FIG. 3.

Effect of motif length and information content on predicting in vitro binding. (a) Performance as a function of motif length. For each PWM, we kept the k most informative contiguous positions and tested the ability of the resulting motif to predict in vitro binding. When the motif length was smaller than k, we used all positions. Average results of three criteria are shown in the graphs. (b) Performance as a function of IC cutoff. For each PWM, we removed all contiguous side positions with IC below the cutoff until reaching the first position with higher IC. The graphs show average results using the same three criteria.