A Disease-related Gene Mining Method Based On Weakly Supervised Learning Model

Prediction of disease-related genes is helpful to understand the molecular mechanisms during the disease progression. However, since the disease data may contain some samples with weak labels and the number of known disease- related genes is quite small, the applications of traditional gene mining methods have certain limitations. Therefore, this paper designs a disease-related gene mining method based on weakly supervised learning model. The method contains two parts. Firstly, screening differentially expressed genes based on the weakly supervised learning model makes full use of the strong and weak label information at different stages of disease progression. The gene set obtained by the algorithm is more stable after convergence. Then, the disease-related genes are predicted in the differentially expressed gene set using transductive support vector machine, in which the difference kernel function can map the input space of the original gene expression data to the difference space. In the difference space, the affinity between the two genes can be measured more accurately. So the known disease-related gene information can be effectively utilized. Using Huntington’s disease gene expression data, this paper compares the designed method with other excellent methods and verifies the validity and accuracy of the method.