Figures from this paperAbstractFigures from this paperAbstractFigures from this paperAbstractFigures from this paper

Abstract

Neuropeptides, small protein-like molecules, play essential roles in cellular communication and modulate physiological processes such as pain, mood, and immune responses. This thesis presents a neuropeptide classification method based on a fine-tuned ESM-1b model. Initially, the ESM-1b model was pre-trained on a neuropeptide-specific dataset, with multiple sequence alignment (MSA) via a Hidden Markov Model (HMM) applied to the training sequences. This generated an output.afa file, which highlighted shared sequence features to enhance model generalization. After fine-tuning, the model was combined with a convolutional neural network (CNN) to extract high-dimensional feature representations that comprehensively characterize the input sequences. These representations were subsequently processed by a gradient boosting tree classifier, which optimized feature weighting and classification, enabling precise differentiation between neuropeptides and other peptides. This multi- step approach leverages the advantages of both deep learning and ensemble learning techniques, enhancing the accuracy and robustness of neuropeptide classification and laying a foundation for deeper insights into their diverse biological roles.