#----------------------------------------------------------------# # SDS code # # # # Version: pre-publication release, Sep 19 2016 # # # # Pre-print: # # Field, Boyle, Telis, ..., Pritchard # # Detection of human adaptation during the past 2,000 years # # http://biorxiv.org/content/early/2016/05/07/052084 # # # # Contact: Yair Field, yairf@stanford.edu # #----------------------------------------------------------------# #---------------------------------------# # List of files and their description # #---------------------------------------# - compute_SDS.R: This is the main script to compute SDS, implemented in R. - Makefile: This includes functions to do two things -- (1) preparing the gamma-shape parameters file, based on simulations. (2) running compute_SDS.R on the example input files. - backward.py: This python scripts wraps the simuPOP package to simulate allele frequency trajectories. - example.testsnp: The example input file with info on the test-SNP(s) - example.singletons: The example input file with info on location of singletons - example.observability: The example input file with specification of the "singleton observability" parameters - example.boundaries: The example input file with specification of the genomic boundaries of interest - example.gamma_shapes: The example input file with the "gamma-shape" parameters - example.output: The example output file #-------------------# # Getting started # #-------------------# First, you'll need to change the first line in compute_SDS.R to specify the correct path to 'Rscript' on your computer. Run "make run_example" to see it running on the example input files. You'll now need to prepare your input files. Run "compute_SDS.R -h" to see a help text with description of the file formats. The one file that might take you longer to prepare is the "gamma-shapes" file, because here you'll need to run some simulations. This will also require to download/install two simulation programs (ms & simuPOP). *** For a quick start, I suggest to just use the "example.gamma_shapes" file! *** The raw SDS will be better calibrated after you do these simulations to get the "gamma shape" estimates that correspond to your demographic model; but it will probably result with very similar standardized SDS scores (standardization will correct for the model-misspecification frequency-dependent biases). In the next release I plan to add pre-computed gamma-shapes files for several demographic models and sample sizes. To estimate these "gamma-shape" parameters yourself: (1) Download Hudson's "ms". Try here: http://home.uchicago.edu/rhudson1/source/mksamples.html (2) Download and install the simuPOP package. Try here: http://simupop.sourceforge.net (3) Modify backward.py: - Change the first line to specify the path to python on your computer. - Run "backward.py -h" to see a help text. - Several demographic models are already implemented. If you want another model you'll need to add a function for it - but it is straightforward (e.g., search "Tennessen_CEU" and follow from there). (4) Modify the Makefile's first lines: - Specify the path to the 'ms' simulation program. - Set the flag for backward.py to specify your demographic model. - Specify the present effective diploid population size. - Specify the list of derived allele frequencies for which you'd like to simulate the gamma-shape parameters. - Specify the sample size. - Specify the number of simulations. (5) Run "make sim". This will take a few hours to run (assuming ~1000 simulations for several frequencies). You can change the Makefile code to make it run in parallel (e.g., for different frequencies)... When you work on genome-wide predictions, it will be useful to run the predictions on different genomic regions in parallel. Rememebr that you'll need to standardize the scores within bins of derived allele frequecy. So even if you're interested in a single locus, you are advised to predict the raw SDS genome-wide (or at least chromosome-wide) and use that for standardization. For small samples, many SNPs do not have individuals from all three genotypes. I think this might have an effect on the distribution of raw SDS - and so this should probably be taken into consideration in the standardization step. I'm still looking at this, so I don't have yet a clear recommendation for how to do it. I'll soon release a script to compute SDS for phased (haplotype) data too. Good luck -- and please email me for questions, suggestions and bug reports.