Abstract
Kinase fusion genes were the most targeted fusion gene group among multiple major cellular gene groups. Kinase inhibitors disrupt aberrant signaling cascades and inhibit tumor progression, yet the specific mechanisms of action of the U.S. Food and Drug Administration (FDA)-approved inhibitors in the context of kinase fusion oncoproteins remain largely unknown. This gap limits our ability to develop personalized therapies and next-generation kinase inhibitors. To address this, we developed a novel in silico pipeline for predicting 3D structures of kinase fusion proteins and performing structure-based virtual screening. This approach enables large-scale structural annotation and drug screening across pan-cancer kinase fusions. We present KinaseFusionDB, available at https://compbio.uth.edu/KinaseFusionDB, a comprehensive knowledgebase providing functional annotation of 7680 kinase fusion genes, 1399 predicted fusion protein structures, predicted Local Distance Difference Test (pLDDT)-based confidence scoring, and virtual screening data using FDA-approved kinase inhibitors. Our analysis revealed that most predicted structures showed high pLDDT scores (pLDDT >70) within conserved kinase domains. Structural alignment with known Protein Data Banks demonstrated shared structural motifs despite variation in fusion breakpoints. Virtual screening results highlighted repurposing opportunities and isoform-specific binding preferences. KinaseFusionDB is a valuable resource for investigating kinase fusion structure-function relationships and guiding the design of personalized and next-generation kinase inhibitor therapies.