Machine learning-based classification of viral membrane proteins

Viral membrane proteins (VMPs) are essential viral components located on the surface of enveloped viruses, playing a critical role during the viral infection of host cells. Based on their structural characteristics, membrane localization, and functional properties, VMPs can be categorized into lipid anchor chains, multichannel membrane proteins, peripheral membrane proteins, and single-channel proteins. These classifications directly correlate with their biological functions. Therefore, accurately identifying and analyzing the functional types of VMPs is crucial for advancing antiviral drug discovery and vaccine development. In this study, we developed a multiclass machine learning (ML) model to classify VMPs types based on sequence-derived information. The proposed approach integrated compositional and physicochemical features with a transformer-based multi-layer perceptron (MLP) model. Its performance was benchmarked against conventional ML classifiers, including support vector machines (SVM), decision trees (DT), random forests (RF), and naive bayes (NB). Our transformer-based MLP model significantly outperformed these baseline classifiers, achieving an overall area under the receiver operating characteristic curve (AUC) of 0.94 and an overall area under the precision-recall curve (AUPRC) of 0.89. These promising results highlight the effectiveness of transformer architectures in classifying VMPs by capturing subtle yet critical feature relationships necessary for accurate predictions. Ultimately, our findings contribute valuable insights toward antiviral drug development and effective disease-control strategies.