AI-driven multi-omics drug repurposing nominates AZD7762 as a multitarget inhibitor of IL22RA1 and FAM221A in esophageal squamous cell carcinoma

Esophageal squamous cell carcinoma (ESCC) is an aggressive malignancy with limited targeted treatment options and poor clinical outcomes. We developed an AI-driven multi-omics pipeline that links prognostic modeling to multitarget drug repurposing for ESCC. Summary-data-based Mendelian randomization was integrated with bulk transcriptomic datasets to identify esophageal cancer-related druggable genes that are differentially expressed. Cox regression and non-negative matrix factorization were then used to define prognostic genes and molecular subgroups, and a Lasso Cox model with SHapley Additive explanation provided an interpretable prognostic signature. Single-cell RNA sequencing analysis mapped the hub genes interleukin 22 receptor subunit alpha 1 (IL22RA1) and family with sequence similarity 221 member A (FAM221A) to epithelial cell populations and associated them with proliferative and DNA repair programs, supporting their role in tumor progression, supporting their role in ESCC progression. To translate these targets into a therapeutic strategy, we applied machine learning-based drug sensitivity prediction, ADMET-AI toxicity, pharmacokinetic profiling, and molecular docking, which converged on the checkpoint kinase inhibitor AZD7762 (3-(carbamoylamino)-5-(3-fluorophenyl)-N-(3S)-piperidin-3-yl thiophene-2-carboxamide) as a promising multitarget inhibitor of IL22RA1 and FAM221A. In vitro assays confirmed that IL22RA1 and FAM221A promote ESCC cell proliferation, migration, and invasion. Taken together, this AI-driven multi-omics framework delivers a prognostic model, defines biologically distinct ESCC subgroups, and nominates AZD7762 as a rational multitarget drug repurposing candidate, providing a precision oncology strategy.