Study On Prediction Of Protein Contact Based On Markov Logic Network

Protein tertiary structure prediction is a major challenge in bioinformatics, and sofar we have not a good way to get the desired prediction.When we use Ab-initioprediction method to predict the three-dimensional structure of the protein, if we canget the protein contact information, the prediction accuracy can be greatly improved.The protein contact prediction plays a very important role in the proteinthree-dimensional structure prediction.In the same protein, the contact between the protein residues is not independent ofeach other, while the traditional machine learning methods, they require the sampleinstance are independent and identically distributed, so this type of method can not bea good solution to the protein contact prediction problem. At the same time, proteincontact is often subject to the constraints of the rules, many rules are a priori.TheMarkov logic network method, which is based on statistical relational learningframework, can overcome the traditional machine learning for independent consistencyrequirement, and can capture the constraints in the protein contact rules. The methoduses a logical language based on weighted rules to express the protein contact with thedomain knowledge, and it is a good solution for the protein contact prediction.The main research work is as follows:①Making a comprehensive overview of protein structure prediction and proteincontact prediction, including the research background, status, and significance.②Conducted a comprehensive review of statistical relational learning and thetheory of Markov logic network. Markov logic network’s concepts and characteristics,and Markov logic network learning and inference algorithms.③Introducing the data used in the protein contact prediction. Making specificanalysis of the information used in the protein contact prediction.④Making specific and in-depth research about the protein contact rules from theprotein space physics structure and biochemical characteristics, and constructingMarkov logic network predicate formula.⑤Appling Markov logic network to the protein contact prediction andanalysising the experimental results.The experimental results show that the method used in this paper can achievebetter results than the other prediction methods.Compared with the BetaPro method, it can improve the prediction accuracy by eight precentage points, which furthervalidates the value of the Markov logic network.