A new approach to the assessment of the quality of predictions of transcription factor binding sites

In this paper, we describe a novel method called Secondary Verification which assesses the quality of predictions of transcription factor binding sites. This method incorporates a distribution of prediction scores over positive examples (i.e. the actual binding sites) and is shown to be superior to p-value, routinely used statistical significance assessment, which uses only a distribution of prediction scores over background sequences. We also discuss how to integrate both distributions into a framework called Secondary Verification Assessment method which evaluates the quality of a model of a transcription factor. Based on that we create a hybrid representation of a transcription factor: we select the description (with or without dependencies) which is best for the transcription factor considered.