Pulmonary infections are a major global health problem, particularly in patients with chronic diseases such as cystic fibrosis and bronchiectasis, where structural abnormalities and impaired mucociliary clearance lead to recurrent infections. These infections are often complicated by antibiotic resistance, making treatment difficult. Phage therapy appears to be a promising alternative for resistant lung infections. The integration of artificial intelligence has improved phage selection, but the accuracy of predicting phage-bacterial host interactions is still limited. The authors propose a deep Siamese network framework for precise phage selection in lung infections. Phage and host genome sequences are coded using k-mer segmentation and the skip-gram model, then processed by convolutional neural networks (CNNs) and transformers to extract local and global features. The extracted features are fused to predict phage-host interactions. Experiments on a dataset from the NCBI database showed excellent performance of the method in identifying phages targeting specific bacterial hosts, supporting its use in phage therapy of lung infections.