PSEV

The data provided here is not the same as that used in the paper. Patient IDs have been scrambled. This is only a fraction of the patient data used.

Clone repository

git clone https://github.com/baranzini-lab/PSEV.git

Go to constants directory

cd PSEV/constants/

Unzip lab path file

tar -xvzf cui_to_loinc_path.tsv.tar.gz

Go to build directory

cd ../build/

Unzip EHR files

tar -xvzf EHR_DIAGNOSIS.tsv.tar.gz

tar -xvzf EHR_LABS.tsv.tar.gz

tar -xvzf EHR_MEDICATION.tsv.tar.gz

Go to SPOKE directory

cd ../spoke_v_1/

Unzip SPOKE edge file

tar -xvzf neo4j_edges.tsv.tar.gz

Go to main PSEV directory

cd ..

Create matrix that contains patient connections to SPOKE nodes

python connect_all_patients.py

stdout (using default settings):

# of LOINC mapped to SPOKE: 557

Mapped SPOKE breakdown from LABS: Counter({'SideEffect': 226, 'Compound': 136, 'Gene': 49, 'Symptom': 12})

SEP STATS: (3183,) Counter({'SideEffect': 1851, 'Compound': 922, 'Symptom': 257, 'Disease': 137, 'Gene': 16})

Create PSEV matrix for specific SEPs (by node type: -t Gene)

python make_psevs_by_node_type.py

NOTE

Default creates PSEVs for Gene SEPs. To change default settings (including number of cores to use) please see python make_psevs_by_node_type.py --help

Beware this program will use a lot of memory and take a lot of time. Run on cluster if possible.

stdout (using default settings):

make_psevs_by_node_type.py:83: RuntimeWarning: invalid value encountered in divide add_val = np.nan_to_num(direct_hit.astype(float)/np.sum(direct_hit))

make_psevs_by_node_type.py:111: RuntimeWarning: invalid value encountered in divide connectivity_matrix = np.transpose(np.nan_to_num(connectivity_matrix/np.sum(connectivity_matrix, axis=0)))