MSLP: mRNA subcellular localization predictor based on machine learning techniques.

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• Subcellular localization of messenger RNA (mRNA) is crucial for gene expression, cell migration, and adaptation.
• Experimental methods for determining mRNA localization are laborious, time-consuming, and expensive.
• In silico approaches are gaining traction in the RNA community for predicting mRNA localization.

Purpose of the Study:

• To develop a machine learning-based method (MSLP) for predicting mRNA subcellular localization.
• To integrate diverse sequence features for enhanced prediction accuracy.
• To provide an accessible computational tool for researchers.

Main Methods:

• MSLP utilizes a machine learning algorithm trained on four feature types: k-mer, pseudo k-tuple nucleotide composition (PseKNC), nucleotide physicochemical properties, and Z-curve based 3D sequence representation.
• Ensemble-based models were employed to combine these features for prediction.
• The method was evaluated across ten distinct subcellular locations.

Main Results:

• Ensemble models achieved state-of-the-art performance in mRNA subcellular localization prediction.
• Ablation studies revealed k-mer and PseKNC features are dominant for predicting cytoplasmic, nuclear, and ER localization.
• Physicochemical properties and Z-curve features were most important for extracellular region and mitochondria localization.
• SHAP analysis provided insights into feature importance.

Conclusions:

• MSLP offers an accurate and efficient computational tool for predicting mRNA subcellular localization.
• The study highlights the importance of integrating multiple sequence feature types for robust predictions.
• A Docker container and API are available, along with datasets and code, to facilitate community use and further research.