Schematic of LogoAlign "hill-climbing" methodology. A. Each of the ten promoter subfamilies is aligned in order to produce one input Logo per promoter subfamily. B. A random starting position of motif-sized (12-bp) windows is established for each of the input Logos, and total information is calculated based on the alignment of these windows. C. Beginning with a randomly selected Logo, the window is moved to every possible position in the Logo, and for each position, the total information is calculated based on the alignment of this window to all of the other stationary windows. The window is moved to the position at which the total information is maximized (red). D. In a randomly determined order, each Logo undergoes exactly the same window-sliding procedure as described in C above, and each time, the window is set to the position where information is maximized. One cycle is completed after every Logo has undergone one window-sliding procedure. E. Multiple cycles are executed, each time using the final resting positions from the previous cycle as starting positions for the new cycle, and each time using a random order of cycling through the ten Logos, until a complete cycle is completed during which none of the windows are moved. The resulting motif (a local maximum of total information) is kept as the result of one trial (2000 trials were conducted in this study).
Stewart and Lane BMC Genomics 2007 8:253 doi:10.1186/1471-2164-8-253