PsyOPS

PsyOPS (short for Psychiatric Omnilocus Prioritization Score) is a software package for prioritizing causal genes from psychiatric genome-wide association studies (GWAS).

For a detailed description of PsyOPS, check out our paper!

Getting started

Download this repository, which contains the PsyOPS code (PsyOPS.py) and all necessary data files:

git clone https://github.com/Wainberg/PsyOPS
cd PsyOPS

PsyOPS requires Python >= 3.6 with the numpy, scipy, pandas, and scikit-learn packages. You can install Python and the required packages with Miniconda as follows:

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh -bp ~/.anaconda3
rm Miniconda3-latest-Linux-x86_64.sh
export PATH="~/.anaconda3/bin:$PATH"
conda install -y numpy scipy pandas scikit-learn

Running PsyOPS

PsyOPS.py takes two arguments, both of which are mandatory:

1. --GWAS-hit-file: a tab-separated file of lead variants. Must contain 3 columns: "rs" (the variant's rs number), "chrom" (the variant's chromosome; only autosomal variants will be analyzed), and "bp_hg19" (the variant's hg19/GRCh37 base pair coordinates). Chromosomes can be specified either with or without the "chr" prefix (e.g. both "1" and "chr1" are fine).
2. --output-file: a tab-separated file where results will be output. Each row lists a gene within 500 kilobases of one of the lead variants. Columns denote the lead variant, the distance to the gene, the gene, the gene's chromosome, transcription start site and transcription end site (in hg19/GRCh37 coordinates), whether the gene has each of the three PsyOPS features (extreme pLI, brain-enriched expression, neurodevelopmental disorder), and the gene's PsyOPS score.

As an example, let's run PsyOPS on the 102 lead variants from a recent major depressive disorder (MDD) GWAS , provided in the file Howard_102_MDD.tsv in this repository:

python PsyOPS.py --GWAS-hit-file Howard_102_MDD.tsv --output-file PsyOPS_MDD.tsv

This takes about 5 seconds to run and prints the following output:

Loaded 102 autosomal GWAS hits from Howard_102_MDD.tsv

88 of 708 genes (12.4%) are positives (i.e. nearest genes)

Model coefficients (when trained without cross-validation):
                                   OR  lower_CI   upper_CI             p
Intercept                    0.071024  0.051358   0.098221  1.531120e-57
Extreme pLI                  2.710843  1.491015   4.928634  1.076882e-03
Brain-enriched expression    2.927626  1.661044   5.160004  2.033198e-04
Neurodevelopmental disorder  5.458068  2.909110  10.240418  1.249739e-07

WARNING: only one class represented for chromosome chr15; excluding from accuracy calculation

Average performance across chromosomes: AUC 0.71 +/- 0.05, AUPRC 0.53 +/- 0.07

Prioritized genes:
ADARB2, AGBL4, ASTN2, ATP2A2, BRINP2, CABP1, CACNA1D, CACNA1E, CDC42BPB, CELF4, CHD6, CNTLN, CRB1, CRYBA1, CTNND1, CTTNBP2, CYP7B1, DAAM1, DAGLA, DCC, DDX6, DNAH3, DOCK9, DPF2, DRD2, EIF3B, ELAVL2, EP300, ESR2, EYS, FHIT, GRIK2, GRIK3, GRM5, H4C12, HIVEP2, KCNG2, KIF2A, KIRREL3, KLF7, LHX2, LIN28B, LRFN5, MEF2C, MEIS2, MPPED2, MRTFB, MSANTD1, NCOA2, NEGR1, NR2F1, NR4A2, PAX5, PCDH8, PCDH9, PCLO, PHOX2B, PPFIA1, PROX2, QRICH1, RAB3B, RBFOX1, REEP1, RERE, RFX3, RSRC1, SEMA6D, SGIP1, SHISA9, SLC12A5, SORCS3, SOX5, SPRY2, STXBP5, TCF4, TEX26, THSD7A, TNR, TRPC3, TUSC1, VRK2, WDR17, ZNF445, ZNF536

Exported results to PsyOPS_MDD.tsv

We also provide lead variant files for the other two psychiatric GWAS analyzed in the PsyOPS paper, for bipolar disorder (Mullins_64_BIP.tsv) and schizophrenia (Pardinas_145_SCZ.tsv).