The stacking strategy-based hybrid framework for identifying non-coding RNAs

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Next-generation sequencing generates vast transcriptomic data, complicating the identification of non-coding ribonucleic acids (ncRNAs).
• Existing methods struggle with ncRNA identification, particularly for non-model organisms lacking extensive transcriptional data.

Purpose of the Study:

• To develop a robust and universal framework for identifying ncRNAs across diverse species.
• To enhance ncRNA identification accuracy by incorporating novel peptide-based features.

Main Methods:

• A hybrid stacking framework combining Random Forest, LightGBM, XGBoost, and logistic regression classifiers was developed.
• The framework utilized DNA-based, RNA-based, and eight novel predicted peptide-based features.
• Two models were trained: one for cross-species ncRNA identification and another for plant-specific ncRNA identification.

Main Results:

• The cross-species model achieved superior performance in Arabidopsis, worm, and zebrafish datasets (98.36%, 99.65%, 94.12% accuracy, respectively).
• Across all six tested species, the model demonstrated top performance in sensitivity, accuracy, precision, and F1 scores.
• The plant-specific model consistently yielded average metric values exceeding 95% across two experiments.

Conclusions:

• The proposed hybrid framework is effective for identifying ncRNAs in both animal and plant species.
• The framework exhibits significant advantages for cross-species ncRNA identification, addressing limitations of existing methods.