Simple = 5.5, BrW = 1.3, MLP = 3.5)
grid <- sort(unique(c(seq(0, 24, by = 0.01))))
human_sim <- dplyr::bind_rows(lapply(names(human_models), function(m) {
  ev <- rxode2::et(amt = human_doses[[m]], cmt = "depot") |> rxode2::et(grid, cmt = "central")
  out <- as.data.frame(rxode2::rxSolve(readModelDb(human_models[[m]]), events = ev, params = c(WT = 70)))
  out$method <- m
  out
}))

animal_typ <- dplyr::bind_rows(lapply(
  list(
    c("Chae_2025_beeVenomPLA2_mouse", "Mouse (0.1 mg/kg)", 0.1),
    c("Chae_2025_beeVenomPLA2_rat", "Rat (0.7 mg/kg)", 0.7),
    c("Chae_2025_beeVenomPLA2_dog", "Beagle (0.07 mg/kg)", 0.07)
  ),
  function(x) {
    ev <- rxode2::et(amt = as.numeric(x[3]), cmt = "depot") |> rxode2::et(grid, cmt = "central")
    out <- as.data.frame(rxode2::rxSolve(readModelDb(x[1]) |> rxode2::zeroRe(), events = ev))
    out$profile <- x[2]
    out
  }
))
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalcl'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalq', 'etalka'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc'
fig5 <- dplyr::bind_rows(lapply(names(human_models), function(m) {
  dplyr::bind_rows(
    animal_typ |> dplyr::select(time, Cc, profile),
    human_sim |>
      dplyr::filter(method == m) |>
      dplyr::mutate(profile = paste0("Human (", human_doses[[m]], " mg)")) |>
      dplyr::select(time, Cc, profile)
  ) |>
    dplyr::mutate(method = m)
})) |>
  dplyr::filter(time > 0) |>
  dplyr::mutate(method = factor(method, levels = names(human_models)))

ggplot(fig5, aes(time, Cc, colour = profile, linetype = grepl("Human", profile))) +
  geom_line() +
  facet_wrap(~method, nrow = 1) +
  scale_y_log10(limits = c(0.1, 1000)) +
  scale_linetype_manual(values = c("solid", "dashed"), guide = "none") +
  labs(
    x = "Time (h)", y = "Plasma bvPLA2 (ng/mL)", colour = NULL,
    caption = "Replicates Figure 5 (A Simple, B BrW, C MLP) of Chae 2025; typical-value profiles."
  ) +
  theme(legend.position = "bottom")

The Simple and BrW human curves match Figures 5A-B (peak near 75 ng/mL at 1 h, a few ng/mL at 24 h). The MLP curve does not match Figure 5C: see the next section.

