Analysis and prediction of exon skipping events from RNA-Seq with sequence information using rotation forest

In bioinformatics, exon skipping (ES) event prediction is an essential part of alternative splicing (AS) event analysis. Although many methods have been developed to predict ES events, a solution has yet to be found. In this study, given the limitations of machine learning algorithms with RNA-Seq data or genome sequences, researchers from Anhui University constructed a new feature, called RS (RNA-seq and sequence) features. These features include RNA-Seq features derived from the RNA-Seq data and sequence features derived from genome sequences. The researchers propose a novel Rotation Forest classifier to predict ES events with the RS features (RotaF-RSES). To validate the efficacy of RotaF-RSES, a dataset from two human tissues was used, and RotaF-RSES achieved an accuracy of 98.4%, a specificity of 99.2%, a sensitivity of 94.1%, and an area under the curve (AUC) of 98.6%.

 The framework of Rotation Forest classifier to predict ES events with RS features
(RotaF-RSES), showing both the training and testing stages

RotaF-RSES involves two steps. Step 1: Obtaining known exons, their upstream and downstream introns, and then extract RNA-Seq features and sequence features according to their RNA-Seq data and sequence information. The above two features, called RS features, were used to build a classification model based on a Rotating Forest algorithm (RotaF-RSES). Step 2: After obtaining the RS features of an unknown type of exon, the RotaF-RSES model was used to determine the type of exon.

When compared to the other available methods, the results indicate that RotaF-RSES is efficient and can predict ES events with RS features.

Availability – The source code and data of this approach can be used via http://ailab.ahu.edu.cn:8087/RotaF-RSES/index.html.