ProSVM and ProK-means: Novel methods for promoter prediction

It is important to identify promoter regions to improve genome annotation and understand transcriptional regulation. In order to identify such regions accurately, transcription start sites (TSS) need to be identified correctly. Looking at the current genome annotation projects, it is not yet a common solution for the problem which is about identification of the transcription initiation regions. There are some drawbacks of the current methods which identify the core promoter regions. First, most of such methods require huge amounts of training data. Second, they are similar to black box methods, so output predictions are difficult to interpret. In this work, for identification of core promoter regions, we propose a supervised and an unsupervised method. We use support vector machines as a supervised method and k-means as an unsupervised method using physical properties of DNA sequences. Finally, we evaluate and compare our results with ProSOM [1] results. We show that ProSVM is able to achieve much higher recall rates compared to ProSOM and, therefore, is more accurate compared to ProSOM overall.