Toward high-efficiency, low-resource, and explainable neuropeptide prediction with MSKDNP

Area of Science:

• Neuroscience
• Biochemistry
• Computational Biology

Background:

• Neuropeptides are vital signaling molecules in the nervous system, regulating physiological processes.
• Neuropeptides are implicated in neurodegenerative and neuropsychiatric disorders, making their accurate identification crucial for research and drug development.
• Current AI methods for neuropeptide identification are computationally intensive and lack user-friendly interfaces.

Purpose of the Study:

• To develop an efficient and accurate artificial intelligence model for neuropeptide identification.
• To address the limitations of existing neuropeptide prediction methods, such as high computational cost and slow processing speed.
• To provide a user-friendly web server for practical application in biomedical research.

Main Methods:

• Proposed MSKDNP, a neuropeptide prediction model utilizing a multi-stage knowledge distillation framework.
• Implemented knowledge distillation to create a computationally efficient model with fewer parameters.
• Developed a web server for accessible use of the MSKDNP model.

Main Results:

• MSKDNP achieved performance comparable to a fully fine-tuned protein language model with only 1.2% of the parameters.
• The model demonstrated state-of-the-art results in neuropeptide recognition.
• MSKDNP offers favorable interpretability, aiding in biological understanding.

Conclusions:

• MSKDNP represents a significant advancement in efficient and accurate neuropeptide identification.
• The model's efficiency and interpretability facilitate deeper biological insights and therapeutic strategy development.
• A publicly accessible web server is available for widespread use in neuropeptide research.