Impute GWAS summary statistics using reference genotype data

pip install pyfizi==0.7.2


Functionally-informed Z-score Imputation (FIZI)

FIZI leverages functional information together with reference linkage-disequilibrium (LD) to impute GWAS summary statistics (Z-score).

This README is a working draft and will be expanded soon.


The easiest way to install fizi and pyfizi is through conda and conda-forge:

conda config --add channels conda-forge
conda install pyfizi

Alternatively you can use pip for installation:

pip install pyfizi

Or directly from the github repository:

git clone
cd fizi
pip install .

Check that FIZI was installed by typing

fizi --help

If that did not work, and pip install pyfizi --user was specified, please check that your local user path is included in $PATH environment variable. --user location and can be appended to $PATH by executing

export PATH=`python -m site --user-base`/bin/:$PATH

which can be saved in ~/.bashrc or ~/.bash_profile. To reload the environment type source ~/.bashrc or source ~/.bash_profile depending where you entered it.

We currently only support Python3.7+. Python2.7 and below is not supported


fizi has two main functions: munge and impute. The munge subcommand is a pruned down version of the LDSC munge_sumstats software with a few bells and whistles needed for our imputation algorithm. The impute subcommand performs summary statistic imputation using either the functionally informed algorithm (i.e. fizi) or using only reference-LD-only algorithm (i.e. ImpG). For a full list of features please refer to the help command: fizi munge -h or fizi impute -h.

Imputing summary statistics using only reference LD

When functional annotations and LDSC estimates are not provided to fizi, it will fallback to the classic ImpG algorithm described in ref 1. To impute missing summary statistics only for chromosome 1 using the ImpG algorithm simply enter the commands

1. fizi munge gwas.sumstat.gz --out cleaned.gwas
2. fizi impute cleaned.gwas.sumstat.gz plink_data_path --chr 1 --out imputed.cleaned.gwas.chr1.sumstat

By default fizi requires that at least 50% of SNPs to be observed for imputation at a region. This can be changed with the --min-prop PROP flag in step 2.

Incorporating functional data to improve summary statistics imputation

Usage consists of several steps. We outline the general workflow here when the intention to perform imputation on chromosome 1 of our data:

  1. Munge/clean all GWAS summary data before imputation

    fizi munge gwas.sumstat.gz --out cleaned.gwas

  2. Partitioning cleaned GWAS summary data into chr1 and everything else (loco-chr1).

  3. Run LDSC on locoChr to obtain tau estimates

  4. Perform functionally-informed imputation on chr1 data using tau estimates from loco-chr

Software and support

If you have any questions or comments please contact and/or

For performing various inferences using summary data from large-scale GWASs please find the following useful software:

  1. Association between predicted expression and complex trait/disease FUSION
  2. Estimating local heritability or genetic correlation HESS
  3. Estimating genome-wide heritability or genetic correlation UNITY
  4. Fine-mapping using summary-data PAINTOR