pNPs-CapsNet: Predicting Neuropeptides Using Protein Language Models and FastText Encoding-Based Weighted Multi-View...

Area of Science:

• Biochemistry and Bioinformatics
• Computational Biology
• Drug Discovery

Background:

• Neuropeptides (NPs) are crucial signaling molecules with therapeutic potential.
• Experimental NP identification is resource-intensive.
• Computational methods offer a cost-effective alternative for NP prediction.

Purpose of the Study:

• To develop a novel deep capsule neural network model, pNPs-CapsNet, for accurate prediction of neuropeptides (NPs) and non-NPs.
• To leverage advanced protein language models and feature selection strategies for enhanced predictive performance.
• To establish a robust computational tool for accelerating NP identification in drug discovery.

Main Methods:

• Numerical encoding of peptide sequences using pretrained protein language models (ESM, ProtBERT-BFD, ProtT5).
• Weighted feature integration using differential evolution to create a multiview vector.
• Two-tier feature selection (MRMD and SHAP) followed by training a capsule neural network (CapsNet).

Main Results:

• The pNPs-CapsNet model achieved 98.10% accuracy and 0.98 AUC on training data.
• Independent validation demonstrated 95.21% accuracy and 0.96 AUC.
• pNNPs-CapsNet outperformed existing models by 4% and 2.5% on training and independent datasets, respectively.

Conclusions:

• pNNPs-CapsNet is a highly accurate and robust computational model for neuropeptide prediction.
• The model offers a significant advancement over current state-of-the-art methods.
• pNNPs-CapsNet shows considerable potential for facilitating drug discovery and academic research.