Many　traditional　drug　prediction　models　mostly　focus　on　analyzing　single　omics　data,　while　ignoring　the　rich　multi-omics　data　in　bioinformatics.　Moreover,　they　failed　to　make　full　use　of　the　drug　complementary　information　of　sequence　features　and　graphical　features　when　considering　the　SMILES(Simplified　molecular　input　line　entry　system)　features.　In　view　of　this,　we　propose　a　deep　learning　model　GSDRP　that　effectively　integrates　omics　data　and　drug　attribute　information.　SA-BiLSTM　is　used　to　extract　one-dimensional　sequence　features　of　drugs,　and　Graph　Transformer　and　GAT_GCN　capture　two-dimensional　structural　features,　which　are　then　fused　by　the　graph　sequence　attention　module.　At　the　same　time,　the　omics　data　features　are　processed　by　convolutional　neural　network.　Finally,　the　cross-attention　module　in　GSDRP　facilitates　the　fusion　of　omics　and　drug　features　to　enhance　interactions　for　better　prediction.　Experiments　on　the　Cancer　Drug　Sensitivity　Database　(GDSC)　show　that　GSDRP　can　effectively　combine　multi-omics　information　such　as　genome　aberration　(MUT_CNA)　and　gene　expression　(GE)　with　the　1-D　and　2-D　features　of　SMILES　to　significantly　improve　the　accuracy　of　drug　response　prediction.　The　comparison　results　with　four　other　state-of-the-art　methods　further　demonstrate　the　superior　performance　of　GSDRP　in　drug　response　prediction.　In　addtion,we　identify　important　omics　markers　and　important　characteristics　of　drugs　that　affect　HCC　celllines　response　prediction.　This　will　not　only　help　to　understand　the　therapeutic　mechanism　of　hepatocellular　carcinoma,　but　also　provide　strong　support　for　future　individualized　treatment.