mRCat: A Novel CatBoost Predictor for the Binary Classification of mRNA Subcellular Localization by Fusing Large...

Area of Science:

• Biomolecular science
• Computational biology
• Genomics

Background:

• Subcellular localization of messenger RNAs (mRNAs) is crucial for gene regulation and protein synthesis.
• Accurate prediction of mRNA localization is vital for disease diagnosis and drug development.
• Existing computational methods for mRNA localization prediction lack sufficient accuracy.

Purpose of the Study:

• To develop a novel computational predictor, mRCat, for accurate subcellular localization of mRNAs.
• To enhance the prediction of mRNA localization in the nucleus versus the cytoplasm.

Main Methods:

• Utilized large language models to extract hidden information from mRNA sequences.
• Integrated traditional sequence features for comprehensive mRNA characterization.
• Employed the CatBoost algorithm as the base classifier for prediction.

Main Results:

• mRCat achieved an accuracy of 0.761 on an independent test set.
• The predictor demonstrated strong performance with an F1 score of 0.710 and AUROC of 0.751.
• mRCat outperformed existing state-of-the-art methods in accuracy and robustness.

Conclusions:

• The mRCat predictor offers a significant advancement in accurately determining mRNA subcellular localization.
• This method provides valuable insights for biomolecular research, disease diagnosis, and drug development.
• The integration of LLMs and traditional features enhances predictive capabilities for mRNA localization.