The invention discloses a phage
receptor binding protein host prediction method based on cross-
modal multi-source transfer learning, which is characterized in that the host prediction precision is enhanced by utilizing
chemical toxicity data, and the challenge of scarcity of phage sequence data is overcome. First, a shared bidirectional long short-
term memory network
encoder is used to process features from chemical and biological modalities, and structure and sequence information is captured. Next,
domain adaptation is performed through a maximum
mean difference loss function, effectively aligning chemical and biological feature representations. According to the method, knowledge migration of cross-
modal data is realized, the accuracy of host prediction is remarkably improved, and particularly, the precision is improved by about 30% compared with that of a traditional model in the early stage of training. Experimental results show that the method provided by the invention can realize
rapid convergence and is excellent in performance when being used for
processing the problem of data scarcity, a novel efficient method is provided for host prediction in the field of biological medicines, especially in
phage therapy, and the method has a wide application prospect.