picopili

Purpose

• Provide a pipeline for GWAS of family data with arbitrary pedigrees, from sib-pairs to extended pedigrees, to work in parallel to ricopili for case/control and trio data

Primary Functions

• qc_rel.py: runs QC for family GWAS data. Designed to be consistent with ricopili trio QC, with minor adjustments to mendelian error handling and streamlined output.
• pca_rel.py: main driver script to run PCA on family GWAS data. Includes strict QC (with LD pruning, etc), extraction of a set of unrelated individuals to compute PCs, and projection of PCA to full dataset. Jobs are submitted via UGER, and ricopili-like success/failure emails are sent on completion. See tasks strict_qc.py and imus_pca.py.
• admix_rel.py: estimates relatedness for admixed samples. Relatedness is estimated by starting with a subset of individuals who are unrelated, running an Admixture analysis, projecting that Admixture solution to the remaining samples, and using those Admixture results as the basis for ancestry-adjusted relatedness estimation using REAP.
• filter_ped.py: uses genetic relatedness information to flag (a) cryptic relatedness across FIDs, (b) unrelated individuals within FIDs, (c) possible parent/offspring pair not identified in .fam file, and (d) parent/offspring pairs indicated by pedigree that aren't supported by genetic relatedness. Provides a suggested sample exclusion list based on weighted preferences for phenotype, pedigree, and genotyping rate.
• ped_confirm.py: confirms that reported pedigrees in .fam file are consistent with genetic relatedness. Works as a wrapper to find_expected_pedigree.pl from PRIMUS.
• gwas_rel.py: Genome-wide association analysis (in parallelized chunks) with either plink --dfam, a GEE model, or a logistic mixed model (using GMMAT).
• impute_rel.py: Imputation pipeline for related samples using SHAPEIT's --duohmm and IMPUTE2. Includes build check (with liftOver if needed), alignement to reference, phasing, imputation, and best-guess genotype calls with MAF/info score/mendelian error filtering.

Installation

1. The following assumes:

• You have git configured with a github account.
• You have dependencies installed, or can install them (e.g. Admixture, REAP, PRIMUS).
• You're either on one of the default clusters (Broad, Lisa) or can define appropriate configuation files for your local compute environment (see step 6).

2. Download picopili from github

mkdir ~/github
cd ~/github
git clone https://github.com/Nealelab/picopili.git

3. Run the configuration scripts

cd picopili
./CONFIG
./GET_REFS

Follow on screen instructions for each script. If dependencies are missing or need to be adjusted, see ~/.picopili.conf.

4. Add picopili to search path

In ~/.my.bashrc (or equivalent for your shell) add the line:

PATH=$HOME/github/picopili/bin:$PATH

5. Logout and log back in.

Applies the changes in ~/.my.bashrc.

6. Define cluster configuration

You will also need to define a cluster configuration with a .conf configuation file and a .sub.sh sumbission script template. See ./picopili/cluster_templates/ for examples for a SGE/UGE job system (broad_uger.*) and for a TORQUE/Maui job system with tasks parallelized within node as well as across nodes (lisa.*).

Contact

Bug report? Open an issue right here on github!

Questions? Feedback? Email the authors. (rwalters at broad)