Most heritability analyses require a large number of unrelated samples. For example, to reliably estimate SNP heritability (standard deviation less than 5%) typically needs at least 7,000 unrelated samples. If your dataset has fewer than 5000 samples, or contains substantial relatedness (or population structure), then you will not be able to perform heritability analysis, sorry. However, you may instead be able to perform single-SNP or gene/chunk-based association analysis. See Small Datasets for more details.
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
We have spent much time investigating the best Heritability Model. In general, we recommend that you use the LDAK-Thin Model when analysing individual-level data, and the BLD-LDAK Model when analysing summary statistics. Note that the BLD-LDAK Model can only be used with human SNP data, so if using non-human or non-SNP data, you should always use the LDAK-Thin Model. We explain our recommendations in Publications, while you can find out how to implement these models in Heritability Model.
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
When analysing summary statistics, SumHer provides the option to allow for confounding bias. However, estimates of confounding bias are based on an assumption that the inflation of association study test statistics caused by relatedness or population structure is constant across predictors. This assumption is hard to test, and it is easy to find situations where it is violated. Therefore, we recommend that you do not use this feature, and that you only use summary statistics from association studies that performed strict quality control.
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
There are help pages for most of the LDAK functions. Furthermore, Protocols contains step-by-step scripts for a range of analyses, while there are practical slides within Short Course. Nonetheless, I think the easiest way to learn LDAK is by using the software. Once you have have LDAK running (see Downloads for instructions), simply specify the main argument, and LDAK will tell you what options are required.
For example, suppose you are using the Linux version of LDAK, and you wish to perform single-SNP linear regression. Main Arguments says you should use --linear <outfile>, so you can start by trying
./ldak5.1.linux --linear results
Note that arguments must be in pairs, with the main argument followed by the output name (here, the output files will have prefix results). This command will fail, with the message Error, you must use "--pheno" to provide phenotypes. Therefore, if your phenotype file is called phen.pheno, you would then try
./ldak5.1.linux --linear results --pheno phen.pheno
This command will also fail, with the message Error, you must provide a set of genetic data files using "--bfile", "--sp", "--sped", "--speed" or "--gen". Suppose your data are in Binary PLINK format, with prefix data, then you can try
./ldak5.1.linux --linear results --pheno phen.pheno --bfile data
This works, and LDAK performs single-SNP linear regression, testing each SNP the data for association with the phenotype, and saving the output in files with prefix results.
