This second kind of analysis is done by lining up lots of known binding sites for a particular protein, comparing them position by position, and so finding out which letter is most likely to occur at which position, and how probable it is that a different letter may sometimes crop up instead.