LEAP

LEAP is a program for liability estimation in ascertained case-control genome-wide association studies, written in the Python language. LEAP estimates liabilities, that can then be treated as phenotypes in a GWAS context, which can greatly increase power. LEAP was published in: Accurate liability estimation improves power in ascertained case-control studies. Nat Methods 12, 332-4 (2015)

Installation

The easiest way to install LEAP is via pip, by typing the command:

pip install --user leap_gwas

Typically, the LEAP scripts will be installed at:

~/.local/lib/python2.7/site-packages/LEAP/

LEAP is particularly easy to install using the Anaconda Python distribution. The numerically optimized version of Anaconda can speed LEAP up by several orders of magnitude.

Alternatively (if numerically optimized Anaconda can't be installed), for very fast performance it is recommended to have an optimized version of Numpy/Scipy installed on your system, using optimized numerical libraries such as OpenBLAS or Intel MKL (see Compilation instructions for scipy with Intel MKL).

To install LEAP manually, you need the following dependencies:

• Python 2.7
• Numpy and Scipy
• Scikits-learn
• The Python FaST-LMM package.

Please make sure these are installed prior to using LEAP. To verify that everything is correctly installed, please run the script test.py in the regression directory. It will run a small example analysis and print an error message if any problem is found.

Usage instructions

There are two ways to run LEAP. The first is via a Python API. A detailed explanation about this option is provided in the LEAP Ipython notebook.

The second option is to run LEAP though a series of Python scripts, as detailed below. This option is more suitable for those not familiar with Python. The script leap_pipeline.sh runs the full LEAP pipeline on a small example dataset, and can be used for reference.

Generally, LEAP uses the same file formats as FaST-LMM. Namely, input files are in binary Plink format. When there is a contradiction between file formats used by Plink and by FastLMM, LEAP uses the convention adopted by FastLMM. Explanations about the parameters used by all the scripts can be seen by typing

python <script_name> --help

The command-line options for LEAP largely follow the options of the C++ version of FaST-LMM.

The LEAP pipeline

1) (optional): Find related individuals to be removed:

python findRelated.py --bfilesim <Plink base file> --out <output file>

This script creates a file marking the individuals that need to be removed to eliminate relatedness.

2) Compute an eigendecomposition of the kinship matrix:

python eigenDecompose.py --bfilesim <Plink base file> --out <output file> [--extractSim <SNPs used to estimate kinship> --pheno <phenotype file>]

This script computes a kinship matrix and its eigendecomposition, and saves them to speed up subsequent stages. It is recommended to perform a leave-one-chromosome-out (LOCO) analysis by computing a different kinship matrix for each chromosome, each one consisting of all SNPs except those on the selected chromosome. The optional extractSim file is a text file with a list of SNP names (one SNP per line) to facilitate this. The optional phenotype file is only used to exclude individuals with an unknown phenotype.

3) Compute heritability using the method of Golan et al.:

python calc_h2.py --bfilesim <Plink base file> --prev <prevalence> --pheno <phenotype file> --h2coeff 1.0 [--eigen <eigen file> --extractSim <SNPs used for heritability estimation>  --related <relatedness file> --h2coeff <heritability coefficient>]

This script outputs the heritability estimate. The optional eigen file is the one created in stage 2. The optional extractSim file is a text file with a list of SNP names (one SNP per line) that will be used for heritability estimation. It is recommended to perform a different heritability and liability estimation for every excluded chromosome, and then testing the SNPs on the excluded chromosome for association with the estimated liabilities. The bfilesim and extractSim parameters must be the same as the ones used in stage 2. The optional relatedness file should be the output of stage 1, and is used to exclude related individuals from the analysis, which improves analysis results.

4) Estimate liabilities:

python probit.py --bfilesim <Plink base file> --pheno <phenotype file> --prev <prevalence> --out <output base file> --h2 <heritability> [--eigen <eigen file> --extractSim <SNPs used in the liability estimation> --related <relatedness file>]

This script creates a file called <output base file>.liabs, with estimated liabilities for every individual. The estimated liabilities can be used directly for GWAS by using them as a standard phenotype file. The eigen file is the one computed in stage 2, and the h2 parameter should be the heritability estimate from stage 3. The extractSim and relatedness file parameters should be the same as in stage 3.

5) Test for Associations:

python leap_gwas.py --bfilesim <Plink base file for kinship estimation> --bfile <Plink file with tested SNPs> --pheno <estimated liabilities file> --out <output file> --h2 <heritability> [--eigen <eigen file> --extractSim <SNPs used in the LMM kinship matrix>  --extract <SNPs to test>]

This script performs GWAS with a prespecified heritability level (as computed in stage 2). The eigen file is the one from stage 2, and the pheno parameter is the liabilities file computed in stage 4. The syntax largely follows that of the C++ version of FaST-LMM. The bfile and bfilesim parameters can both point to the same file. In this case, the extract and extractSim parameters should be used to guarantee that kinship estimation doesn't use SNPs on the excluded chromosome, and that all tested SNPs are on the excluded chromosome.

General comments and tips

1) Fixed effects can be included in stages 3-5 by adding the flag --covar. Please type

python probit.py --help

and

python leap_gwas.py --help

for instructions. However, we note that under extreme ascertainment, it is recommded to use covariates only in stages 3-4 (see the paper for details).

2) As described in the main text, it is recommended to perform a different liability estimation for every excluded chromosome, and then testing the SNPs on the excluded chromosome for association with the estimated liabilities. The -extractSim flag is useful for this. Please see the example file leap_pipeline.sh for a usage example.

3) A complete end-to-end usage example is provided with the LEAP source files, and can be invoked via the script leap_pipeline.sh. This example estimates liabilities for a small balanced case-control dataset. The dataset was simulated with 50% heritability and 0.1% prevalence. It included 500 cases, 500 controls, 499 causal SNPs and 10000 SNPs differentiated with FST=0.01. Causal SNPs are called csnp<i>. The original liabilities for this file are available in the file dataset1.phe.liab (but this file is not used by LEAP).

Contact

For questions and comments, please contact Omer Weissbrod at omerw[at]cs.technion.ac.il