This code has been deprecated and exists solely for reference and documentation.
The model prediction pipeline (with nested cross validation) is available in github here, and a brief overview is available here (PrediXcan google group).
This directory can be used for creating models for PredictDB with GTEx data.
These models can be used to predict gene expression in 44 different tissues, based on individual genotype. To train the models, we used version 6p of the GTEx data and applied 10-fold cross validation elastic-net within SNPs that are located within 1mb of the gene as features.
The pipeline takes 4 different sets of inputs (all in the data/inputs/ directory, possibly in the form of symbolic links), does some preprocessing, stores the results in the data/intermediate/ directory. Once this is all done, we can apply our machine learning technique, output the results in the data/output/ directory. There's also a script to turn this data into SQLite databases, to be used by PrediXcan.
With some modifications, this pipeline can be extended to analyze other studies which have genotype and expression data.
See our wiki for a tutorial with example files and a detailed description of these scripts.