Toolkit for ChIP-Seq based comparative analysis of the PWM methods for prediction of transcription factor binding sites

Despite wide application of the powerful ChIP-Seq technology for experimental identification of transcription factor (TF) binding sites, the computational prediction of the TF-binding sites is also relevant. Many methods for the prediction of the TF-binding sites have been developed over the last decades. Some of them represent position weight matrix (PWM) approach that is the most common and widely used. However, there exists little guidance in the choice among these methods because of a comprehensive comparison of existing methods is still challenging in practice. Thus, the direct use of the ChIP-Seq data for assessing predictive ability of the methods does not seem advisable because of such reasons as the tethered binding or false positive rates of peak detection algorithms. We have developed computational toolkit for reliable comparison of prediction methods under condition that unknown fraction of the ChIP-Seq data do not contain genuine TF-binding sites. On the base of developed toolkit, we have performed comparative analysis of three existing methods that represent PWM approach. The analysis has revealed that MATCH performed significantly worse than two other methods while common additive method outperformed others.