Method Research For The Prediction Of MicroRNA-disease Associations Based On Anti-noise Gradient Boosting Tree And Learning Interactions Of Featuresbased Algorithm

MicroRNA(mi RNA)is a single-stranded non-coding RNA and it is involved in many cell life activities.Researchers have found that mi RNA-related malfunctions can cause some specific diseases.This fact provides a new way for diseases diagnosis.Therefore,discovering new mi RNA-disease associations is important and meaningful.One of the effective methods to discover new associations is to predict potential associations by computational methods.In recent years,more and more researchers focus on using ensemble learning algorithms and deep learning algorithms to predict mi RNA-disease associations.Although these methods have achieved excellent results,they did not deal with the noise hiding in the data and learn the potential information between features.At the same time,the lengthy calculation process is also a disadvantage for these types of algorithms.In addition,few mi RNA-disease associations prediction tools have been designed and developed.To tackle these problems,this paper proposes anti-noise gradient boosting tree and learning interactions of features-based algorithm.Light feature algorithm is used to further improve the performance of these two algorithms.What is more,a prediction tool is designed and developed.The main contents of this paper are as follows:(1)Aiming at tackling the noise hiding in the data,this paper proposes anti-noise gradient boosting tree.First,several data subsets are created through sampling with replacement on data for noise smoothing.After that,extreme gradient boosting tree is utilized to learn and fit each data subset.Finally,the voting method is applied to make the final prediction based on the prediction results of each data subset.(2)Aiming at learning potential information between features,this paper proposes learning interactions of features-based algorithm.First,light gradient boosting tree is utilized to extract new features.After that,factorization machine and deep neural network is respectively applied to learn the low-order and high-order information of original features and extracted features.Finally,the final prediction can be obtained by combining the results of these two parts.(3)Aiming at accelerating calculation process,this paper proposes light feature algorithm.First,recursive feature elimination is used to rank all features.After that,several experiments are set up to confirm the best suitable features for target algorithms.Finally,the improved algorithms light anti-noise gradient boosting tree and light learning interactions of features-based algorithm are obtained.(4)This paper designs and develops a visualization tool.The tool encapsulates the light anti-noise gradient boosting tree and light learning interactions of features-based algorithm based on a web server and displays the results in a variety of ways.In summary,this paper proposes new methods on predicting mi RNA-disease associations.The experimental results show that these methods can effectively solve several problems mentioned above.Therefore,it can achieve certain theoretical significance and application value.