Sequence-based prediction of DNA-binding sites on DNA-binding proteins

Motivation: Identification of DNA-binding sites on DNA-binding proteins is important for functional annotation. Experimental determination of the structure of a protein-DNA complex is an expensive process. Reliable computational methods that utilize the sequence of a DNA-binding protein to predict its DNA-binding interface are needed. Results: We present an application of three machine learning methods: support vector machine, kernel logistic regression, and penalized logistic regression to predict DNA-binding sites on a DNA-binding protein using its amino acid sequence as an input. Prediction is performed using either single sequence or a profile of evolutionary conservation. The performance of our predictors is better than that of other existing sequence-based methods. The outputs of all three individual methods are combined to obtain a consensus prediction. This further improves performance and results in accuracy of 82.4%, sensitivity of 84.9% and specificity of 83.1% for the strict consensus prediction. Availability: http://lcg.rit.albany.edu/dp-bind