SAWRPI: A Stacking Ensemble Framework With Adaptive Weight for Predicting ncRNA-Protein Interactions Using Sequence Information.

Zhong-Hao Ren,Yong-Jian Guan,Li-Ping Li,Zhu-Hong You,Chang-Qing Yu,Jie Pan,Yue-Chao Li

doi:10.3389/fgene.2022.839540

Zhong-Hao Ren, Yong-Jian Guan + Show 5 more

Open Access

https://doi.org/10.3389/fgene.2022.839540

Copy DOI

Abstract

Non-coding RNAs (ncRNAs) take essential effects on biological processes, like gene regulation. One critical way of ncRNA executing biological functions is interactions between ncRNA and RNA binding proteins (RBPs). Identifying proteins, involving ncRNA-protein interactions, can well understand the function ncRNA. Many high-throughput experiment have been applied to recognize the interactions. As a consequence of these approaches are time- and labor-consuming, currently, a great number of computational methods have been developed to improve and advance the ncRNA-protein interactions research. However, these methods may be not available to all RNAs and proteins, particularly processing new RNAs and proteins. Additionally, most of them cannot process well with long sequence. In this work, a computational method SAWRPI is proposed to make prediction of ncRNA-protein through sequence information. More specifically, the raw features of protein and ncRNA are firstly extracted through the k-mer sparse matrix with SVD reduction and learning nucleic acid symbols by natural language processing with local fusion strategy, respectively. Then, to classify easily, Hilbert Transformation is exploited to transform raw feature data to the new feature space. Finally, stacking ensemble strategy is adopted to learn high-level abstraction features automatically and generate final prediction results. To confirm the robustness and stability, three different datasets containing two kinds of interactions are utilized. In comparison with state-of-the-art methods and other results classifying or feature extracting strategies, SAWRPI achieved high performance on three datasets, containing two kinds of lncRNA-protein interactions. Upon our finding, SAWRPI is a trustworthy, robust, yet simple and can be used as a beneficial supplement to the task of predicting ncRNA-protein interactions.

Highlights

IntroductionHuman proteins are translated from less than 2% of genome, but more than 80% of genome has biochemical functions (Djebali et al, 2012; Pennisi 2012), which accounts for the large number of non-coding RNA (ncRNA), known as the RNA with little or without ability of encoding proteins, have biological functions
RPI488 is a non-redundant dataset of Long non-coding RNA (lncRNA)-protein interactions, containing 245 negative samples and 243 positive samples among 25 lncRNAs and 247 proteins (Huang et al, 2010; Puton et al, 2012)
To predict ncRNA-protein interactions, we developed a computational method SAWRPI

Summary

Introduction

Human proteins are translated from less than 2% of genome, but more than 80% of genome has biochemical functions (Djebali et al, 2012; Pennisi 2012), which accounts for the large number of non-coding RNA (ncRNA), known as the RNA with little or without ability of encoding proteins, have biological functions. Wet experiments have no ability to examine ncRNA-protein interactions efficiently and effectively because of the large number of unexplored interactions. Due to experimental methods are costly, time-consuming and localized, and sequences of RNA and protein carry sufficient information for predicting interaction between them (Ray et al, 2009; Alipanahi et al, 2015), many computational models have been proposed as alternative methods to overcome the drawbacks of ncRNAprotein interactions prediction

Methods

Results

Conclusion