Skip to content

Prepare deconvoluted data with LolliPop for Covvfit

Here we explain how to prepare input for Covvfit using the LolliPop tool.

We need to deconvolute wastewater data with LolliPop, but we need to do it without smoothing (or with minimal smoothing) to avoid introducing bias by kernel smoothing procedures.

Install LolliPop

Follow installation instructions described here. They consist of the following:

Create environment: 

conda create -n lollipop 
conda activate lollipop

Clone repository:

git clone https://github.com/cbg-ethz/LolliPop.git
cd LolliPop

Install dependencies: 

pip install '.[cli]'

Get and prepare data

Make a directory for deconvolution results:

mkdir lollipop_covvfit
cd lollipop_covvfit

Obtain mutation data, for example from Euler:

rsync -avz --progress euler:/cluster/project/pangolin/work-vp-test/variants/tallymut.tsv.zst .
zstd -d tallymut.tsv.zst

Obtain the latest configuration files from euler:

rsync -avz --progress euler:/cluster/project/pangolin/work-vp-test/var_dates.yaml .
rsync -avz --progress euler:/cluster/project/pangolin/work-vp-test/variant_config.yaml .
rsync -avz --progress euler:/cluster/project/pangolin/work-vp-test/ww_locations.tsv .
rsync -avz --progress euler:/cluster/project/pangolin/work-vp-test/filters_badmut.yaml .

Prepare parameters for the deconvolution: 

cat << EOF > deconv_config.yaml
bootstrap: 0

kernel_params:
  bandwidth: 0.1

regressor: robust
regressor_params:
  f_scale: 0.01

deconv_params:
  min_tol: 1e-3
EOF

Run LolliPop

To deconvolve the data, run:

cd ..
ldata="./lollipop_covvfit"
lollipop deconvolute $ldata/tallymut.tsv \
    -o $ldata/deconvolved.csv \
    --variants-config $ldata/variant_config.yaml \
    --variants-dates $ldata/var_dates.yaml \
    --deconv-config $ldata/deconv_config.yaml \
    --filters $ldata/filters_badmut.yaml  \
    --seed=42 \
    --n-cores=2

Run CovvFit

If you followed the installation guide here, you are ready to use Covvfit. You can have a look at the tutorial here.

Prepare parameters

In the example below, we are looking at tracking the variants KP.2, KP.3, XEC, LP.8, NB.1.8.1 and XFG. We will prepare a config yaml before running:

cat << EOF > covvfit_config.yaml
variants:
  - KP.2
  - KP.3
  - XEC
  - LP.8
  - NB.1.8.1
  - XFG
plot:
  dimensions:
    right: 1.2
  time_spacing: 3
  variant_colors:
    JN.1: '#00E9FF' 
    KP.2: '#7A5B58' 
    KP.3: '#3A38F3'
    XEC: '#6C3072'  
    LP.8: '#379E33'
    NB.1.8.1: '#BDC94D'  
    XFG: '#9E7A29'  
    undetermined: '#969696'
EOF

Run CovvFit on LolliPop output

Run on the last 365 days of data, and provide a 90 days horizon forecast. 

covvfit infer -i deconvolved.csv -o output -c covvfit_config.yaml --max-days 365 --horizon 90

After running we can find the results in the output folder, including the figures together with the pairwise_fitnesses.csv.