Integrating RNA-seq transcript signals, Primary and Secondary Structure Information in Differentiating Coding and Non-coding RNA Transcripts

摘要

Non-coding RNA transcripts play important roles in different regulatory functions inside the cell which were previously treated as "junk". Exploring the differences between the non-coding transcripts with the protein coding transcripts helps researchers annotate many novel RNA transcripts. Researchers already showed that the secondary structures of non-coding RNA transcripts can be used to distinguish them from coding transcripts. But, there is a large class of non-coding transcripts that are completely unstructured, thereby making the structure-based classification results unreliable. However, the advent of Next Generation Sequencing platform, RNA-seq, producing more and more expression data of many organisms under different set of conditions, adds a new dimension to this classification task. In this paper, we presented a way to integrate the primary structures (i.e., the sequence), secondary structures and differential expressions of the two types of transcripts obtained from RNA-seq experiments under three different conditions and design a better classification method. We employed two popular classification algorithms-SVMs and Conditional Random Forest classifier to build the model to discriminate the two classes. The prediction accuracy of these classifiers are 93% and 91% respectively which are much higher than those of the two existing classification systems - CPC and PORTRAIT.