Abstract 1453: Association of interpretable histomorphic features with molecular markers: A Computational Histology Artificial Intelligence (CHAI) biomarker development platform analysis.

Abstract Background: Biomarkers are needed to guide precision oncology. Ideally, tools have rapid turnaround and integrate into existing workflows. Computational pathology analyzing features on routine hematoxylin and eosin (H25,000 pan cancer H500,000 pathologist annotated nuclei and 100 million μm2 tissue, with macro AUC 0.99 for nuclei and tissue segmentation. These models process H30,000 histomorphologic features representing hallmarks of cancer biology (e.g. nuclei size and shape, spatial arrangement, immune infiltration, stromal density). On 41 colorectal cancer WSIs with matched MIF scans from the ORION dataset, CHAI cell/tissue-typing performance was evaluated against MIF by calculating Pearson correlation coefficient between the densities of CHAI’s predicted target and matched MIF markers. Results: Across 289,619 200 x 200 μm2 tissue patches, cell segmentation correlation comparing CHAI-segmented nuclei vs MIF DAPI was 0.899 (95% CI 0.898, 0.900). Cell typing correlation was 0.642 (0.640, 0.644) for CHAI-predicted epithelial cells vs MIF cytokeratin (CK) and 0.587 (0.584, 0.590) for CHAI pan-leukocyte vs MIF CD45. Tissue typing correlation was 0.656 (0.653, 0.658) for CHAI tumor-epithelial regions vs MIF CK and 0.543 (0.541, 0.546) for CHAI stromal regions vs MIF smooth muscle actin. All correlations were statistically significant (p0.001). Conclusion: The CHAI platform measures histomorphic features from H it captures facets of the tumor microenvironment that showed significant overlaps with molecular markers captured by MIF, while also identifying areas to further explore complementarity in these orthogonal modalities. This work underscores the biologic basis of the CHAI system as a novel modality for biomarker development with potential for biomedical research and clinical applications. Citation Format: Asit Tarsode, Haochen Zhang, Viswesh Krishna, Vrishab Krishna, Snehal S. Sonawane, Lesli A. Kiedrowski, Trevor J. Royce, Anirudh Joshi, Richard M. Goldberg, Eric A. Collisson. Association of interpretable histomorphic features with molecular markers: A Computational Histology Artificial Intelligence (CHAI) biomarker development platform analysis abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 1453.