a SNP may need an additional polymorphism(s) to explain the phenotype (multiple genes hypothesis), while the additional genetic variation(s) may not be present in a particular ethnical population. One of the practical problems in dealing with low-MAF SNPs concerns the sample size necessary to obtain reliable association data, i.e. the lower the MAF the more samples are required to reach statistical significance. It is likely that inclusion of potentially functional SNPs (non-synonymous SNPs and SNPs in gene promoter regions or exon-intron boundaries) will increase the success rate in the analysis. At present, chips containing 20,000 non-synonymous human SNPs from ~11,000 genes are already available (www.affymetrix.com).
