forecasts-flu

Seasonal influenza variant growth rates and frequency forecast

Data

We gratefully acknowledge the authors, originating and submitting laboratories of sequences from the GISAID EpiFlu Database on which this research is based. The files produced by this workflow represent heavily derived GISAID data and are equivalent to downloadable results from outbreak.info, cov-spectrum.org, and covariants.org. This use is allowable under the GISAID Terms of Use.

Environment setup

1. Create a GitHub token using this link.

   1. Expiration: 90 days (or however long you expect to be using)
   2. Scroll down and click Generate token

2. Copy the token for the next step.

3. Pull the custom Docker image (replace <GitHub username> and <GitHub token>)

   docker login ghcr.io -u <GitHub username> -p <access token>
   docker pull --quiet ghcr.io/blab/flu-geo-fitness:latest
   

Configure the workflow

Edit config/defaults.yaml to set the minimum number of total sequences per location to use for modeling, the geographic resolution, date range for analysis, and variant column to use from Nextclade annotations. Edit the MLR model config per lineage config/mlr-model/<lineage>.yaml to define the pivot to use for relative growth advantage calculations. Define the generation time in days between infections divided by the aggregation frequency of the input count data. For example, the generation time for H3N2 is 3.1 days, so for a model fit based on 14-day aggregation of counts, set the generation time to 0.22.

We define the variants used in these models in the seasonal-flu repository in one "emerging haplotypes" file per subtype and segment. To update the variants included in these results, update the corresponding files for H1N1pdm, H3N2, or Vic in that repository. Colors per haplotype are also defined in the corresponding Auspice config JSONs in the seasonal-flu repository.

Run the workflow

By default, the workflow will run MLR models for all builds defined in config/defaults.yaml and plot the inferred frequencies and growth advantages.

nextstrain build --docker --image=ghcr.io/blab/flu-geo-fitness:latest .

Alternately, you can run the workflow through the creation of MLR model JSONs and upload these JSONs to S3 by specifying an additional configuration file. Since we only need the custom Docker image for plotting model outputs, we can use the Nextstrain base image in the following command.

nextstrain build --docker . \
    --configfile config/defaults.yaml config/optional.yaml