human_nca_conc <- human_sim |>
  dplyr::filter(!is.na(Cc), time %in% c(0, 0.25, 0.5, 1, 1.5, 2, 3, 4, 6, 8, 10, 12, 16, 20, 24)) |>
  dplyr::mutate(id = 1L) |>
  dplyr::select(id, time, Cc, method)
human_dose <- data.frame(id = 1L, time = 0, amt = unname(human_doses), method = names(human_doses))
h_nca <- PKNCA::pk.nca(PKNCA::PKNCAdata(
  PKNCA::PKNCAconc(human_nca_conc, Cc ~ time | method + id),
  PKNCA::PKNCAdose(human_dose, amt ~ time | method + id),
  intervals = data.frame(start = 0, end = 24, cmax = TRUE, tmax = TRUE, auclast = TRUE, half.life = TRUE)
))
# Table 4 of Chae 2025; AUC converted from ug*h/mL to ng*h/mL.
published_t4 <- tibble::tribble(
  ~method, ~cmax, ~tmax, ~auclast, ~half.life,
  "Simple", 78.4, 1, 409, 7.39,
  "BrW", 73.3, 1, 382, 7.40,
  "MLP", 64.3, 0.5, 378, 3.35
)
cmp_t4 <- nlmixr2lib::ncaComparisonTable(
  simulated = h_nca,
  reference = published_t4,
  by = "method",
  units = c(cmax = "ng/mL", tmax = "h", auclast = "ng*h/mL", half.life = "h"),
  tolerance_pct = 20
)
knitr::kable(cmp_t4, caption = "Typical human profile vs. Table 4 of Chae 2025. * differs by more than 20%.")
Typical human profile vs. Table 4 of Chae 2025. * differs by more than 20%.
NCA parameter method Reference Simulated % diff
Cmax (ng/mL) Simple 78.4 78.3 -0.1%
Cmax (ng/mL) BrW 73.3 73.3 +0.1%
Cmax (ng/mL) MLP 64.3 135 +109.7%*
Tmax (h) Simple 1 1 +0.0%
Tmax (h) BrW 1 1 +0.0%
Tmax (h) MLP 0.5 1.5 +200.0%*
AUClast (ng*h/mL) Simple 409 394 -3.6%
AUClast (ng*h/mL) BrW 382 370 -3.1%
AUClast (ng*h/mL) MLP 378 1670 +341.9%*
t½ (h) Simple 7.39 7.35 -0.5%
t½ (h) BrW 7.4 7.36 -0.6%
t½ (h) MLP 3.35 52.8 +1477.3%*
# Terminal half-life from the disposition eigenvalue of each projection.
terminal_t12 <- function(cl, vc, vp, q) {
  k10 <- cl / vc
  k12 <- q / vc
  k21 <- q / vp
  s <- k10 + k12 + k21
  log(2) / ((s - sqrt(s^2 - 4 * k10 * k21)) / 2)
}
pk_of <- function(m) {
  th <- readModelDb(human_models[[m]])()$theta
  exp(th[c("lcl", "lvc", "lvp", "lq")])
}
t12 <- sapply(names(human_models), function(m) do.call(terminal_t12, as.list(unname(pk_of(m)))))
peak <- human_sim |>
  dplyr::group_by(method) |>
  dplyr::summarise(cmax = max(Cc), tmax = time[which.max(Cc)], .groups = "drop")
pk <- setNames(peak$cmax, peak$method)
tk <- setNames(peak$tmax, peak$method)
knitr::kable(data.frame(
  Method = names(t12),
  "Terminal t1/2 (h)" = unname(round(t12, 2)),
  "Cmax (ng/mL)" = unname(round(pk[names(t12)], 1)),
  "Tmax (h)" = unname(tk[names(t12)]),
  check.names = FALSE
), row.names = FALSE)
Method Terminal t1/2 (h) Cmax (ng/mL) Tmax (h)
Simple 7.39 78.3 0.99
BrW 7.39 73.4 0.99
MLP 53.12 135.0 1.42
# Deterministic typical-value solves: tight tolerances are appropriate.
stopifnot(
  abs(pk[["Simple"]] / 78.4 - 1) < 0.01,
  abs(pk[["BrW"]] / 73.3 - 1) < 0.01,
  abs(tk[["Simple"]] - 1) <= 0.05,
  abs(tk[["BrW"]] - 1) <= 0.05,
  abs(t12[["Simple"]] / 7.39 - 1) < 0.02,
  abs(t12[["BrW"]] / 7.40 - 1) < 0.02,
  # The printed MLP parameter set gives a terminal half-life an order of
  # magnitude longer than Table 4 -- the discrepancy documented below.
  t12[["MLP"]] > 40
)

The Simple and BrW projections reproduce Table 4 Cmax, Tmax and terminal half-life to within 1-2%. The PKNCA AUC to 24 h is about 3% below Table 4 for both; the eigenvalue (terminal) half-life in the second table and the PKNCA half-life over 0-24 h both match Table 4 to within 1%.

