Abstract 4719: Discovery of potential synergistic targets for PKMYT1 inhibitor through AI-assisted drug combination screening in an MPS-based vascularized metastatic gastric cancer model

Abstract Introduction: Many approvals of advanced drugs, such as Antibody-Drug Conjugates (ADCs) or immunotherapies, have not provided significant treatment benefits for patients with metastatic gastric cancer (GC). There remains an urgent need to develop novel therapies. However, the lack of well-defined oncogene targets has hindered progress in therapy development. A patient-like disease model was established using patient samples and a 3D co-culture platform, the Microphysiological System (MPS). Patient cells in this model preserved key properties and successfully replicated clinical phenotypes in the model. Promising therapeutic targets were identified through chemogenomic screening. To address resistance to single-target inhibition, potential combination targets were predicted using AI tools enhanced with target-based knowledge graphs (KG) and active learning algorithms. The top 20 combination targets were selected and tested, resulting in promising combination hits that demonstrated significant efficacy for this challenging and heterogeneous disease. The integration of AI-assisted combination discovery with the patient-like disease model has uncovered new opportunities for developing novel therapies and combination treatments. Methods: For the vascularized GC model, patient cells, isolated from surgical specimens or ascitic fluid, were expanded and formed into spheroids. These spheroids were co-cultured with Human Umbilical Vein Endothelial Cells (HUVECs) in Curiochips by embedding them in a hydrogel. The chemogenomic library was developed based on pathway analyses of RNAseq and proteomics studies of advanced gastric cancer. Additionally, the combination library was designed by the process utilized AI for image feature extraction, supplemented by a target-based knowledge graph (KG) that integrated data from over 20 public databases, encompassing genomics, proteomics, drug interactions, and pathway information. All screens were performed using high-throughput screening (HTS) instruments, including high-content confocal microscopy, leveraging the HTS compatibility of Curiochips. Results: Metastatic GC patient samples exhibited clinical phenotypes of aggressive migration and angiogenesis when co-cultured with endothelial cells in 3D. From pathway analysis, 200 targets were identified, and their inhibitors were screened in the vascularized model. The top targets were prioritized based on image analysis of GC phenotypic landscape. Among these, PKMYT1 was selected as a test case for identifying potential combination targets. Using the target-based KG, 20 top targets for combination with the PKMYT1 inhibitor were identified and screened in the disease model. The combination demonstrated promising efficacy and significantly delayed the onset of drug resistance. Citation Format: Yeongmin Choi, Sumin Kim, Yangji Kim, Youngsook Song, Jinhwan Jang, Delaney Donnelly, Bushra Rajput, Tsung-Li Liu, Kyusuk Baek, Aneesh Sathe, Sanghee Yoo. Discovery of potential synergistic targets for PKMYT1 inhibitor through AI-assisted drug combination screening in an MPS-based vascularized metastatic gastric cancer model [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 4719.