Comparing the Performance of Several Popular Machine Learning Algorithms on Classifying TATA-box from putative TATA box

A TATA box is a common transcription binding site that occurs in the upstream of a transcription start site of many genes. Identifying a TATA box accurately is important since it has been shown empirically that a transcription start site (TSS) occurs in the downstream of a TATA box after a fixed distance that is only dependent on the species. Unfortunately, many substrings of a DNA sequence fit to the profile of a TATA box and such substrings are called putative TATA boxes. Identification of a TATA box among putative TATA boxes will improve the accuracy of determining a TSS and hence the detection of a promoter. In this paper we investigate the effectiveness of several popular machine learning algorithms for discriminating TATA box from putative TATA boxes. These algorithms include naïve Bayes, artificial neural network, decision tree C4.5, random forest and support vector machine. To compare the effectiveness, we use a metric of prediction accuracy, true and false positive. Empirically we have shown that tuned support vector machine has outperformed all other machine learning algorithms