The MLP projection

The MLP file carries Table 3 as printed (CL/F 0.612 L/h, also stated in the Results text). That parameter set does not reproduce the paper’s own MLP simulation:

  • With CL/F = 0.612 L/h the terminal half-life is about 53 h and the AUC to 24 h is about four times the Table 4 value (0.378 ug.h/mL); the 24 h concentration is ~50 ng/mL, whereas Figure 5C shows the human curve near 0.6 ng/mL at 24 h.
  • Table 4 lists “CL/F 0.00917” for MLP. Read on the same scale as its Simple and BrW entries (0.01248 and 0.00316, i.e. one thousandth of the Table 3 L/h values) this is 9.17 L/h, consistent with dose / AUC = 3.5 / 0.378 = 9.26 L/h, and with the 3.5 mg dose back-calculated from the mouse target (0.397 x 9.17 = 3.6 mg).
  • Even with CL/F = 9.17 L/h and the Table 3 volumes, no ka gives the Table 4 combination of Cmax 64.3 ng/mL at 0.5 h with a 3.35 h half-life, so the Table 4 MLP simulation used a parameter set that is not printed.
mlp_alt <- two_cmt_sd(0.803, 9.17, 10.0, 34.7, 7.59, 3.5)
knitr::kable(data.frame(
  "CL/F (L/h)" = c(0.612, 9.17),
  "Cmax (ng/mL)" = round(c(pk[["MLP"]], max(mlp_alt$c)), 1),
  "Tmax (h)" = c(tk[["MLP"]], mlp_alt$t[which.max(mlp_alt$c)]),
  "AUC0-24 (ug*h/mL)" = signif(c(
    trap(human_sim$time[human_sim$method == "MLP"], human_sim$Cc[human_sim$method == "MLP"]) / 1000,
    trap(mlp_alt$t, mlp_alt$c) / 1000
  ), 3),
  check.names = FALSE
), caption = "MLP projection with the Table 3 CL/F (shipped) and with the Table 4 CL/F read as 9.17 L/h (Table 4: Cmax 64.3 ng/mL, Tmax 0.5 h, AUC0-t 0.378 ug*h/mL).")
MLP projection with the Table 3 CL/F (shipped) and with the Table 4 CL/F read as 9.17 L/h (Table 4: Cmax 64.3 ng/mL, Tmax 0.5 h, AUC0-t 0.378 ug*h/mL).
CL/F (L/h) Cmax (ng/mL) Tmax (h) AUC0-24 (ug*h/mL)
0.612 135.0 1.42 1.670
9.170 87.3 0.89 0.366

The maintainers ship the printed Table 3 parameters, which are the paper’s stated projection, and flag the inconsistency here. To reproduce the Table 4 MLP exposure approximately, override the clearance: rxode2::ini(readModelDb("Chae_2025_beeVenomPLA2_human_mlp"), lcl = fixed(log(9.17))). The paper’s recommended FIH projection is BrW (highest R2, FIH dose 1.3 mg), which is unaffected.

Assumptions and deviations

  • Human ka. Not printed for any human projection. All three human files use the rat Table 2 estimate, 0.803 1/h (fixed), which reproduces the Table 4 Simple and BrW Cmax, Tmax and half-life (sweep above). This is a back-solved value, not a reported one.
  • MLP clearance. The MLP file carries CL/F = 0.612 L/h from Table 3 and the Results text, which conflicts with Table 4 and Figure 5C (see above).
  • Residual error. Methods describes a combined additive + proportional residual model, but Table 2 reports only a proportional term for each species, so the animal models carry proportional error only.
  • BSV. Reported as CV%; converted to log-normal variances with log(CV^2 + 1). Parameters without a reported BSV have none.
  • Units. The animal parameters are per kg (Table 2); the rat Vp/F is printed as “6.89 L” in the Results prose and as L/kg in Table 2, and the table unit is used. The Table 2 rat column header reads “0.2-0.5 mg/kg” while Methods, Table 1 and Figure 3 cover 0.2-2 mg/kg.
  • Human weight scaling. Only the simple-allometry exponents are printed, so only Chae_2025_beeVenomPLA2_human_simple scales with WT (reference 70 kg). The BrW and MLP regressions also depend on species brain weight or lifespan, whose values and coefficients are not printed; those two files encode the 70 kg prediction only.
  • Human projections are deterministic. They carry no between-subject variability, and the residual error is fixed to zero.
  • Mouse efficacious AUC. The integration window of the reported 0.397 ug.h/mL is not stated; it lies between the typical mouse AUC to 24 h and to infinity.
  • No erratum or correction notice for Chae 2025 was found as of 2026-11-09.