IRESGet: A transformer framework-based model to predict internal ribosome entry sites.

Area of Science:

• Computational Biology
• Molecular Biology
• Bioinformatics

Background:

• Internal ribosome entry site (IRES) elements are vital for cap-independent translation, particularly in circular RNA protein expression.
• Current methods for IRES identification are experimentally intensive and computationally limited in accuracy and efficiency.

Purpose of the Study:

• To develop a highly accurate and efficient computational model for predicting internal ribosome entry site (IRES) sequences.
• To overcome the limitations of existing machine learning and deep learning approaches for IRES identification.

Main Methods:

• Developed IRESGet, a Transformer-based deep learning model for IRES sequence prediction.
• Utilized RNA-FM for optimal feature encoding and the Optuna library for hyperparameter optimization.
• Trained and validated the model on curated datasets from published literature and public databases.

Main Results:

• IRESGet demonstrated superior accuracy and computational efficiency compared to established tools like DeepCIP and IRESfinder.
• The model outperformed five mainstream machine learning frameworks by effectively avoiding overfitting.
• Interpretability analysis identified a novel de novo motif (SBNCAGVNVNNN) enriched in high-attention regions, potentially recognized by specific RNA-binding proteins.

Conclusions:

• IRESGet represents a significant advancement in computational IRES prediction, outperforming existing methods.
• While dataset limitations may lead to some false positives, the model shows great promise for improving accuracy with enhanced data.
• This tool offers a novel and effective solution for identifying IRES elements in RNA sequences.