BiGM-lncLoc: Bi-level Multi-Graph Meta-Learning for Predicting Cell-Specific Long Noncoding RNAs Subcellular...

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• Long noncoding RNAs (lncRNAs) are crucial in biological regulation.
• Aberrant lncRNA expression and localization are linked to cancer development.
• Existing computational methods often overlook cell line specificity and inter-cell line correlations.

Purpose of the Study:

• To develop a novel computational approach, BiGM-lncLoc, for predicting lncRNA subcellular localization.
• To address limitations of current methods by incorporating cell line specificity and shared information.
• To improve the accuracy and applicability of lncRNA localization prediction in cancer research.

Main Methods:

• BiGM-lncLoc treats lncRNA subcellular localization prediction as a multi-graph meta-learning task.
• It integrates nucleotide sequence localization data and cell line expression data.
• The method employs a cell line-specific optimization network and a graph neural network optimized across cell lines.

Main Results:

• BiGM-lncLoc achieved an average prediction accuracy of 97.7% across various cell lines.
• Even with independent data, accuracy ranged from 82.4% to 94.7%, surpassing existing models.
• Key feature analysis confirmed the necessity of cell line-specific studies.

Conclusions:

• BiGM-lncLoc offers a robust and accurate method for predicting lncRNA subcellular localization.
• The approach effectively leverages cell line-specific and shared information for enhanced prediction.
• This advancement holds promise for understanding lncRNA roles in cancer biology.