In this vignette, we show a proof-of-concept of a new sampling framework in which the mosquito dynamics are decoupled from the human dynamics. The two systems pass relevant information to each other at time step for the duration of the simulation. The motivation for decoupling the human and mosquito dynamics is to be able to incorporate more complex models of disease transmission into MGDrivE-2’s sampling framework. While this vignette shows a simple SIS model, the eventual goal is to incorporate the Imperial College model of malaria transmission (https://www.researchsquare.com/article/rs-72317/v1) to model the epidemiological effects of gene drive organisms. This model, alongside other complex models, are not directly compatible with MGDrivE-2’s stochastic Petri net (SPN) architecture due to continuous-state immunity functions and non-Markovian delays, and therefore are separated into their own module. Future vignettes will showcase the decoupled functionality with the Imperial model, but here we showcase the functionality with an SIS model.
In this way, we can still leverage the entomological simulations furnished by MGDrivE-2 and apply the relevant parameters to the epidemiological module. This framework also allows for other models of disease transmission to be swapped in when needed. Here, only the mosquito component functions as an SPN, whereas the human component is formulated using ODEs. For a more complete overview of the decoupled sampling framework, see: https://www.overleaf.com/read/hhwbxpqnhzfv
We start by loading the MGDrivE2 package, as well as the MGDrivE package for access to inheritance cubes and ggplot2 for graphical analysis. We will use the basic cube to simulate Mendelian inheritance for this example.
# simulation functions
library(MGDrivE2)
# inheritance patterns
library(MGDrivE)
# plotting
library(ggplot2)
# basic inheritance pattern
<- MGDrivE::cubeMendelian() cube
Several parameters are necessary to setup the structural properties
of the Petri Net, as well as calculate the population distribution at
equilibrium, setup initial conditions, and calculate hazards. Again, we
specify all entomological parameters as for the mosquito-only simulation
(see “MGDrivE2: One Node Lifecycle
Dynamics”) as well as additional parameters for the
SEI mosquito dynamics. Like the aquatic stages, \(\frac{1}{q_{\mathit{EIP}}}\) will give the
mean dwell time for incubating mosquitoes, and variance by \(\frac{1}{n_{\mathit{EIP}} \cdot
q_{\mathit{EIP}}^{2}}\). The model requires muH
,
mortality rate in humans, because equilibrium dynamics are simulated
(that is, human populations follow an “open cohort” with equal birth and
death rates). A table of (case-sensitive) epidemiological parameters the
user needs to specify is given below. Note that all parameters must be
specified as a rate per day. For a detailed discussion of these
parameters in the context of malaria models, see Smith and McKenzie
(2004).
Parameter | Description |
---|---|
NH |
total human population size |
X |
equilibrium prevalence of disease in humans |
f |
mosquito feeding rate |
Q |
proportion of blood meals taken on humans (human blood index in field literature) |
b |
mosquito to human transmission efficiency |
c |
human to mosquito transmission efficiency |
r |
rate of recovery in humans |
muH |
mortality rate in humans |
qEIP |
inverse of mean duration of EIP |
nEIP |
shape parameter of Erlang-distributed EIP |
Please note that f
and Q
must be specified;
this is because future versions of MGDrivE2 will
include additional vector control methods such as IRS (indoor residual
spraying) and ITN (insecticide treated nets). In the presence of
ITNs/IRS f
will vary independently as a function of time
depending on intervention coverage.
Additionally, we specify a total simulation length of 300 days, with output stored daily.
# entomological and epidemiological parameters
<- list(
theta # lifecycle parameters
qE = 1/4,
nE = 2,
qL = 1/3,
nL = 3,
qP = 1/6,
nP = 2,
muE = 0.05,
muL = 0.15,
muP = 0.05,
muF = 0.09,
muM = 0.09,
beta = 16,
nu = 1/(4/24),
# epidemiological parameters
NH = 1000,
X = 0.25,
f = 1/3,
Q = 0.9,
b = 0.55,
c = 0.15,
r = 1/200,
muH = 1/(62*365),
qEIP = 1/11,
nEIP = 6
)$a <- theta$f*theta$Q
theta
# simulation parameters
<- 250
tmax <- 1 dt
We also need to augment the cube with genotype specific transmission
efficiencies; this allows simulations of gene drive systems that confer
pathogen-refractory characteristics to mosquitoes depending on genotype.
The specific parameters we want to attach to the cube are b
and c
, the mosquito to human and human to mosquito
transmission efficiencies. We assume that transmission from human to
mosquito is not impacted in modified mosquitoes, but mosquito to human
transmission is significantly reduced in modified mosquitoes. For
detailed descriptions of these parameters for modeling malaria
transmission, see Smith & McKenzie (2004) for extensive discussion.
These genotype-specific transmission efficiencies are used in the human
ODE model to determine the rates of movement between susceptible and
infected compartments.
# augment the cube with RM transmission parameters
$c <- setNames(object = rep(x = theta$c, times = cube$genotypesN), nm = cube$genotypesID)
cube$b <- c("AA" = theta$b, "Aa" = 0.35, "aa" = 0) cube
The SEI disease transmission model sits “on top” of
the existing MGDrivE2 structure, using the default
aquatic and male “places”, but expanding adult female “places” to follow
an Erlang-distributed pathogen incubation period (called the extrinsic
incubation period, EIP). Information on how to choose
the proper EIP distribution can be found in the help
file for ?makeQ_SEI()
.
