Here we use three different approaches to take advantage of variation data in this context. First, we analyze TFBSs position-by-position to confirm that the levels of variation are generally consistent with TFBSs functional constraints predicted by their position weight matrix (PWM) models and highlight some intriguing exceptions. Next, we draw inspiration from Haldane's [31] and Muller's [32] genetic load model to devise a metric of TFBS variation that takes into account the loss of PWM match score associated with a mutation and makes it possible to investigate per-instance TFBS functional constraints. Finally, we take advantage of per-individual binding maps for a human transcription factor (CTCF) to highlight the 'buffering' of genetic variation at TFBSs at the level of binding, particularly in evolutionarily conserved regions.
