RNA secondary structure prediction using deep learning

RNA secondary structure plays an important role in governing RNA’s properties and functions. Experimental assays for detecting RNA secondary structure are very expensive and time consuming. For this reason, computational prediction provides a practical alternative. However, traditional prediction approaches, including physics based approaches and machine learning based approaches, all heavily rely on manually designed features.

We propose a novel data-driven deep learning based approach. This approach utilizes deep learning sequential model LSTM to automatically extract RNA structure features, and utilizes structured SVM for training. It is the first approach to introduce deep learning technique into RNA secondary structure prediction problem. We compared our model with current best model CONTRAfold on a well-known dataset. Our model achieves better prediction accuracy.

Major Advisor: Liang Huang
Committee: Prasad Tadepalli
Committee: Amir Nayyeri