Overview

This repository provides an analytical pipeline for calibrated projections of SARS-CoV-2 transmission and COVID19 burden for specified regions.

Analysis Engineering

This analysis is designed to be run from the command line, specifically in an high performance computing setup, but with allowances for development on a personal machine.

The core model is covidm. There is a convenience target for cloning that repository, but this does not manage installation of covidm dependencies. The .install target will attempt to compile covidm; subsequent uses of covidm will not recompile that library.

This repository does define other dependencies (i.e., R packages) and inputs (e.g. case and death data), and provides targets for their download and cleaning.

This analyses is written in "map-reduce" paradigm, with the parallel steps defined by different ISO country codes. Inputs and outputs are segregated by file paths; within each source (inputs) and sink (outputs) subfolder, there are identical files. Generic inputs and outputs are defined at the source and sink folder level, while locale-specific ones are defined in subfolders or files named by ISO code.

The Makefile defines the relationships between inputs and outputs, in terms of various scripts. Assorted top level variables, e.g. path to the source and sink folders, can be defined in a local.makefile; these variables are distinguishable by the ?= assignment in the Makefile. The example.makefile shows the structure for a local.makefile, and make local.makefile will create a copy for you to edit.

In general, each target represents a step in the analysis, typically accomplished by executing a single script with dependencies and the target as arguments, e.g.:

%/output.last: somescript.R %/someinput %/moreinput
  Rscript $^ $@

A somescript.R generally follows this pattern:

suppressPackageWarnings({
#' require assorted packages
})

.debug <- c("DATADIR", "PATTERN") #' an example % match from Makefile

#' this structure will use `.debug` to create the relevant arguments
#' when working interactively with the script
.args <- if (interactive()) sprintf(c(
  "%s/input/population/%s.csv", "%s/output/intermediate_result/%s.rds",
  "%s/output/next_result_step/%s.rds"
), .debug[1], .debug[2]) else commandArgs(trailingOnly = FALSE)

#' assign `.args` into sensible variables
#' note that typically the target is the final item in `.args`
tarfile <- tail(.args, 1)

#' PERFORM VARIOUS CLEANING / ANALYSIS / PLOTTING / ETC

saveRDS(result, tarfile)
#' or save, qsave, ggsave, etc as appropriate to the script target

Naming

Scripts follow these naming conventions:

get_...: gets data from a remote source; requires open internet access; typically puts products in SOURCES (i.e. inputs) directory
gen_...: synthesizes local data into new local data; only re-organization, no analysis; typically puts products in SOURCES (i.e. inputs) directory, though to SINKS (i.e. outputs) if it's about reorganizing structure of outputs
est_...: imputes some result from local data; maybe either deterministic or stochastic; products to SINKS
sim_...: simulates (projects) some process using assumptions in local data; products in SINKS
fig_...: visualizes some element of the analysis; products in SINKS

Setup

make .install will attempt to install dependencies. make allinputs will attempt to get raw inputs.

Data Sources

Reported Cases and Deaths

We use the JHU Center for Systems Science and Engineering (CSSE) time series for cases and deaths (repository link). Aside from superficial organizational changes, we perform one cleaning step to address negative case counts (if necessary; the process emits a warning message when this occurs). We assume negative reports represent a correction to earlier data, thus we distribute negative incidence into previous daily counts. See get_epi_data.R for specifics.

Sequence Data

We obtain sequence metadata from the (Network for Genomics Surveillance in South Africa)[https://www.krisp.org.za/ngs-sa/index.php], via NextStrain.

Analysis Steps

determine when pre-/post- [and optionally, modification] intervention periods occur
calculate static Rt for these periods
use the static Rt as an approximate R

this implies pre-R == R0 with no interventions
post-R == R0 with interventions in place
optional modification-R == Reff (i.e. with both susceptible depletion and some intervention)

using R values, fit assorted intervention scenarios

for the modification-R, this means fitting the pre-/post-R first, then projecting (to deplete susceptibles)

with intervention fits, run projections

cmmid/SA2UK