IA-Transformer: Prediction and Classification of β-Lactamase Proteins Using Transformer Model with Integrated Attention Mechanism

β-Lactamase proteins are the primary mediators of bacterial resistance to β-Lactam antibiotics, posing a severe threat to global public health. Accurate prediction and classification of β-Lactamase proteins are crucial for the development of novel antibiotics and the formulation of clinical treatment strategies. Traditional machine learning methods for β-Lactamase analysis often rely on manual feature engineering, which fails to fully capture the complex sequence patterns and contextual information of proteins. To address this limitation, this study proposes a Transformer model integrated with a multi-head attention mechanism (IA-Transformer) for the prediction and classification of β-Lactamase proteins. The IA-Transformer model innovatively integrates three attention modules: sequence-wise self-attention, residue-wise attention, and channel-wise attention. The sequence-wise self-attention captures long-range dependencies between amino acid residues in the protein sequence; the residue-wise attention emphasizes key functional residues related to β-Lactam hydrolysis; and the channel-wise attention optimizes the feature representation of different sequence motifs. Experimental results show that the IA-Transformer model achieves an accuracy of 98.2%, a sensitivity of 97.8%, a specificity of 98.5%, and an F1-score of 98.0% in β-Lactamase prediction, outperforming traditional methods such as SVM, Random Forest, and single-attention Transformer by 3.5% − 7.2%.