Prediction Of MicroRNA Based On Computational Intelligence

With the successful implementation of the Human Genome Project, researchers getmassive bio-molecular data, and then how to analysis, process and extract biological feature ofthese information, is another problem that scientists have to face. Bioinformatics as theinterdisciplinary subject of biology, computer science and applied mathematics providemethods to solve the problems, and get more and more attention from human. Bioinformaticsnot only provide great help for human to understand the genetic information, but to clearlyrecognize the importance of gene regulatory networks.It has been long that the understanding of the biological central dogma is that geneticinformation is generated by the DNA transcribed to RNA, and then translated to protein. But thediscovery of non-coding RNA has changed people’s original cognitive model.MicroRNA(microRNA) can suppress or crack the target messenger RNA by the principle of thebase complementary, so as to affecting the biological gene regulation. In recent years,microRNA has become one of the hot issues in the bioinformatics research. It has played animportant regulatory role in the growth and development of plants and animals, cellproliferation and apoptosis, organ formation, virus defense, and also is closely related to humandisease and cancer.Till now the discovered number of microRNA takes up only a very little part of the entiremicroRNA family. Because of the base number of microRNA is just about21-25, the commonpredicting methodology to detecting microRNA is by forecasting the sequence of pre-miRNAwhether containing microRNA indirectly. Machine learning method is an efficient, convenientand accurate method to predict microRNA. There are two key issues affect the predicted resultsin this method, one is whether the extracted features can represent the pre-miRNA sequencesaccurately, the other is prediction model.The study found that the stem-loop structure of the pre-miRNA is highly conserved andthe adjacent base pairs in the stem have important complementary effect. So we encode threeadjacent base pairs to characterize the stem-loop structure of the pre-miRNA. And another fourfeatures about sequence information of the pre-miRNA have been extracted together to form 36-dimensional feature vector in this experiment.For the artificial neural network has self-organizing, self-learning and adaptivecharacteristics and is very good at dealing with nonlinear optimization problem ofbioinformatics. So we first select feed forward artificial neural network model with mentortraining to predict the microRNA. At the same time in order to avoid falling into the localoptimum, we use particle swarm optimization to train parameters. The accuracy of theexperiment has been improved by this method. In order to improve the accuracy of thisexperiment furthermore, we adopted the integration approach. The integration which has beenproved to be an effective machine learning method can significantly improve the predictionaccuracy than single base classifier. Neural network has been used as base classifier tointegration and the selective ensemble algorithm based on genetic algorithm has been used topredict microRNA. The experiment proves that the predicting result is better than singleartificial neural network. Finally, we use the flexible neural tree as experiment model to predictmicroRNA. The flexible neural tree’s structure and parameters can be automatic optimized. Itsolves the problem that the network structure and the number of neuron of hidden layer have tobe set in advance. The flexible neural tree also has feature selection function and can reduce thedimensionality of the original data.Several authentic and pseudo test sets have been used to test artificial neural network,integrated network and flexible neural tree model. The prediction accuracy has been improvedmore, which validates that the structure of neural network model in the prediction of microRNAis effective, and opens up a new way for microRNA predict.