GM-lncLoc: LncRNAs subcellular localization prediction based on graph neural network with meta-learning

Area of Science:

• Computational biology
• Genomics
• Molecular biology

Background:

• The function of long non-coding RNAs (lncRNAs) is closely linked to their subcellular localization.
• Accurate prediction of lncRNA subcellular localization is crucial for understanding their roles.
• Existing computational methods face challenges due to limited sample sizes and reliance on basic sequence information.

Purpose of the Study:

• To develop a novel computational method, GM-lncLoc, for accurate prediction of lncRNA subcellular localization.
• To address the issue of limited samples in lncRNA localization prediction using meta-learning.
• To improve upon existing prediction models by incorporating higher-level features.

Main Methods:

• Utilized initial sequence information from lncRNAs.
• Integrated graph structure information for extracting high-level lncRNA features.
• Employed a meta-learning training approach to efficiently learn parameters for similar tasks, addressing the few-samples problem.

Main Results:

• GM-lncLoc achieved high accuracy, reaching 93.4% and 94.2% on benchmark datasets with 5 and 4 subcellular compartments, respectively.
• Demonstrated superior performance on an independent dataset compared to previous methods.
• Validated the effectiveness and potential of the proposed GM-lncLoc model.

Conclusions:

• GM-lncLoc offers a robust and effective solution for lncRNA subcellular localization prediction.
• The integration of graph structures and meta-learning significantly enhances prediction accuracy, especially with limited data.
• The developed method holds great potential for advancing research in lncRNA function and genomics.