Statistical Model for Predicting TALEN-DNA Binding Sites Based On Moving Average

Abstract

In this paper, we propose a new approach to the in-silico prediction of any possible DNA binding sites for the user-defined artificial TALENs. This approach based on the exponential moving average model and developed as an online service TANDIS. The direct validation of our prediction model based on the direct matching with the known results of the certain in-vitro experiments, while for the verification of its accuracy we use comparative analysis against other similar popular services like TALE-NT and TALENoffer. So thus, we have found out that the exponential moving average model brings very good results comparable with those of the Markov chain model used in TALENoffer, but TANDIS can do it much more easily because its model is much simpler. The TALE-NT prediction is even faster than ours for it has an utmost simple position-independent scoring system and drastically simplified filtering rules for the case of paired TALEs, which makes however, on the other hand, the results of such TALE-NT 's prediction much less competitive. Besides being the compromise between accuracy and efficiency, the exponential moving average model has only five parameters, so in future, it could be easily used for more intense prediction, and probably later, it can be used to cast some light on our understanding of real physical principles of the attractive interaction between a certain TALE and a random DNA site.