Download
Examples
Conditional test tutorial
Usage
Warnings
Future developments
Licence
Citation
Contact
Warnings
- Check the data thoroughly before running whap, e.g use PEDSTATS or similar.
- Exclude rare haplotypes from analysis to improve numerical stability
- WHAP is not suited to very large datasets, where large here means more than a few thousand SNPs on a thousand or more individuals. In particular, WHAP is not to whole genome association analysis, due to memory and speed issues (see PLINK for software that is).
- Running time for the example dataset of an omnibus test for a 7 SNP haplotype on 300 individuals for a quantitative trait is around 0.8 of a second to phase and perform the association test (on a 3Gz Linux workstation). To perform 100 permutations on this same dataset takes around 1 minute.
- In addition to not handling datasets with many thousands of SNPs, WHAP should not be used to attempt to phase more than a dozen or so SNPs at a time. The precise number will depend on the LD structure of the region (i.e. how many common/rare haplotypes there are). As a rule of thumb, most analyses should probably try to look at 10 or fewer SNPs in any one window resultsing in 10 or fewer common haplotypes.
- Run whap with the --repeat N option where N is set to some value such as 50 or 100 to check the numerical stability of important results. This option will slow performance, but increase the chance of convergence at a global minimum, especially for models with many parameters.
- Quantitative traits and covariates should be one an approximately standard normal scale (i.e. a mean near 0 and a variance near 1) -- traits with order-of-magnitude different means and/or variances are likely to show problems with numerical stability
- Fix the trait mean and variance or prevalence where appropriate
- Use of permutation is necessary to obtain significance values for sliding window analyses
- Use the --cond, --prev and --model w options combined to perform a TDT-like test
- Remember that including covariates when performing conditional (--cond) analyses can cause problems
Created by Shaun Purcell; Last updated by Lori Thomas: March 2006