PrediXcan is a command-line tool that predicts gene expression from genotype data and performs gene-based association tests, allowing researchers to prioritize genes that are likely to be causal for a phenotype.

Gamazon ER†, Wheeler HE†, Shah KP†, Mozaffari SV, Aquino-Michaels K, Carroll RJ, Eyler AE, Denny JC, Nicolae DL, Cox NJ, Im HK*. (2015) A gene-based association method for mapping traits using reference transcriptome data. Nat Genet. doi:10.1038/ng.3367.

† equal contribution

An open access preprint can be found on BioRxiv

To run PrediXcan you will need

Software Requirements:

• Linux or Mac OS
• Python 2.7
  • numpy package
• R

Scripts:

• PrediXcan.py
• PrediXcanAssociation.R

Input Files:

• genotype file
• sample file
• transcriptome prediction model (sqlite db to be downloaded from PredictDB.)
• phenotype file
• filter file - Specifies a subset of rows on which to perform association tests (optional)

To predict the transcriptome from a given genotype file, include the --predict flag when running PrediXcan.py and specify the following arguments:

1. genelist: list of genes. Optional. By default it will use all available genes in the model database
2. dosages: imputed genotype file path. Default value: 'data/dosages/'
3. dosage_prefix: prefix of dosage file. Default value: 'chr'
4. weights: full path to the prediction model database. Default value: 'data/weights.db'
5. output_prefix: prefix for output files. This includes the path to the output files as well as the prefix for the file name

• Columns are chromosome rsid position allele1 allele2 MAF id1 ..... idn.
• Dosage for each person refers to the number of alleles for the 2nd allele listed (between 0 and 2).
• It is expected that there will be one file per chromosome.
• Files must be gzipped, and their names are expected to end with ".gz"
• In the dosages directory, there must be a file of the individuals with id #'s listed in the same order as the genotype columns.
  • The first column must contain the family ID, and the second must contain the individual ID.
  • If the the family ID is unavailable, it is ok if the individual ID column is copied over to the FID
  • The remaining columns of the sample file are not used in the creating the output, so it is possible to have a file with only two columns, but a PLINK .fam file is also an acceptable format for the samples file.

./PrediXcan.py --predict --dosages dosagefile_path --dosages_prefix chr --samples samples_file --weights prediction_db --output_prefix results/tissue

To perform an association test between the predicted expression levels and phenotype, include the --assoc flag when running PrediXcan.py and specify the following arguments:

1. pred_exp: predicted transcriptome from a previous run of PrediXcan. Default value: 'predicted_expression.txt'.
2. pheno: phenotype file. No default value. See below for file format.
3. filter: filter file to specify which rows to include in test and a number to filter on. Optional. See below for details.
4. linear or logistic: specify one of these to perform a linear or logistic regression between the expression levels of each gene and phenotype. Default is linear.
5. output_prefix: prefix for output files. This includes the path to the output files as well as the prefix for the file name

This will produce a file with suffix association.txt, containing summary statistics on the association between each gene and the phenotype.

Phenotype files are expected to be in a format similar to the format required for PLINK. Most commonly, the phenotype file is tab delimited, and preferably has a header. By default, PrediXcan will assume the first column is the Family ID, the second column is the Individual ID, and the last column is the phenotype column.

Note: If the phenotype file has a header line, which preferably it will, the first two columns must be labeled FID and IID, respectively. If there are multiple phenotype columns, you can specify which column to perform the association on with the --pheno_name flag.

If there is more than one phenotype column in the file, you can specify which phenotype to perform the association on with the --mpheno option. For example --mpheno 1 will do the association with the 3rd column in the phenotype file, as columns 1 and 2 are ID numbers, --mpheno 2 does the association on 4th, etc. This option will mainly be used for when there is no header line, and may behave unexpectedly if the user does not specify options carefully.

By default, PrediXcan performs a linear regression for association tests, and assumes quantitative traits in the phenotype file.

Unlike PLINK, for logistic tests on qualititative traits, by default the trait is assumed to be encoded as 0 for unaffected and 1 for affected. 0 is NOT a missing value.

By default, NA specifies a missing phenotype value. To specify a missing phenotype value that is encoded numerically, say -9 for example, include --missing_phenotype -9.

If a logistic test is specified and there are more than two levels of the phenotype, the user will recieve an error.

Filter files can specify a subset rows in the pheno file to perform the association on. It is a tab delimited file with the first 2 columns identical to the pheno file. The third column holds numerical values on which to filter. If the filter file is called filter.txt, with filter values 1 and 2, including --filter filter.txt 2 will perform the association test only on individuals marked 2 in the filter file.

Header rows are optional for the filter file, but if they are included, the first two columns must be labeled FID and IID.

./PrediXcan.py --assoc --pheno phenotype_file --pred_exp predicted_expression_file --linear --filter filter_file filter_val --output_dir output_dirExample for Prediction and Association

• Download and untar/unzip this file PrediXcan Example tar file
• Go to folder and run the following

./PrediXcan.py --predict --assoc --linear \
               --weights weights/TW_Cells_EBV-transformed_lymphocytes_0.5.db \
               --dosages genotype \
               --samples samples.txt \
               --pheno phenotype/igrowth.txt \
               --output_prefix results/Cells_EBV-transformed_lymphocytes 

Conversion from Plink to Dosage (provided by scottritchie73 via pull request, thank you!) link