EL-RMLocNet

EL-RMLocNet predicts multi-compartment subcellular localization of RNA molecules using an explainable Long Short-Term Memory (LSTM) network to reveal nucleotide k-mer contributions for studying RNA function and disease associations.

Key Features:

• Explainable Predictive Model: Uses an explainable Long Short-Term Memory (LSTM) network to extract sequence-derived features for multi-compartment RNA localization prediction.
• GeneticSeq2Vec Statistical Representation: Employs GeneticSeq2Vec to convert raw RNA sequences into statistical vectors that capture short- and long-range relationships of nucleotide k-mers.
• Attention Mechanism: Integrates an attention mechanism to weight discriminative features extracted by LSTM layers for improved localization accuracy.
• Reverse Engineering for Transparency: Maps model weights in the statistical feature space back to nucleotide k-mer patterns to enable interpretation of predictions.
• Empirical Performance: Demonstrates improved accuracy over state-of-the-art predictors across four RNA classes, with reported average accuracy gains of 8% for Homo sapiens and 6% for Mus musculus.

Scientific Applications:

• RNA Functionality Insights: Provides localization-based insights into RNA functionality by linking sequence patterns to subcellular compartments.
• Disease Association Exploration: Supports exploration of RNA co-localization across compartments to investigate associations with diseases.
• Therapeutic Optimization: Facilitates investigation of localization mechanisms that can inform optimization of RNA-targeted therapeutics.

Methodology:

Raw RNA sequences are converted into statistical vectors using GeneticSeq2Vec; those vectors are processed by LSTM layers to extract features; an attention mechanism weights discriminative features for localization prediction; reverse engineering maps feature weights back to nucleotide k-mer patterns.