Research Of Sequence Specificities Based On Convolutional Neural Network

Predicting the sequence specificities of genes is essential for gene analyzing and gene regulatory. The sequence specificities of DNA-and RNA-binding proteins are used to identify causal disease variants. However, the developments of molecular biology brings explosive growth of biological data. It is really hard to discover some useful message through traditional biological experiments or statistics, and it costs to much.Convolution neural network is a deep learning technology emerging in recent years. It has been widely used in image recognition, speech recognition, and natural language processing. This kind of technique based on multi-layer perceptron, which has been proposed by Rumelhart, Willams, Hinton and LeCun, aimed to simulating the information processing of human being’s neurons, and use the neural networks to represent some complex rules(in some functional way) of real world. But due to the naive structure, the performance is not so good. To address this problem, we need more layers and more functions, that leads to convolution neural network.In this article, we propose a predict model based on the state-of-the-art deep learning technique, convolution neural network. By learning from the real world, it can performance well in predicting sequence specificities. There are several stages in this process:Firstly, transform the sequences to an encoding matrix, secondly, convolve this matrix using a matrix kernel which has been encoded by the motif pattern, and abstracts features from sequences automatically. Then rectified by some given threshold, pooling at the same time. Finally, a score will be output.