MHAM-NPI: Predicting ncRNA-protein interactions based on multi-head attention mechanism

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• Non-coding RNA (ncRNA) molecules are crucial for gene regulation and biological processes.
• Understanding ncRNA-protein interactions is vital for elucidating ncRNA functions.
• Accurate prediction of these interactions remains a significant challenge in bioinformatics.

Purpose of the Study:

• To develop an accurate computational method for predicting ncRNA-protein interactions.
• To leverage deep learning techniques for automatic feature extraction from ncRNA and protein sequences.

Main Methods:

• Utilized a multi-head attention mechanism integrated with residual connections.
• Employed node feature projection into multiple spaces to capture diverse interaction patterns.
• Stacked interaction layers to derive higher-order feature interactions while preserving initial information.

Main Results:

• Achieved high prediction accuracy with AUC values of 97.4% on NPInter v2.0, 98.5% on RPI807, and 94.8% on RPI488.
• Demonstrated the method's effectiveness in capturing complex, high-order sequence features.
• The proposed method shows significant promise for ncRNA-protein interaction studies.

Conclusions:

• The developed multi-head attention method is a powerful tool for predicting ncRNA-protein interactions.
• The approach effectively utilizes sequence information to identify hidden high-order features.
• The findings contribute to advancing the understanding of ncRNA functions through accurate interaction prediction.