This distribution consists of three parts. One, a program (sampleSNPs) that creates and saves ordered random samples of SNPs from a variety of formats. Two, a program (sampleLD) that calculates linkage disequilibrium (LD) among randomly chosen (without replacement) pairs of SNPs from a binary variant format file. Three, a library (libsampFiles.a) that allows users to build similar applications taking advantage of the fast sampling algorithms used in the two programs mentioned above.
The software is build for Unix-like systems. Neither compilation nor running was checked under Windows and there are reasons to believe it will not compile on that OS. There are no dependencies other than a compiler that understands the C++11 standard. Random number generation employs the RDRAND CPU instruction if supported by the processor, otherwise an implementation of the 64-bit Mersenne Twister \cite matsumoto98a is substituted. Intel Ivy Bridge or later support RDRAND. With AMD it is not completely clear. Opteron definitely does not support it. Zen architectures (Ryzen) claim to support it, but I did not have access to one so I cannot personally vouch for it. The RNG choice is made automatically at run time.
The simplest way to install everything is to run
make all
sudo make install
in the directory with the source code. This will install the executables (sampleSNPs and sampleLD) in /usr/local/bin/ and the library in the appropriate folders in /usr/local/. The included Makefile can be modified to change where things go. The headers varfiles.hpp, populations.hpp, and random.hpp have to be included in your code as necessary.
The tests/ directory contains example .bed, .tped, .vcf, and .hmp.txt files to try running the programs on. To keep sizes manageable for distribution, the .bed file has 50,000 loci, while the text files have only 5,000. Make sure your samples do not exceed these values. Uncompress the directory and run, for example,
./sampleSNPs -i tests/sample -t BED -s 5000
This should sample 5,000 SNPs from the included sample_ALL.bed and sample_ALL.bim files and save the results into files with the sample_ALL_s5000 prefix.
Note that sampleLD supports only the .bed format.
Running sampleSNPs and sampleLD without flags will cause these programs to print flag descriptions and exit.
sampleLD calculates LD between sampled pairs of loci. The paper cited below gives the details. The flag -s controls the number of locus pairs picked for estimates, and as this number increases, the average distance between pairs gets smaller. This leads to an increased precision of LD estimates between SNPs close to each other, at the expense of undersamling and hence diminished accuracy of disequilibrium calculations between distant loci. The user should keep this trade-off in mind when selecting the sample size. A further increase in LD estimate precision between distant variants can be achieved if one collects several sparse samples. However, care must be taken to eliminate redundant locus pairs because sampling without replacement is not guaranteed in this case.
An optional population index file can be provided after the -p flag. This file should contain space-delimited integer values that relate each individual in the .fam file to a population. An example file is in the archived tests directory.
Expanding the timingTrials.tar.gz archive will generate a directory with separate software, depending only on random.cpp and .hpp, that performs analyses of execution time using Vitter's Method D and Method S. The README.md file included there explains how to compile and run these analyses.
The paper describing this work is available, and can be referenced, from bioRxiv and arXiv.