PhageVir: An evaluation of computational intelligence models for the precise identification of phage virion proteins

This study presents PhageVir, an enhanced computational model developed to predict Phage Virion Proteins (PVPs), which are essential for bacteriophage infection and replication. PhageVir integrates advanced feature selection methods, including the Position Relative Incidence Matrix (PRIM) and the Reverse Position Relative Incidence Matrix (RPRIM), to effectively capture key sequence features and positional dependencies within protein sequences. Several machine learning and deep learning algorithms were employed, including LightGBM, Random Forest, Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Recurrent Neural Network (RNN), and Artificial Neural Network (ANN), to classify PVPs based on sequential data. Model performance was evaluated through independent set testing, self-consistency testing, and cross-validation, using metrics such as accuracy (ACC), specificity (Sp), sensitivity (SN), Z-score, and Matthews correlation coefficient (MCC). The CNN model demonstrated strong performance in cross-validation, achieving an accuracy of 0.833, sensitivity of 0.832, specificity of 0.834, a correlation coefficient of 0.665, an AUC score of 0.927, and a Z-score of 1.37. The results confirm the effectiveness of the proposed computational approach for accurate PVP classification. Beyond its predictive power, PhageVir offers valuable biological insights into phage infection mechanisms, supporting advancements in phage therapy and antibacterial treatments.