To　solve　the　problem　of　how　to　screen　out　more　effective　antiviral　compounds,　the　relationship　between　dissociation　rate　(koff)　and　antiviral　drug　structure　was　studied　in　this　research.　The　theoretical　basis　is　that　koff　is　often　used　to　evaluate　the　activity　of　the　drug　in　the　open　system　of　the　human　body.　The　molecular　descriptor　of　each　antiviral　compound　was　calculated　by　using　molecular　descriptor　software,　and　the　descriptors　were　screened　by　multiple　stepwise　regression　analysis,　partial　least　squares　method　and　genetic　algorithm.　Then,　the　support　vector　machine　(SVM)　and　BP　neural　network　were　used　to　establish　the　prediction　model　of　antiviral　compound　structure　and　dissociation　rate　koff　value,　and　the　model　was　verified.　Results　show　that　this　experiment　screens　out　the　descriptors　with　good　predictive　power,　and　the　two　predictive　models　are　proved　to　be　reasonable　and　have　guiding　significance　for　the　future　development　of　antiviral　drugs.　©　2017,　Editorial　Department　of　Journal　of　Beijing　University　of　Technology.　All　right　reserved.