Abstract IA05: Computational pathology in ADC drug development

Abstract Technological advances in computer vision and artificial intelligence (AI) are reshaping how therapeutic targets are assessed in cancer tissues. When applied to immunohistochemistry (IHC), AI-based computational pathology delivers higher precision, accuracy, reproducibility, and a broader dynamic range of target quantification than visual scoring alone. Crucially for antibody–drug conjugates (ADCs), AI can measure membrane and cytoplasmic expression with continuous resolution, capture intratumoral heterogeneity, and extract biologically meaningful, human-interpretable features that inform drug design, patient selection, and response prediction. We will present the Quantitative Continuous Score (QCS), a strongly supervised AI method trained on expert pathologist–annotated whole-slide images. QCS first detects and segments invasive tumor regions, then quantifies therapeutic targets within specific subcellular compartments (membrane, cytoplasm) at single-cell resolution. Using pre-specified, human-interpretable features, QCS not only measures target intensity and prevalence but also infers attributes directly relevant to ADCs, including target internalization, cell-to-cell variability, and spatial heterogeneity across tumor and microenvironment. These capabilities can directly address several ADC development bottlenecks: 1.) Target selection and validation: By quantifying membrane staining intensity and internalization surrogates, QCS prioritizes targets with favorable binding–internalization–trafficking profiles, de-risking early discovery; 2.) Linker–payload optimization: Spatial maps of expression and internalization guide the choice of payload potency and linker stability to balance on-target efficacy and bystander effect, especially in heterogeneous tumors; 3.) Patient enrichment and trial design: Continuous, standardized QCS metrics support companion diagnostic cut-offs that reflect true biological gradients rather than coarse ordinal scores, increasing the probability of technical and clinical success; 4.) Resistance mechanisms: Longitudinal QCS analyses could reveal shifts in subcellular localization, loss of membrane expression, or altered heterogeneity that signal adaptive resistance and inform next-line strategies; and 5.) Manufacturing and site-to-site harmonization: Robust, reproducible quantification across scanners, stains, and sites improves assay transferability, enabling global trials with consistent eligibility criteria. Beyond quantification, in the future QCS-derived features can be integrated with genomics, transcriptomics, and pharmacological data to build multimodal models of ADC response. For example, cases with high membrane expression but low inferred internalization may benefit from alternative linkers or payloads, whereas tumors with pronounced heterogeneity might require payloads that permit bystander killing. Because QCS outputs are continuous and interpretable, they are suitable for regulatory-grade biomarker development and can evolve into scalable CDx solutions. Citation Format: Mark Gustavson. Computational pathology in ADC drug development [abstract]. In: Proceedings of the AACR-NCI-EORTC International Conference on Molecular Targets and Cancer Therapeutics; 2025 Oct 22-26; Boston, MA. Philadelphia (PA): AACR; Mol Cancer Ther 2025;24(10 Suppl):Abstract nr IA05.