The　selection　of　drug　targets　is　a　key　link　in　drug　design.　The　correct　selection　of　targets　depends　on　biomolecular　activity,　which　refers　to　the　ability　or　property　of　interaction　between　biomolecules　(such　as　proteins,　enzymes,　receptors,　etc.)　and　other　molecules　(such　as　drugs,　compounds).　It　describes　the　effects　and　reactions　of　biomolecules　on　other　molecules　within　an　organism　or　in　a　laboratory　environment.　Biomolecular　activity　can　include　multiple　aspects　of　properties　and　effects,　such　as:　binding　activity,　enzymatic　activity,　activation　or　inhibition　activity,　cell　action　activity.　Biomolecular　activity　refers　to　the　binding　activity　of　molecules　in　drug　design,　so　the　selection　of　appropriate　targets　needs　to　predict　the　biological　activity　of　molecular　proteins　(the　binding　activity　of　molecules).　In　this　paper,　we　present　an　FPGA　hardware　accelerator　for　predicting　molecular　protein　activity　prediction,　which　deploies　lasso　and　deep　neural　network　(SED)　algorithms　for　screening　extended　connectivity　fingerprints.　In　order　to　speed　up　the　algorithm　to　process　data　more　efficiently,　it　used　the　minimum　integer　bit　width　and　decimal　point　width　to　reduce　the　hardware　processing　of　data　under　the　condition　of　reasonable　accuracy　loss,　and　used　AXI　MASTER　interface　to　improve　the　bandwidth　of　the　algorithm　to　transfer　data　from　memory.　Experimental　results　show　that　our　FPGA　accelerator　implemented　on　Xilinx　Zynq　UltraScale+　XCZU7EV　is　4.3　times　faster　than　the　algorithm　implemented　on　Intel　i7-7700K　CPU@4.2GHz.