NeuroPpred-MSN: A Neuropeptide Prediction Model Based on Multi-feature Fusion and Siamese Networks

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Neuropeptides are crucial targets for novel drug development.
• Existing computational methods for neuropeptide prediction require performance enhancement.

Purpose of the Study:

• To introduce NeuroPpred-MSN, an innovative and efficient neuropeptide prediction model.
• To improve the accuracy and robustness of neuropeptide identification using advanced computational techniques.

Main Methods:

• Utilized multi-feature fusion and Siamese networks for neuropeptide representation.
• Employed four encoding schemes: token embedding, word2vec, protein language embedding, and handcrafted features.
• Integrated ProtT5-XL-UniRef50 for embedding generation and Bi-GRU for feature processing.

Main Results:

• NeuroPpred-MSN achieved an AUROC of 98.3% on an independent test set.
• Demonstrated significant improvements in accuracy (1.52%), F1 score (1.52%), and MCC (3.2%) over state-of-the-art predictors.
• Showcased excellent performance on imbalanced datasets, highlighting robustness and generalization.

Conclusions:

• NeuroPpred-MSN offers a significant advancement in neuropeptide prediction accuracy.
• The model's superior performance and robustness make it a valuable tool for drug discovery and disease treatment.
• The developed model is publicly accessible for further research and application.