RNALoc-LM: RNA subcellular localization prediction using pre-trained RNA language model.

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Accurate prediction of RNA subcellular localization is vital for understanding RNA functions and regulation.
• Existing computational methods often focus on single RNA types, leaving a gap for multi-type prediction.
• Pre-trained RNA language models show promise in bioinformatics but are underutilized for subcellular localization prediction.

Purpose of the Study:

• To develop an interpretable deep-learning framework for predicting RNA subcellular localization.
• To leverage pre-trained RNA language models for enhanced prediction accuracy.
• To enable simultaneous prediction for multiple RNA types.

Main Methods:

• Developed RNALoc-LM, a framework utilizing pre-trained RNA language models for sequence encoding.
• Employed TextCNN and BiLSTM modules to capture local and long-range dependencies in RNA sequences.
• Incorporated a multi-head attention mechanism to focus on critical RNA regions.

Main Results:

• RNALoc-LM significantly outperformed existing deep-learning baselines and state-of-the-art predictors.
• Motif analysis demonstrated RNALoc-LM's capability in identifying important RNA motifs.
• An ablation study validated the effectiveness of RNA sequence embeddings from the pre-trained language model.

Conclusions:

• RNALoc-LM represents a significant advancement in predicting RNA subcellular localization.
• The framework's interpretability and performance highlight the potential of pre-trained RNA language models in this field.
• RNALoc-LM offers a powerful tool for both prediction and motif discovery in RNA biology.