Predicting Quorum Sensing Peptides Using Machine Learning Models

Quorum sensing micropeptides play important roles in bacterial communication. The identification of these micropeptides from amino acid sequence data remains challenging due to their short lengths and limited distinguishing features. In this study, machine learning approaches were used to classify quorum sensing micropeptides using sequence-derived feature representations. A dataset of 1229 micropeptide sequences, ranging from 5 to 15 amino acids in length, was analyzed, including both QSPs and non-QSPs. Two feature extraction methods were investigated: k-mer representations with k=3, which capture local sequence patterns, and amino acid composition, which counts individual frequencies. These features were combined with logistic regression and neural network classifiers, resulting in four predictive models. Model performance was evaluated using classification accuracy, precision, recall, AUC, and MCC. Among the evaluated approaches, the k-mer based neural network achieved the highest overall performance, demonstrating superior AUC and MCC values compared to other models. Models trained on amino acid composition features, particularly using logistic regression, demonstrated a comparatively lower performance. Overall, the results demonstrate that machine learning models using k-mer sequence features and non-linear classifiers can effectively distinguish quorum sensing micropeptides from non-quorum sensing peptides. This study emphasizes the importance of sequence pattern information for micropeptide classification and proves that machine learning models can be used for QSP prediction in the future.