A Computationally De-Risked CD47 Inhibitor for Glioblastoma

Abstract Background Glioblastoma (GBM) is a lethal primary brain tumor that leverages the CD47-SIRPα "don't eat me" signal to evade the innate immune system. First-generation antibody therapies targeting this pathway have failed due to significant hematological toxicity and limited efficacy, creating an urgent need for safer, more effective therapeutic modalities. Methods I employed a "computation-first" drug discovery pipeline to screen a novel library of 525 secreted cephalopod proteins against human CD47. The lead candidate was identified via AlphaFold2 structure prediction and ClusPro docking, followed by a comprehensive in silico immunogenicity assessment using the IEDB and NetMHCpan servers. A rational deimmunization strategy was then executed, and the binding affinity of the engineered variants was validated using MM/GBSA calculations. Results The initial screen identified AOA_UR2 as a potent binder, but subsequent analysis revealed a catastrophic immunogenicity flaw, with a predicted 99.99% population coverage. A rational engineering campaign produced the double-mutant, AOA_UR2_F7P_M16C, which emerged as the optimal candidate. This engineered protein, which lacks an Fc region, retains a strong predicted binding free energy of -45.05 kcal/mol. The final in silico safety profile is vastly improved, with the universal 99.99% population coverage of the parent molecule being reduced to a predicted low, non-universal coverage driven by a few minor, sporadic epitopes. Conclusion AOA_UR2_F7P_M16C represents a novel molecular class of CD47 inhibitor, computationally validated as a lead candidate with high predicted potency and significantly reduced immunogenic potential. Its non-Fc-mediated binding suggests a promising avenue for GBM therapy, warranting preclinical development and experimental validation.