Supporting Information

Files in this Data Supplement:

Fig. 7. Parameters of four profile models learned from training motifs. Each of the eight histogram under arepresents the 4D parameter vector of a Dirichlet component (the height of the bar represents the magnitude of the corresponding element in the vector); vector pand matrix B are represented by color images, of which each element of por B specifies the color of a rectilinear patch in the image.

Fig. 8. Evaluation of position weight matrix (PWM) estimations by four different schemes. Cyan, symmetric-Dirichlet smoothing; green, maximum likelihood (ML); red, mixture of profile models; black, ML profile model out of the mixture. Motifs are listed along the x axis, ordered by the log likelihood odds of their PWM based on the "true" (according to their original family label) profile prior model.