A Pipeline for Predixcan

To know about PrediXcan, please refer the link.

Key things to know before using this script

The script is only compatible with

1. the imputation dosage files produced by ricopilli pipeline (link) and it assumes a file structure derived from ricopilli pipeline.

2. a linux environment with SLURM job sheduler

This script does four things

1. Converts genotype dosage format (refer here for more details ) to allele dosage format required by Predixcan

2. Predicts the gene expression using Predixcan software.

3. Runs an association analysis between the predicted expression data and a phenotype.

4. Creates a manhattan plot of the results obtained from the step 3.

Requirements

1. Python ≥ 2.7 and numpy package.

2. R and packages(data.table, doParallel, foreach, qqman) Note: the script will install the packages if not installed already

3. Linux environment

Installation

Clone the repository and run the install.sh script once.

git clone https://github.com/veera-dr/Predixcan.ipsych.git 
cd predixcan.ipsych
sh install.sh

The install.sh script appends an alias to your bashrc file, so you can call the script by predixcan

You can test if installed by

predixcan -h 

Usage instructions

The help info from the script is detailed and you can jump start from there quickly.

Extraction

Usage:

predixcan --extract --chunkdir <value> --outdir <value> 

Argument	Value
--outdir	full path to the folder where you need the outputs
--chunkdir	full path to the 'qc1' folder containing the dosage chunks

Note: the script by default will look for the 'info' folder in the same directory where 'qc1' folder is located. This is for quality filtering based on infoscore thresholds. You can skip this filtering step with --noinfo argument

Details

This will extract a subset of hapmap variants (predixcan requires only subset of variants for prediction) from the dosage files and will convert them to allele dosage format required by predixcan software. In addition, the script also filters out the variants with info score < 0.60 (you can change this with --infoscore argument) before conversion. The converted dosage files will be placed in a folder named "dosages" inside the outdir you specify.

Prediction

Usage:

predixcan --predict --outdir <value> --weights <value> 

Argument	Value
--outdir	full path to the folder where you need the outputs
--weights	full path to the 'weights' folder containing the '.db' files

Note: Be default the script looks for folder "dosages" inside the output folder (outdir) that you specify. You can override this with argument --dosagefolder. Inside dosage folder there should be a file with name 'samples.txt', containing FIDs and IIDs in the same order as in the dosage files. One of the fam files inside qc1 folder is copied to this folder during the extraction step.

Details

The script will use the Predixcan.py (link) script to predict the gene expressions using the genotype data. The --weights argument requires a path to a folder (not a file) containing pre calculated weights for various tissues. The files have .db extension and more details can be found in Predixcan page (link).

Association

Usage:

predixcan --assoc --outdir <value> --pheno <value> 

Argument	Value
--outdir	full path to the folder where you need the outputs
--pheno	full path to the pheno file (same as plink format)

Note: the script by default looks for a folder 'expression' inside the output folder(outdir) that you specify. The expression folder should contain the predicted expressions (one to many), which are the output of the prediction step. You can override this with --expression argument pointing to a folder with predicted expressions located elsewhere.

Details

The script will run an association analysis using the PrediXcanAssociation.R Rscript. The Rscript is modified from the actual version. By default the Rscript will run a logistic regression analysis with the phenotype. The phenotype should be coded as 0 and 1 (not 1 and 2). In case, if you are using a continuous parameter as a phenotype, just add --linear tag in the command.

Manhattan plot

Usage:

predixcan --manhattan --outdir <value>

Argument	Value
--outdir	full path to the folder where you need the outputs

Note: The script by default looks for a folder 'results' inside the output folder (outdir) that you specify. You can override this using argument --results with the path to a folder (not a file) containing the results.