NeuroPred-PLM

NeuroPred-PLM predicts neuropeptides from protein sequences to enable large-scale, interpretable identification of signaling peptides for research and drug discovery.

Key Features:

• Interpretable model: A global multi-head attention network produces attention scores that capture position-wise contributions to neuropeptide prediction.
• Robust predictive performance: Demonstrates superior performance compared to existing state-of-the-art predictors via benchmarking on an independent test set derived from the NeuroPep 2.0 database.
• Semantic representation with ESM: Uses the ESM (Evolutionary Scale Modeling) protein language model to obtain semantic embeddings of neuropeptides, reducing reliance on traditional feature engineering.
• Enhanced local feature extraction: Integrates a multi-scale convolutional neural network to improve local feature representation of neuropeptide embeddings and prediction accuracy.

Scientific Applications:

• Drug discovery: Enables large-scale identification of neuropeptides to support discovery of peptide-based therapeutics.
• Target identification: Facilitates identification of signaling peptides relevant to biological pathways and therapeutic targets.
• Neuropeptide discovery: Supports systematic cataloging and analysis of neuropeptides across diverse datasets for disease research.

Methodology:

NeuroPred-PLM derives semantic representations using the ESM protein language model, applies a multi-scale convolutional neural network for local feature extraction, and employs a global multi-head attention network to generate attention scores for position-wise interpretability.