Privacy-preserving prediction of phenotypes from genotypes using homomorphic encryption

The implementation uses C++ and Microsoft SEAL library.

How to run

SEAL library

We use SEAL library for HE scheme. After pulling the code from github, build SEAL manually following the guide of its README.

    git clone https://github.com/microsoft/SEAL.git
    cd SEAL
    cmake -S . -B build
    cmake --build build
    cd ..

Build

Run CMake file to make binary file

    cmake -S . -B bin
    cmake --build bin

As a result, you get two outputs: binary and continous.

Run

continuous is for the models with continous phenotypes (1-3), and binary is for binary ones (4-5). Run the following arguments.

    for pheno in 1 2 3;
        do ./bin/continuous $pheno;
    done
    for pheno in 4 5;
        do ./bin/binary $pheno;
    done

Otherwise, of course you can run each model one by one, or run one of following argument if you want to see the result of one model.

    ./continuous 1
    ./continuous 2
    ./continuous 3
    ./binary 4
    ./binary 5

The time is measured at each step and printed at the end of the executions. During this process, ciphertexts are stored in the folder savefiles/ and loaded to be used for the next required step. The outputs of each model are stored in the folder output/.

Accruacy

We provide python code for computing accruacies and combining outputs. Run following python code.

    python combine_result.py

As a result, the combined outputs are saved in the file ./all_models_final.csv.