The invention relates to the technical field of
bioinformatics, in particular to a
drug response prediction method based on
knowledge graph and multi-view comparative learning, which comprises the following steps: acquiring
drug response information and a target
knowledge graph, and further constructing low-order and high-order views according to the
drug response information and the target
knowledge graph; extracting the embedded representation of each entity in the low-order view and the high-order view by using the knowledge graph
attention network, and carrying out iterative updating to generate the embedded representation of the target
biological entity; performing comparative learning tasks in a
single view and among a plurality of views in the low-order view and the high-order view at the same time to update the embedded representation of the entity; and carrying out
nonlinear transformation on the updated embedding representation to obtain a final embedding representation, and further obtaining a reaction prediction result between the drug and the
cancer cell line. According to the method, information is captured through an attention mechanism, more high-order knowledge graph information is explored by utilizing a multi-view contrast learning mechanism, and the response of a
cell line to a drug can be accurately predicted.