GMFR-CNN: An Integration of Gapped Motif Feature Representation and Deep Learning Approach for Enhancer Prediction

• Main Authors: Yu, Shiong Wong, Nung, Kion Lee, Norshafarina, Omar
• Format: E-Article
• Language: English
• Published: ACM New York, NY, USA 2016
• Subjects: T Technology (General)
• Online Access: http://ir.unimas.my/id/eprint/16666/1/GMFR-CNN%20%28abstract%29.pdf; http://ir.unimas.my/id/eprint/16666/; http://dl.acm.org/citation.cfm?id=3029380
• Institution: Universiti Malaysia Sarawak

Unravelling gene expression has become a critical procedure in bioinformatics world today and required continuous efforts to form a complete picture of enhancers. Enhancers are explicit patterns of gene expression that bound by activators to stimulate transcription. It could reside in upstream or downstream thousands of base pairs away without any fixed position. Therefore, the identification task of enhancers is extremely challenging. The inclusion of gaps in motif identification improved the overall accuracy and sensitivity, however, this feature is not fully utilised in deep learning method yet. Deep learning, is a powerful machine learning technique that has been actively used in image recognition and this technique has begun to shed light in bioinformatics. The expressiveness of deep learning enables higher feature learning from lower level ones. As a result, an integration of gapped motif feature representation (GMFR) and deep learning approach called deep convolutional neural networks (CNNs) is introduced to improve the predictive power of enhancer prediction. We called this method as GMFR-CNN. Comparative studies indicate that GMFR-CNN outperforms the other deep learning and gapped k-mer SVM tools with average 98% prediction accuracy. Breakthrough in deep learning technique certainly improves the performance in the near future.