Simple Encoding Of Secondary Structure Prediction

Bioinformatics is a new subject, which adopts computers to deal with and research into bioinformation. And protein secondary structure prediction has long been a fundamental problem of it and plays an important role. It is an important task for bioinformatical researchers to develop an efficient predicting method to get the more accurate results.The author first briefly introduces the basic content and fundamental problems in bioinformatics, the importance of bioinformatics and biological experiments as well as the research methods of them, and the mainstream of the development in protein secondary structure predicting and dimensional structure predicting methods. In the following theoretical part, the author mainly talks about the theory of artificial neural network modeling to predict protein secondary structure. What is focused on is how to better understand neural network, which is usually considered as having nothing to do with probability, on the basis of traditional probability system. Such understanding will be applied to dictate the designing of neural network structure and the choosing of expense function in the machine learning.Thus, considering the fact that when neural network is applied to protein secondary structure predicting, orthogonal encoding is usually chosen, and the physical and chemical characteristics of different protein, a new simplified encoding technology is put forward, and compared with other protein secondary structure predicting. And calculating and experiments prove that this new method makes more sufficient use of the information provided by protein initial structure, and gets better results. This coding method improves the predicting accuracy, and at the same time cuts the number of the input nodes. Thus it provides new way of thinking in the coding problem in protein secondary structure prediction.