LncLocFormer: a Transformer-based deep learning model for multi-label lncRNA subcellular localization prediction by...

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Subcellular localization of long non-coding RNAs (lncRNAs) is crucial for understanding their biological functions.
• lncRNAs often exhibit multiple subcellular localizations with distinct patterns.
• Existing computational methods lack the ability to predict multi-label localization and consider motif specificity.

Purpose of the Study:

• To develop a novel deep learning model, LncLocFormer, for predicting multi-label subcellular localization of lncRNAs using only their sequences.
• To incorporate motif specificity into the prediction of lncRNA localization.
• To improve the accuracy of lncRNA subcellular localization prediction.

Main Methods:

• Proposed LncLocFormer, a deep learning model utilizing eight Transformer blocks to capture long-range dependencies in lncRNA sequences.
• Implemented a localization-specific attention mechanism to model relationships between different subcellular localizations.
• Used lncRNA sequences as input for multi-label prediction.

Main Results:

• LncLocFormer demonstrated superior performance compared to state-of-the-art predictors on a hold-out test set.
• Motif analysis revealed LncLocFormer's capability to identify known functional motifs within lncRNA sequences.
• Ablation studies confirmed the significant contribution of the localization-specific attention mechanism to prediction accuracy.

Conclusions:

• LncLocFormer provides an effective approach for predicting multi-label subcellular localization of lncRNAs.
• The model's ability to capture motif information enhances prediction accuracy and biological relevance.
• LncLocFormer advances computational methods for lncRNA functional genomics research.