Abstract

PREDICTION OF RNA-BIDING PROTEINS BASED ON GRADIENT TREE BOOSTING

Xiaolin Wang* and Baoguang Tian

ABSTRACT

The interaction between RNA and proteins is of great significance to the regulation of gene expression, cell defense and development regulation and other life activities. Therefore, the use of machine learning methods to predict RBPs has become a hot spot in biological information. In this paper, we propose an RNA-binding proteins prediction model RBPro-GTB based on machine learning. Firstly, fusion feature coding, Pseudo-Position Specific Scoring Matrix (PsePSSM) and Grouped Tri-Peptide Composition (GTPC) are fused to extract features from protein sequences. Secondly, the Least Angle Regression-Least Absolute Shrinkage and Selection Operator (LARS- LASSO) is used to reduce high-dimensional feature vectors to low dimensions, and the cluster-based undersampling algorithm (ClusterCentroids) is used to overcome the impact of imbalanced samples. Finally, the optimal feature vectors are input into gradient tree boosting (GTB) classifier to predict RBPs. Based on the ten-fold cross validation, the prediction accuracy of the training set reaches 95.65%, and the matthews correlation coefficient reaches 0.9135. In addition, the prediction model of this paper is tested by an independent test set, and the ACC reaches 91.21%. Compared with other methods using the same dataset, the results show that our method has better performance.

Keywords: Feature extraction, Data Resampling, Gradient Tree Boosting, Pseudo Position Specific Scoring Matrix.