Identifying effective druggable targets with disease-specific for diseases is a tremendous challenge in new drug development. However, current studies of druggable targets identification are most based on either druggability or disease-specific, lacking a combination of two factors. To further improve the accuracy of druggable targets discovery, a druggable target discovery strategy for diseases (DTDS) was proposed, which combined druggable targets prediction by machine learning and key targets identification by tissue-level and cellular-level transcriptomics analysis. Rheumatoid arthritis (RA), an autoimmune disease that cannot be treated entirely, was taken as a case. First, the protein-protein interaction network was constructed as the disease background network, and the classification models were established based on the topological parameters of known RA-druggable targets with druggability and non-RA targets without therapeutic effects on RA. 168 potential druggable targets were predicted by the classification models from 264 RA-related targets. Subsequently, 40 RA-specific targets were identified by tissue-level and cellular-level transcriptomics analysis from 168 potential druggable targets. Most of them were RA-druggable targets except PSMB9 and PTPRC. Finally, PSMB9 and PTPRC were further verified by in vitro experiments. The results showed that the inhibitor of PSMB9 or PTPRC could effectively inhibit inflammation and abnormal proliferation of synovial cells, proving that PSMB9 and PTPRC were potential RA-druggable targets, and further indicating that DTDS had high accuracy. In conclusion, the DTDS strategy established in this study is reliable and has been proven in identification of potential RA-druggable targets, which is expected to provide ideas and methods for systematic discovery of potential druggable targets for diseases.