MVSLLnc: LncRNA subcellular localization prediction based on multi-source features and two-stage voting strategy.

Area of Science:

• Bioinformatics
• Molecular Biology
• Computational Biology

Background:

• Understanding long non-coding RNA (lncRNA) function requires accurate prediction of their subcellular localization.
• Traditional experimental methods for lncRNA localization are time-consuming.
• Existing computational methods may require substantial computing resources.

Purpose of the Study:

• To develop a simple, efficient, and easy-to-implement computational model for predicting lncRNA subcellular localization.
• To improve upon existing methods by utilizing multi-source features and a novel voting strategy.

Main Methods:

• Proposed the MVSLLnc model, integrating k-mer frequency, Chaos Game Representation (CGR) coordinates, and physicochemical properties (PhyChe).
• Employed a two-stage voting strategy combining Random Forest (RF), Support Vector Machine (SVM), and XGBoost classifiers.
• Validated the model on benchmark and independent test datasets.

Main Results:

• Achieved high accuracy on benchmark datasets (0.829, 0.793, 0.968).
• Demonstrated competitive performance on independent test sets (0.642, 0.737, 0.518).
• Ablation studies confirmed the effectiveness of the two-stage voting strategy and multi-source feature integration.

Conclusions:

• The MVSLLnc model provides an efficient and robust approach for predicting lncRNA subcellular localization.
• The proposed method effectively leverages diverse features and multiple classifiers for improved prediction accuracy.
• MVSLLnc offers a valuable tool for lncRNA research, complementing experimental approaches.