HIV-1 integrase is the third enzyme essential for viral replication. It represents an attractive target for new anti-HIV drugs. Diketo acids represent the most prospective class of integrase inhibitors; one of them (S1360) is currently under clinical trials. We prepared the training set containing 90 diketoacid derivatives and performed re-training of computer program PASS. Average accuracy of prediction in LOO cross-validation for HIV-1 integrase inhibitors was shown to be 99%. We selected eight new potential HIV-1 integrase inhibitors among about the 700,000 substances from the databases of commercially available chemicals; anti-integrase activity was found experimentally in two selected compounds.