A Gene Similarity Algorithm Based on Autocorrelation of Diseases and Phenotypes

Gene similarity mining through meta-paths in the heterogeneous network composed of genes-disease-phenotypes is one of the important methods in the field of gene similarity search. Since the current meta-path-based algorithms do not consider the influence of autocorrelation between disease and disease, and between phenotype and phenotype on gene similarity, the potential and implicit similar genes in the heterogeneous network cannot be fully mined. To address this issue, this paper first explores the autocorrelation expressions of diseases and phenotypes by constructing a semantic contribution graph and calculating the similarity of Gaussian kernel topology, and then by integrating the autocorrelation into the heterogeneous network, a fully linked network model with correlation weights containing autocorrelation information between diseases and phenotypes is constructed. This network is then utilized to calculate gene similarity based on the gSim-Search principle. The proposed method makes full use of autocorrelation information between diseases and phenotypes to search for potentially similar genes, and indirectly solves the problem of sparseness of heterogeneous network links. Experiments show that the proposed method improves the accuracy of prediction when calculating and ranking the similarity of the two pathogenic genes of breast cancer and obesity.