The transitions between states is also expanded, providing
transitions for females to progress in infection status, adding human
dynamics, and allowing interaction between mosquito and human states.
All of these additions are handled internally by
spn_T_epiSIS_node()
. Since only the mosquito portion is
stochastic, the SPN will only include the mosquito states. Human states
will be handled by the sampling algorithm in the form of a deterministic
ODE.
# Places and transitions
# note decoupled sampling is only supported currently for one node.
<- spn_P_epi_decoupled_node(params = theta, cube = cube)
SPN_P <- spn_T_epi_decoupled_node(spn_P = SPN_P, params = theta, cube = cube)
SPN_T
# Stoichiometry matrix
<- spn_S(spn_P = SPN_P, spn_T = SPN_T) S
Now that we have set up the structural properties of the Petri Net, we need to calculate the population distribution at equilibrium and define the initial conditions for the simulation.
The function equilibrium_SEI_SIS()
calculates the
equilibrium distribution of female mosquitoes across
SEI stages, based on human populations and
force-of-infection, then calculates all other equilibria. We set the
logistic form for larval density-dependence in these examples by specify
log_dd = TRUE
.
# SEI mosquitoes and SIS humans equilibrium
# outputs required parameters in the named list "params"
# outputs initial equilibrium for adv users, "init
# outputs properly filled initial markings, "M0"
<- equilibrium_SEI_decoupled_mosy(params = theta, phi = 0.5, log_dd = TRUE,
initialCons spn_P = SPN_P, cube = cube)
# augment with human equilibrium states
$H <- equilibrium_SEI_decoupled_human(params = theta) initialCons
With the equilibrium conditions calculated (see
?equilibrium_SEI_SIS()
), and the list of possible
transitions provided by spn_T_epiSIS_node()
, we can now
calculate the rates of those transitions between states.
# approximate hazards for continuous approximation
<- spn_hazards_decoupled(spn_P = SPN_P, spn_T = SPN_T, cube = cube,
approx_hazards params = initialCons$params, type = "SIS",
log_dd = TRUE, exact = FALSE, tol = 1e-8,
verbose = FALSE)
Similar to previous simulations, we will release 50 adult females
with homozygous recessive alleles 5 times, every 10 days, but starting
at day 20. Remember, it is critically important that the event
names match a place name in the simulation. The simulation
function checks this and will throw an error if the event name does not
exist as a place in the simulation. This format is used in
MGDrivE2 for consistency with solvers in
deSolve
.
# releases
<- seq(from = 20, length.out = 5, by = 10)
r_times <- 50
r_size <- data.frame("var" = paste0("F_", cube$releaseType, "_", cube$wildType, "_S"),
events "time" = r_times,
"value" = r_size,
"method" = "add",
stringsAsFactors = FALSE)
As a further example, we run a single stochastic realization of the
same simulation, using the tau-decoupled
sampler with \(\Delta t = 1\), approximating 10 jumps per
day. This means that we use a tau-leaping sampler in the mosquito
states’ SPN, and ODE integration the human model. As the adult male
mosquitoes do not contribute to infection dynamics, we will only view
the adult female mosquito and human dynamics here.
# delta t - one day
<- 0.1
dt_stoch
# run tau-leaping simulation
<- sim_trajectory_R_decoupled(
tau_out x0 = initialCons$M0,
h0 = initialCons$H,
SPN_P = SPN_P,
theta = theta,
tmax = tmax,
inf_labels = SPN_T$inf_labels,
dt = dt,
dt_stoch = dt_stoch,
S = S,
hazards = approx_hazards,
sampler = "tau-decoupled",
events = events,
verbose = FALSE,
human_ode = "SIS",
cube = cube
)
# summarize females/humans by genotype
<- summarize_females_epi(out = tau_out$state, spn_P = SPN_P)
tau_female <- summarize_humans_epiSIS(out = tau_out$state)
tau_humans
# plot
ggplot(data = rbind(tau_female, tau_humans) ) +
geom_line(aes(x = time, y = value, color = inf)) +
facet_wrap(~ genotype, scales = "free_y") +
theme_bw() +
ggtitle("SPN: Tau Decoupled Solution")
Analyzing one stochastic realization of this system, we see some similarities and some striking differences. The releases are clearly visible, lower left-hand plot, and we see that the initial dynamics are similar to the ODE dynamics. However, it is quickly apparent that the releases are not reducing transmission adequately, that in fact, disease incidence is increasing rapidly in human and female mosquitoes. There are two main possibilities for this: first, that the stochastic simulation just happens to drift like this, a visual reminder that there can be significant differences when the well-mixed, mean-field assumptions are relaxed, or that the step size (\(\Delta t\)) is too large, and the stochastic simulation is a poor approximation of the ODE solution. Further tests, with \(\Delta t = 0.05\) and \(\Delta t = 0.15\), returned similar results, indicating that this is an accurate approximation but still highlighting the importance of testing several values of \(\Delta t\) for consistency.