NeuroPred-GMC: a dual-branch deep learning architecture for neuropeptide prediction based on gated dilated...

Area of Science:

• Neuroscience
• Computational Biology
• Drug Discovery

Background:

• Neuropeptides are crucial signaling molecules regulating diverse physiological functions.
• Predicting neuropeptides accelerates the identification of novel therapeutic targets for conditions like pain and anxiety.
• Traditional experimental methods for neuropeptide identification are slow and resource-intensive.

Purpose of the Study:

• To develop an efficient computational method for neuropeptide prediction.
• To introduce NeuroPred-GMC, a novel deep learning architecture for enhanced neuropeptide discovery.
• To provide a robust and generalizable tool for identifying potential drug candidates.

Main Methods:

• A dual-branch deep learning architecture, NeuroPred-GMC, was designed.
• The architecture integrates gated dilated convolutional networks with ESM-2 features and multi-scale convolutional networks with Prot-T5 features.
• Key components include dilated convolution for expanded receptive fields, gating for selective feature enhancement, and multi-scale convolution for contextual information capture.

Main Results:

• NeuroPred-GMC achieved high performance on an independent test set: 93.24% accuracy, 93.69% sensitivity (Sn), 92.79% specificity (Sp), 92.86% precision (Pre), 0.8649 MCC, and 0.9667 auROC.
• Cross-validation and independent testing confirmed the model's robustness and generalizability.
• The model demonstrates significant potential as a supplemental predictor for neuropeptides.

Conclusions:

• NeuroPred-GMC offers a powerful computational approach for neuropeptide prediction, significantly outperforming traditional methods.
• The model's architecture effectively leverages advanced deep learning techniques for feature extraction and prediction.
• This tool can accelerate drug discovery by rapidly identifying novel neuropeptide targets for therapeutic development.