Prediction of transcription factor binding to DNA using rule induction methods

In this study, we seek to develop a predictive model for finding the strength of binding between a particular transcription factor (TF) variant and a particular DNA target variant. The DNA binding paired domains of the Pax transcription factors, which are our main focus, show seemingly fuzzy and degenerate binding to various DNA targets, and paired domain-DNA binding is not a problem well suited for previously proposed algorithms. Here, we introduce a simple way to use rule induction for predicting the strength of TFDNA binding. We have created a dataset consisting of 597 example cases for paired domain-DNA binding by collecting information about all published and quantified interactions between TF and DNA sequence variants. Application of the rule induction based method on this dataset yields a high, although far from ideal accuracy of 69.7% (based on cross-validation), but perhaps more importantly, several useful rules for predicting the binding strength have been found. Although the primary motivation for introducing the rule induction based methods is the lack of efficient algorithms for paired domain-DNA binding prediction, we also show that the method can be applied with some success to a more well-studied TF-DNA binding prediction task involving the early growth response (EGR) TF family. Summary The transcription of DNA into mRNA is initiated and aided by a number of transcription factors (TFs), proteins with DNA-binding regions that attach themselves to binding sites in the DNA (transcription factor binding sites, TFBSs). As it has become apparent that both TFs and TFBSs are highly variable, tools are needed to quantify the strength of the interaction resulting from a certain TF variant binding to a certain TFBS. Ideally, one would like to have a method where any combination of TF amino acids are allowed to interact with any TFBS nucleotide, and vice versa. Rule induction algorithms might be such a method. We used a simple way to predict interactions between protein and DNA: given experimental cases from the literature where the interaction strength between two sequences has been quantified, we created training vectors for rule induction by regarding each amino acid and nucleotide position as a single feature in the example vector. The resulting interaction strength was used as the target class or value. These training vectors were then used to build a rule induction model. We applied the rule induction method to two protein families – transcription factors from the Pax and the early growth response (EGR) families – and their corresponding DNA targets.