Identifying Best Machine Learning Models to Distinguish Neuropeptides from Non-Neuropeptides

Neuropeptides are long-term signaling molecules. In this experiment, machine learning models were used to differentiate between neuropeptides and non-neuropeptides. The goal of this study was to compare model performance on a binary classification task. The NeuroPred-FRL dataset from Kyushu Institute of Technology, containing labeled neuropeptides and non-neuropeptides, was used. An additional dataset containing physicochemical properties for each amino acid was incorporated. Peptide sequences were decomposed into individual amino acids, annotated with physicochemical properties, aggregated to form sequence-level representations, and standardized. The models compared were random forests, support vector machines (SVM), logistic regression, k-nearest neighbors (kNN), neural networks, and gradient boosting. Cross-validation was applied to the training set, and 20% of the data was held out for testing. Model performance was evaluated using the AUC, F1, and MCC metrics. Neural networks achieved the highest performance, while random forest and gradient boosting models also performed well. These results demonstrate that multiple machine learning approaches are effective for neuropeptide classification and provide a foundation for more detailed and biologically meaningful prediction tasks.