cobeylab / antigen-vaccine Goto Github PK

This is a modification of Trevor Bedford's antigen code that implements vaccination.

The program requires Python 2.7.x and Java 1.7+ to compile and run.

The program can be built using:

[path-to]/antigen-vaccine/makeVaccine.py

This will create a dist (distribution) directory that can be used as the basis for a simulation run or a parameter sweep. Inside dist:

• run.py: a Python script that calls java to actually run the simulation. It can be modified to change, e.g., the amount of memory Java uses.
• vaccine: a directory containing tools to run a parameter sweep of the model on a SLURM cluster
• lib: the .jar files containing the simulation code as well as third-party libraries
• src: a copy of the source code, for the sake of knowing what version of the code generated your results.

Parameters are loaded from the file parameters.json into an object defined by the Parameters class defined in src/antigen/Parameters.java. You must specify all parameters for every simulation run; no defaults are provided in order to prevent confusion about which parameters are being used. If a parameter is left out, the simulation will throw an exception.

The one exception to this rule is the randomSeed parameter: if left out of the file, one will be generated at runtime. As an extra verification, all loaded parameters will be written into parameters_out.json, including the random seed actually used for the simulation, so that individual runs can be replicated.

To run a single simulation, execute run.py in a directory containing a parameters.json file. This will generate all output in that directory.

To run a parameter sweep, do the following:

1. Make a copy of the generated dist directory to the cluster.
2. Modify vaccine_parameters.json to include the parameter values shared among all runs.
3. Modify the run_vaccine.py script to perform the desired sweep. Specifically, you need to modify the generateJobs() function to yield a SLURM job name, directory name, and dictionary containing all the parameter values that vary from run to run.
4. Modify vaccine.sbatch to match the memory requirements for the job.
5. To start, the variable DRY is set to True, allowing you to do a "dry run" of the sweep where output directories and parameter files are generated, but jobs are not actually submitted. Try it by running run.py.
6. If the dry-run directory hierarchy and parameter files look good, you can submit the jobs by removing the test results directory, setting DRY to False, and running run.py again.

The simulation will output a timeseries of region-specific prevalence and incidence to out.timeseries. It will also sample viruses periodically and output their geographic and antigenic locations to out.tips and a tree connecting these samples to out.branches. This file contains pairs of viruses, child and parent, representing nodes in a genealogy. Average values are output to out.summary.

The vaccine strains will be periodically written to vaccine.timeseries. They are also written to a table in samples.sqlite, which may later be expanded to include all simulation output.

An example script vaccine_database.py is provided to show how to gather together results and summary statistics from output runs into a single SQLite database.

The script write_params_to_sqlite.py will record all the parameters in parameters_out.json from every simulation into a single SQLite database, in order to mitigate the need to keep a large number of files.

The -Xmx22G is required, because as an individual-based model, the memory requirements are typically quite large. Each host requires a minimum of 40 bytes of memory, plus 8 bytes per Phenotype recorded in its immune history. If the yearly attack rate is 15% and the host life span is 30 years, at equilibrium the average size of the immune history will be 4.5 references. This gives memory usage of: population size x 76 bytes. With 50 million hosts (used in the default parameters), the equals 3.8GB. Large epidemics, either stochastic or due to seasonal forcing, require even more memory.

In addition to hosts and immune histories, the simulation tracks the virus genealogy through VirusTree. This is harder to profile, and will continually grow in memory usage throughout the simulation. With the default parameters, VirusTree takes 5.5 MB at the end of a simulated year and may up to 110 MB at the end of the default 20 simulated years.

Memory can be easily profiled by calling jmap -histo <PID>.

Copyright Trevor Bedford, Sarah Cobey, Frank Wen, and Ed Baskerville, 2010-2017. Distributed under the GPL v3.