MEG-mod: A Multiview Enhanced Graph Neural Network for Knockdown Efficiency Prediction of Chemically Modified siRNA.

Area of Science:

• Biotechnology
• Bioinformatics
• Molecular Biology

Background:

• Chemical modification enhances siRNA stability and gene knockdown.
• Designing effective modified siRNAs is challenging due to complex variables.

Purpose of the Study:

• To develop a deep learning framework, MEG-mod, for predicting the knockdown efficiency of chemically modified siRNAs.
• To provide a practical tool for rational design and prioritization of modified siRNA candidates.

Main Methods:

• Constructed an expanded dataset using a literature-reading agent, manual curation, and public databases.
• Employed a structure-aware Transformer-based graph neural network to model siRNA sequence, chemical modifications, and structural relationships.
• Integrated a modification-base fusion module to capture context-dependent modification effects.

Main Results:

• MEG-mod achieved a high prediction accuracy with a Pearson correlation coefficient of 0.9171.
• The model's attention mechanism identified key modification sites and types, aligning with experimental findings.
• Demonstrated superior performance compared to existing methods in predicting siRNA knockdown efficiency.

Conclusions:

• MEG-mod offers a powerful and interpretable approach for designing chemically modified siRNAs.
• The developed web server provides a valuable resource for researchers in siRNA therapeutics development.