Fig. 2
From: Modeling binding specificities of transcription factor pairs with random forests
A Sub-sequences with maximum PWM scores are chosen from the DNA double-strands according to orientations 1 and 2 (TF1 + TF2 and TF2 + TF1, respectively). PWMs are extended until 25 nucleic acids to cover binding of both TFs. Within an orientation, ComBind scores DNA double-strands with two PWMs and selects the sub-sequence with the highest per-position score. B Forests are trained with the chosen sub-sequences separately for both orientations. C ComBind scores unseen DNA double-strands with both RFs. Maximum per-position scores are considered as the orientation specific scores, whose average ComBind outputs as the sequence score. [PWM visualization [38], and PWM structures [7].]