Abstract 754: AI-powered spatial inference platform for qualifying ADCs and multispecific therapeutic targets

Abstract Over the past two decades, genomic alterations have become important biomarkers for predicting response to targeted therapies; however, the emergence of immunotherapy, antibody-drug conjugates and multispecifics may require a different class of biomarkers that can quantify and localize drug targets within tumor and immune cells. Genomic testing frequently lacks correlation with protein expression, fails to capture membranous versus cytoplasmic localization, and has limited ability to decipher spatial interactions between targets. Multiplex immunofluorescence (mIF) is an established technology capable of quantifying and capturing the subcellular localization of dozens of proteins (>25 proteins per slide), enabling parallel assessment of a large number of drug targets. Here we describe a framework for leveraging high-plex spatial proteomics with an AI-driven platform to develop rational approaches for evaluating individual targets and combinations of targets within and across disease indications. We designed a 30-plex mIF panel to perform sequential immunofluorescence (seqIF™) on the COMET™ platform to profile known tumor antigens and targets (n = 18) currently in development along with the tumor microenvironment (TME), immune cells, and key signaling cascades (n = 12). We tested this panel on 217 samples (>1, 700, 000 cells) from lung (n = 90), gastric (n = 11), and colorectal (n = 68) cancers, as well as normal tissue specimens (n = 48, 22 sites). Utilizing Nucleai’s deep-learning-based multiplex imaging analysis pipeline, we quantified the protein abundance, localization, and interactions among 9 different cell types. Tissue areas were divided into subregions and neighborhoods to characterize localization of these cell types to distinct areas including the tumor core (TC), tumor invasive margin interface (TIM), and stroma. We confirmed known drug targets with historical approvals in each of these indications, which validated that the pipeline was able to correctly identify key targets based on subcellular localization, spatial proximity analysis, and analysis of off-target cell type expression. We also identified potential combination partners for multi-specifics based on target colocalization, which were found to have reductions in off-target cell colocalization, as determined by protein expression in the right compartment and in the expected cell types. Nucleai’s platform enables deep spatial proteomic profiling by deriving AI-driven insights for actionable characterization of protein interactions and patient heterogeneity across tissues. We demonstrate that this framework can aid in the prioritization and derisking of potential drug targets used by key classes of therapeutics currently in development. Citation Format: Michael Mints, Ettai Markovits, Ron Elran, Shai Bookstein, Yuval Shachaf, Elad Amsalem, Mor Kenigsbuch, Jason W. Reeves, Ken Bloom, Oscar Puig. AI-powered spatial inference platform for qualifying ADCs and multispecific therapeutic targets [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 754.