Predicting TWAS features (FeaturePred)

FeaturePred is a tool designed to simplify the process of predicting features (e.g. gene expression) in a target sample. It is designed to work with FUSION formated SNP-weights files and PLINK formatted genotype data. It uses a FUSION released script to convert the weights files into a PLINK .SCORE file, which is then used to predict the feature in a target sample using PLINK. The script harmonises the target sample to the reference data automatically and efficiently handles very large target sample datasets.

Access to weights files and more information on FUSION can be found here.

Getting started

Prerequisites

• R and the required packages:

install.packages(c('data.table','optparse','foreach','doMC'))

• FUSION software:

git clone https://github.com/gusevlab/fusion_twas.git

• FUSION LD reference data (download)

• PLINK 1.9 software

• pigz software for parallel gz compression

• Target sample genetic data

  • Binary PLINK format (.bed/.bim/.fam)
  • RSIDs should match the FUSION LD reference data (1000 Genomes phase 3)

• FUSION formatted SNP-weights

  • Can be downloaded from the FUSION website
  • To make your own FUSION format SNP-weights, see the FUSION website and a pipeline that I wrote here

Parameters

Flag	Description	Default
--PLINK_prefix	Path to genome-wide PLINK binaries (.bed/.bim/.fam) [required]	NA
--PLINK_prefix_chr	Path to per chromosome PLINK binaries (.bed/.bim/.fam) [required]	NA
--weights	Path for .pos file describing features [required]	NA
--weights_dir	Directory containing the weights listed in the .pos file [required]	NA
--ref_ld_chr	Path to FUSION 1KG reference [required]	NA
--score_files	Path to SCORE files corresponding to weights [optional]	NA
--ref_expr	Path to reference expression data [optional]	NA
--n_cores	Specify the number of cores available for parallel computing [optional]	1
--memory	RAM available in MB [optional]	2000
--plink	Path to PLINK software [required]	NA
--save_score	Specify as T if temporary .SCORE files should kept [optional]	TRUE
--save_ref_expr	Save reference expression data [optional]	TRUE
--output	Name of output directory	NA
--pigz	Path to pigz binary [required]	NA
--targ_pred	Set to FALSE to create SCORE file and expression reference only [optional]	TRUE
--ref_maf	Path to per chromosome PLINK freq files [required]	NA
--chr	Specify chromosome number [optional]	NA

Output files

In the specified output directory, the following files will be produced:

Name	Description
FeaturePredictions_<PANEL>_chr<chr>.txt.gz .csv	Space delimited file containing FID, IID, and the predicted values for each feature.
FeaturePredictions.log	Log file.
SCORE_failed.txt	Text file listing weights which couldn't be converted to a .SCORE file (if any).
Prediction_failed.txt	Text file listing features that couldn't be predicted (if any).

Examples

These examples use the weights and .pos file provided here.

When using default settings:

Rscript FeaturePred.V2.0.R \
	--PLINK_prefix_chr FUSION/LDREF/1000G.EUR. \
	--weights test_data/CMC.BRAIN.RNASEQ/CMC.BRAIN.RNASEQ.pos \
	--weights_dir test_data/CMC.BRAIN.RNASEQ \
	--ref_ld_chr FUSION/LDREF/1000G.EUR. \
	--plink plink \
	--ref_maf FUSION/LDREF/1000G.EUR. \
	--pigz pigz \
	--chr 1 \
	--output demo

Running in parallel on cluster:

sbatch -p shared -n 6 --mem 50G Rscript FeaturePred.V2.0.R \
	--PLINK_prefix_chr FUSION/LDREF/1000G.EUR. \
	--weights test_data/CMC.BRAIN.RNASEQ/CMC.BRAIN.RNASEQ.pos \
	--weights_dir test_data/CMC.BRAIN.RNASEQ \
	--ref_ld_chr FUSION/LDREF/1000G.EUR. \
	--plink plink \
	--ref_maf FUSION/LDREF/1000G.EUR. \
	--pigz pigz \
	--chr 1 \
	--output demo \
	--n_cores 6 \
	--memory 50000

Help

This script was written by Dr Oliver Pain ([email protected])

If you have any questions or comments use the google group.