EL-RMLocNet: An explainable LSTM network for RNA-associated multi-compartment localization prediction

Area of Science:

• Computational Biology
• Bioinformatics
• Molecular Biology

Background:

• Subcellular localization of Ribonucleic Acid (RNA) is crucial for understanding RNA function and disease association.
• Existing single-compartment predictors are limited, and multi-compartment predictors often lack accuracy and explainability.
• Current computational methods offer limited practical utility due to poor model interpretability.

Purpose of the Study:

• To develop an explainable deep learning model for accurate multi-compartment RNA subcellular localization prediction.
• To enhance the interpretability of RNA localization prediction models for better therapeutic optimization.
• To leverage raw RNA sequences for predicting localization across different RNA classes and species.

Main Methods:

• Developed an explainable Long Short-Term Memory (LSTM) network named EL-RMLocNet.
• Utilized a novel GeneticSeq2Vec scheme for statistical representation learning of RNA sequences.
• Incorporated an attention mechanism to optimize feature weighting for accurate prediction.

Main Results:

• EL-RMLocNet achieved superior predictive performance compared to state-of-the-art methods.
• Demonstrated an average accuracy improvement of 8% for Homo Sapiens and 6% for Mus Musculus.
• Enabled transparent and explainable multi-compartment localization predictions by mapping feature weights to nucleotide k-mer patterns.

Conclusions:

• EL-RMLocNet offers a significant advancement in explainable RNA subcellular localization prediction.
• The model's interpretability aids in understanding RNA behavior and potential disease mechanisms.
• The freely available web server facilitates broader research applications in RNA biology and therapeutics.