SparSNP fits lasso-penalized linear models to SNP data. Its main features are:
- it can fit squared hinge loss for classification (case/control) and linear regression (quantitative phenotypes)
- takes PLINK BED/FAM files as input
- the amount of memory is bounded - can work with large datasets using little memory (typically <1GB, more for better performance)
- fits a model over a grid of penalties, and writes the estimated coefficients to disk
- it can also do cross-validation, using the estimated coefficients to predict outputs for other datasets
- efficient - it uses warm-restarts plus an active-set approach, the model fitting part of 3-fold cross-validation for a dataset of 2000 samples by 300,000 SNP dataset takes ~5min, and about 25min for ~6800 samples / ~516,000 SNPs
Gad Abraham, email@example.com
G. Abraham, A. Kowalczyk, J. Zobel, and M. Inouye, ``SparSNP: Fast and memory-efficient analysis of all SNPs for phenotype prediction'', BMC Bioinformatics, 2012, 13:88, doi:10.1186/1471-2105-13-88
This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 3 of the License, or (at your option) any later version.
Copyright (C) 2011-2012 Gad Abraham and National ICT Australia (http://www.NICTA.com.au). All rights reserved.
For the post-analysis scripts: R packages ggplot2 >=0.9.3, scales, grid, abind, ROCR
A 64-bit operating system is recommended; we have tested SparSNP on 64-bit OSX and Linux.
To get the latest version:
git clone git://github.com/gabraham/SparSNP
cd SparSNP make
Run (assuming a PLINK BED/BIM/FAM dataset named MYDATA, i.e. MYDATA.bim)
export PATH=<PATH_TO_SPARSNP>:$PATH crossval.sh MYDATA sqrhinge 2>&1 | tee log eval.R
Documentation: see the document https://github.com/gabraham/SparSNP/blob/master/workflow.pdf
Changelog: see https://github.com/gabraham/SparSNP/blob/master/CHANGELOG
- Marco Colombo, patches for consistent lambda1 path