EnsembleNPPred: A Robust Approach to Neuropeptide Prediction and Recognition Using Ensemble Machine Learning and Deep Learning Methods

Abstract

Neuropeptides (NPs) are a diverse group of signaling molecules involved in regulating key physiological processes such as pain perception, stress response, mood, appetite, and circadian rhythms. Acting as neurotransmitters, neuromodulators, or neurohormones, they play a critical role in modulating and fine-tuning neural signaling networks. Despite their biological significance, identifying NPs through experimental techniques remains time-consuming and resource-intensive. To support this effort, computational prediction tools have emerged as a cost-effective approach for prioritizing candidate sequences for experimental validation. In this study, we propose EnsembleNPPred, an ensemble learning framework that integrates traditional machine learning (ML) models with a deep learning (DL) component. By combining the complementary strengths of these approaches, the model aims to improve generalization and predictive robustness. EnsembleNPPred employs a majority voting mechanism to aggregate the outputs from three classifiers: Support Vector Machine (SVM), Extra Trees (ET), and a CNN-based DL model. When evaluated on independent datasets, EnsembleNPPred demonstrated consistently competitive performance, achieving improvements in both accuracy and sensitivity-specificity balance compared to several existing methods. Furthermore, testing on multiple neuropeptide families from the NeuroPep database yielded an average accuracy of 91.92%, suggesting the model’s potential to generalize across diverse peptide classes. These results suggest that EnsembleNPPred may be a useful tool for early-stage neuropeptide candidate identification and for supporting downstream experimental validation. © 2025 by the authors.