Promoter Recognition Based On The Maximum Entropy Hidden Markov Model

With the completion of human genome sequencing, one of the current research in bioinformatics is to recognize promoter, which play an important role in gene regulations. At present, there are two kinds of promoter recognition method:the content-based method and the signal-based method. Content-based methods such as hidden Markov model(HMM) often bring some signal noise and ignore the biological features. Signal-based methods overcome the shortages above and utilized the biological features of the promoter for the condition. However, it leads to a high false positive rate (FPR). In order to overcome the shortages above and improve the recognition performance constantly, we introduce two new methods in this paper:One is based on the maximum entropy Markov model (MEMM), another is based on the maximum entropy hidden Markov model (ME-HMM).The new methods have some innovations as follows:First, both of the methods are applied in the field of promoter recognition for the first time; Second, the two methods are both accommodate the biological features and establish feature templates to select features; Last but not the least, in order to apply the new techniques to promoter recognition, we improved forward algorithm to develop the MEMM-forward algorithm and MEHMM-forward algorithm.Furthermore, ME-HMM method has its own innovations:First, it utilized motif-based HMM instead of profile HMM. Second, it is not only overcome the shortages of content-based methods but also reduce the FPR effectively. This paper build the two model and implemented the algorithm by R language for the first time. To demonstrate the precision, we introduced hidden Markov model for comparison. The experimental results indicate that, the new methods performed much better than HMM with less training sets.