Diabetic retinopathy is a common complication of diabetes mellitus that causes pathogenic damage to the retina. Particularly, the proliferative diabetic retinopathy (PDR) state can cause abnormal angiogenesis in the retina tissues and trigger the retina destruction in advanced stage. In the clinic, the symptoms during the initiation and progression of PDR are relatively unrecognizable. Therefore, various studies have focused on the pathogenesis of PDR. According to published literature, genetic contributions play an irreplaceable role in the initiation and progression of PDR. Although many computational methods, such as shortest path- and random walk with restart-based methods, have been applied in screening the potential pathogenic factors of PDR, advanced computational methods, which may provide essential supplements for previous ones, are still widely needed. In this study, a novel computational method was presented to infer novel PDR-associated genes. Different from previous methods, the method used in this work employed a different network algorithm, that is, the Laplacian heat diffusion algorithm. This algorithm was applied on the protein-protein interaction network reported in the STRING database. Three screening tests were performed to filter the most likely inferred genes. A total of 26 genes were accessed using the proposed method. Compared with the two previous predictions, most of the identified genes were novel, and only one gene was shared. Several inferred genes, such as CSF3, COL18A1, CXCR2, CCR1, FGF23, CXCL11, and IL13, were related to the pathogenesis of PDR.