An ensemble deep learning framework for multi-class LncRNA subcellular localization with innovative encoding...

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Long non-coding RNAs (lncRNAs) are crucial regulators of cellular processes.
• Determining lncRNA subcellular localization is key to understanding their function.
• Existing computational methods face challenges in sequence data representation and nucleotide distribution analysis.

Purpose of the Study:

• To develop a novel, accurate computational model for predicting lncRNA subcellular localization.
• To address limitations in current sequence encoding and feature extraction for lncRNA analysis.

Main Methods:

• Proposed MGBLncLoc, a deep learning model.
• Introduced a generalized encoding based on Distribution Density of Multi-Class Nucleotide Groups (MCD-ND) for nucleotide distribution.
• Integrated advanced neural network modules: Multi-Dconv Head Transposed Attention, Gated-Dconv Feed-forward Network, CNN, and Bidirectional GRU.

Main Results:

• The MCD-ND encoding precisely reflects nucleotide distributions and identifies key sequence regions.
• The integrated deep learning architecture effectively captures lncRNA sequence features.
• MGBLncLoc demonstrated superior prediction performance compared to existing methods.

Conclusions:

• MGBLncLoc offers an effective and accurate solution for lncRNA subcellular localization prediction.
• The novel encoding and deep learning architecture advance the field of lncRNA computational analysis.
• This work provides valuable support for biological investigations involving lncRNAs.