Deep learning for biological sequences

Nowadays, with the increase in biological knowledge, the use of deep learning in bioinformatics and computational biology has increased. Newly, deep learning is widely used to classify and analyze biological sequences. In recent years, deep neural network architectures such as Convolutional and Recurrent Neural Networks have been developed in order to achieve more successful results when compared to classical machine learning algorithms. In this thesis, the discussed problem is a bioinformatics problem. Therefore, it is discussed whether the given microRNA molecule binds to the mRNA molecule. MicroRNAs (miRNAs) are non-coding and small RNA molecules of ~23 base length that play an important role in gene expression cycle. After transcription, they bind to target mRNAs and cause mRNA cleavage or translation inhibition. Rapid and efficient determination of the binding sites of miRNAs is a major problem in molecular biology. In this thesis study, Long Short Term Memory (LSTM) network which is based on deep learning, has been developed with the help of an existing duplex sequence model. The study provides a comparative approach based on different data sets and configurations. In addition, a web tool has been developed to effectively and quickly identify human microRNA target sites and provide a visual interface to the end-user. Compared to the six classical machine learning methods, the proposed LSTM model gives better results in terms of some evaluation criteria.