NeuroCL: A deep learning approach for identifying neuropeptides based on contrastive learning

Area of Science:

• Neuroscience
• Computational Biology
• Artificial Intelligence

Background:

• Neuropeptides (NPs) are crucial signaling molecules in neurotransmission, endocrine regulation, and mood control.
• Accurate NP identification is vital for disease diagnosis, targeted therapy, and personalized medicine.
• Existing models struggle with complex feature relationships and inter-sample connections in NP identification.

Purpose of the Study:

• To develop an advanced deep learning model, NeuroCL, for efficient and accurate neuropeptide identification.
• To overcome limitations of previous models in capturing intricate data relationships.

Main Methods:

• Implemented NeuroCL, a deep learning model utilizing contrastive learning and a cross-attention mechanism.
• Employed multifaceted attribute representation for comprehensive NP analysis.
• Integrated pre-trained large models with manually encoded features via cross-attention.

Main Results:

• NeuroCL achieved 93.8% accuracy and 87.8% MCC on an independent test set.
• Contrastive learning improved class distinction and coherence.
• Cross-attention enhanced feature integration and connections, outperforming state-of-the-art predictors.
• UMAP visualization confirmed distinct segregation of positive and negative NPs.

Conclusions:

• NeuroCL demonstrates superior performance in neuropeptide identification.
• The model's architecture effectively captures complex data nuances and strengthens feature associations.
• A web-based platform is available for NeuroCL accessibility and application.