Context-aware semi-supervised motif detection approach

Motif detection has raised as an important task in bioinformatics. Recently, the discovery of motifs that are localized relative to a certain biological area has become an important task in many applications. For example, it is used to discover regulatory sequences beside the transcription start site and the neighborhood of known transcription factor binding sites [1]. Therefore, the idea of context aware motif detection approach is needed. Moreover, there is an interest to use both labeled and unlabeled sets to enhance the motif detection approaches. In this paper, three novel context aware semi-supervised motif detection approaches are proposed, which are self-learning, context aware and co-training context aware systems. In self-learning motif Hidden Markov Model (HMM) is enhanced independently using unlabeled sets. While in co-training, three different models are trained based on three different views which are pre-motif sequences, motif sequences and post-motif sequences. Moreover, our co-training context aware system is suitable for parallelization to enhance its execution time. The approaches were evaluated using human motif sequences and the results show that co-training context aware system has achieved the best results. The results also show that our approach outperforms other related works in [1], [2] and [3].