macpan2

McMasterPandemic was developed to provide forecasts and insights to Canadian public health agencies throughout the COVID-19 pandemic. Much was learned about developing general purpose compartmental modelling software during this experience, but the pressure to deliver regular forecasts made it difficult to focus on the software itself. The goal of this macpan2 project is to re-imagine McMasterPandemic, building it from the ground up with architectural and technological decisions that address the many lessons that we learned from COVID-19 about software.

Impactful applied public health modelling requires many interdisciplinary steps along the path from epidemiological research teams to operational decision makers. Researchers must quickly tailor a model to an emerging public-health concern, validate and calibrate it to data, work with decision makers to define model outputs useful for stakeholders, configure models to generate those outputs, and package up those insights in an appropriate format for stakeholders. Unlike traditional modelling approaches, macpan2 tackles this challenge from a software-engineering perspective, which allows us to systematically address bottlenecks along this path to impact in ways that will make future solutions easier to achieve. The goal is to enable researchers to focus on their core strengths and fill knowledge gaps efficiently and effectively.

Although macpan2 is designed as a compartmental modelling tool that is agnostic about the underlying computational engine, it currently uses template model builder as the sole engine. Template model builder (TMB) is an R modelling package based on a C++ framework incorporating mature automatic differentiation and matrix algebra libraries.

The Public Health Risk Sciences Division at the Public Health Agency of Canada uses macpan2 (for example, here).

Documentation

• Package reference
• Quick-start guide
• TMB engine [specification document]
• Project history and trajectory [slides]

Installation

The standard recommended way to install macpan2 is with the following command.

repos = c('https://canmod.r-universe.dev', 'https://cloud.r-project.org')
install.packages('macpan2', repos = repos)

This command will install the current version of macpan2. There is no need to use remotes::install_github for the latest development version. For projects in production that need to keep track of specific versions of macpan2, snapshots and other reproducibility information can be obtained here. Please see this article for an explanation of how to manage reproducibility using r-universe.

Many workflows with macpan2 also make use of the following packages.

install.packages(c("dplyr", "ggplot2", "tidyr", "broom.mixed"))

Hello World

The following code specifies an SI model, which is I think is the simplest possible model of epidemiological transmission.

library(macpan2)
si = mp_tmb_model_spec(
    before = list(
        I ~ 1
      , S ~ N - I
    )
  , during = list(
        infection ~ beta * S * I / N
      , S ~ S - infection
      , I ~ I + infection
    )
  , default = list(N = 100, beta = 0.25)
)
print(si)

## ---------------------
## Default values:
## ---------------------
##  matrix row col  value
##       N         100.00
##    beta           0.25
## 
## ---------------------
## Before the simulation loop (t = 0):
## ---------------------
## 1: I ~ 1
## 2: S ~ N - I
## 
## ---------------------
## At every iteration of the simulation loop (t = 1 to T):
## ---------------------
## 1: infection ~ beta * S * I/N
## 2: S ~ S - infection
## 3: I ~ I + infection

Simulating from this model requires choosing the number of time-steps to run and the model outputs to generate. Syntax for simulating macpan2 models is designed to combine with standard data prep and plotting tools in R, as we demonstrate with the following code.

library(ggplot2)
library(dplyr)
(si
 |> mp_simulator(time_steps = 50, outputs = c("I", "infection"))
 |> mp_trajectory()
 |> mutate(quantity = case_match(matrix
    , "I" ~ "Prevalance"
    , "infection" ~ "Incidence"
  ))
 |> ggplot() 
 + geom_line(aes(time, value)) 
 + facet_wrap(~ quantity, scales = "free")
 + theme_bw()
)

Product Management

The project board tracks the details of bugs, tasks, and feature development.