Prediction of Transcription Factor Families Using DNA Sequence Features

摘要

Understanding the mechanisms of protein-DNA interaction is of critical importance in biology. Transcription factor (TF) binding to a specific DNA sequence depends on at least two factors: A protein-level DNA-binding domain and a nucleotide-level specific sequence serving as a TF binding site. TFs have been classified into families based on these factors. TFs within each family bind to specific nucleotide sequences in a very similar fashion. Identification of the TF family that might bind at a particular nucleotide sequence requires a machine learning approach. Here we considered two sets of features based on DNA sequences and their physicochemical properties and applied a one-versus-all SVM (OVA-SVM) with class-wise optimized features to identify TF family-specific features in DNA sequences. Using this approach, a mean prediction accuracy of ～80% was achieved, which represents an improvement of ～7% over previous approaches on the same data.