Prediction of LncRNA Subcellular Localization with Deep Learning from Sequence Features

Abstract

Long non-coding RNAs are involved in biological processes throughout the cell including the nucleus, chromatin and cytosol. However, most lncRNAs remain unannotated and functional annotation of lncRNAs is difficult due to their low conservation and their tissue and developmentally specific expression. LncRNA subcellular localization is highly informative regarding its biological function, although it is difficult to discover because few prediction methods currently exist. While protein subcellular localization prediction is a well-established research field, lncRNA localization prediction is a novel research problem. We developed DeepLncRNA, a deep learning algorithm which predicts lncRNA subcellular localization directly from lncRNA transcript sequences. We analyzed 93 strand-specific RNA-seq samples of nuclear and cytosolic fractions from multiple cell types to identify differentially localized lncRNAs. We then extracted sequence-based features from the lncRNAs to construct our DeepLncRNA model, which achieved an accuracy of 72.4%, sensitivity of 83%, specificity of 62.4% and area under the receiver operating characteristic curve of 0.787. Our results suggest that primary sequence motifs are a major driving force in the subcellular localization of lncRNAs.

Predictive models of subcellular localization of long RNAs

“…We describe here an attempt to use the existing information about the maturation level, chromatin marks, gene architecture and sequence features to predict the simplest dimension of subcellular localization of long RNAs in cells-nucleus versus cytoplasm. This attempt complements the recent development of machine learning approaches that attempt to predict subcellular localization using sequence features alone (Cao et al 2018;Gudenas and Wang 2018;Su et al 2018). Our study and others are based on ENCODE data, which include very high-quality RNAseq on subcellular fractions, but is presently limited to human cancer cell lines.…”

Competing Endogenous RNA Networks in the Epithelial to Mesenchymal Transition in Diffuse-Type of Gastric Cancer

“…LncRNAs can be found in both the nucleus and the cytoplasm and are classified into guides, dynamic scaffolds, and molecular decoys. LncRNAs participate in important regulatory roles in biological processes, such as X inactivation, imprinting, development, mRNA processing, epigenetic modifications, and organization of nuclear architecture [ 54 , 55 ].…”

“…In the cytoplasm, lncRNAs generate a large-scale of trans-regulatory crosstalk through their capacity to communicate with coding mRNA [ 55 , 56 , 57 , 58 ]. In this crosstalk and through the MREs, lncRNAs act as sponges to hijack miRNAs, resulting in the reduction of an available miRNA pool to target mRNAs [ 59 ].…”

Genomic data mining for functional annotation of human long noncoding RNAs

“…We have recently developed a deep neural network model, called DeepLncRNA, to predict the subcellular localization of a lncRNA from its transcript sequence (Gudenas and Wang, 2018). The model was constructed using a comprehensive dataset of…”