where w0 is the PWM score of the major allele, and wi and pi are the score and frequency of each allele, respectively. Classically, genetic load is expressed with respect to the maximum observed value (w0 = wmax). However, we have instead chosen to express it relative to the major allele (w0 = wmaj). The main reason for this is that, in the absence of ChIP data for each individual or isogenic line, TFBSs whose minor alleles have a higher PWM score than the major allele are subject to a significant ascertainment bias. Indeed, only TF-bound TFBS instances are included in the analysis, and we are much more likely to detect TFBSs as 'bound' when their weaker major alleles are also strong enough to ensure TF binding. Additionally, for reasons explained in the main text, we have postulated that TFBSs with stronger-scoring minor alleles have a zero load irrespective of frequency. Using the human data presented an additional challenge of interpreting heterozygous genotypes. Since the immediate phenotypic trait associated with TFBS's match to consensus (that is, TF binding) occurs in
