Glide companion scripts & tutorial

Summary

These Python scripts are provided for basic tasks around the usage of the GLIDE package for the detection of epistasis in GWAS data.

T. Kam-Thong, C.-A. Azencott, L. Cayton, B. Puetz, A. Altmann, N. Karbalai, P. G. Saemann, B. Schoelkopf, B. Mueller-Myhsok, and K. M. Borgwardt. (2012) GLIDE: GPU-Based Linear Regression for Detection of Epistasis. Human Heredity, 73 (4), 220-236 doi: 10.1159/000341885

They were mostly developed at the MLCB (Machine Learning for Computational Biology) research group of the Max Planck Institutes in Tuebingen (Germany). Thanks to Damian Roqueiro for stress-testing them and providing useful feedback!

Documentation

The documentation, which contains additional CLI instructions and short scripts, is available under documentation/HowToGLIDE.pdf

Compile the documentation with pdflatex -shell-escape documentation/HowToGLIDE.tex

Preprocessing scripts

• qqplot.py: Plot a Q-Q plot for (single SNP) p-values, from PLINK output.
• naive_impute.py: Impute missing values in GLIDE input file, using the most frequent value for this phenotype.
• plink2h5.py: Convert a binary PLINK file into a .h5 file.
• transpose.py: Transpose a PLINK .raw file into a .glideIn file.
• split_glide.py: Write the batch file that will run GLIDE sequentially on the "tiles" created beforehand by splitting the input file.

Postprocessing scripts

• compute_all_pvalues.py: Compute all p-values corresponding to the t-test scorse output by GLIDE. Run compute_pvalues.py on all outputs and match indices back to SNP names.
• compute_pvalues.py: Compute the p-values corresponding to the t-test scorse output by GLIDE.
• compute_block_to_index.py: Compute the dictionary (stored in CONST.py) used to convert between tile IDs and SNP indices.
• CONST.py: Provide the dictionary necessary to the conversion between tile IDs and indices.
• rerun_confounder_h5.py: Run linear regression again on a small number of SNP pairs. Include confounders. Read genotypes from .h5 file (created with plink2h5.py).
• rerun_confounder.py: Run linear regression again on a small number of SNP pairs. Include confounders.
• rerun_h5.py: Run linear regression again on a small number of SNP pairs. Read genotypes from .h5 file (created with plink2h5.py).
• rerun.py: Run linear regression again on a small number of SNP pairs.
• utils.py: Utilities that can be reused by several other scripts.
• meta_analysis.py: An example of running a meta-analysis of the pairs of SNPs selected by running GLIDE on 4 different cohorts.

Contact

Any questions can be directed to Chloe-Agathe Azencott: chloe-agathe [dot] azencott [at] mines-paristech [dot] fr