Abstract B061: Path2Omics–TIME_ACT: Pan-cancer prediction of tumor hotness and immune checkpoint blockade response from pathology slides

Abstract Background: Immune checkpoint blockade (ICB) has reshaped cancer treatment, yet only a minority of patients respond. This underscores the need for robust biomarkers capable of broadly identifying tumors with high immune activation. Existing immune gene signatures and transcriptomic predictors often lack generalizability across cancer types and independent cohorts. To overcome this limitation, we developed TIME_ACT, an unsupervised 66-gene signature that quantifies tumor immune activation (“hotness”) by integrating signals of immune infiltration, inflammatory activity, and spatial TIL density. TIME_ACT provides a consistent, pan-cancer measure of immune activity and is highly inferable from pathology slides, enabling a low-cost and widely accessible approach for predicting tumor hotness and ICB response using routine histopathology. Methods: We first evaluated TIME_ACT in 22 publicly available pre-treatment ICB transcriptomic cohorts (n=1,416) across seven cancer types. We then combined TIME_ACT with Path2Omics, a recently published deep-learning model that imputes gene expression from pathology slides, enabling direct prediction of ICB response from pathology. This integrated framework, Path2Omics-TIME_ACT, was tested across nine external treatment cohorts (n=459). Results: Using measured gene expression data, TIME_ACT accurately distinguished immune-hot versus immune-cold tumors with AUCs above 0.96 in all 10 evaluated cohorts, and achieved a mean AUC of 0.74 and a mean odds ratio of 5.49 across 22 transcriptomic cohorts, outperforming 26 established ICB response signatures and methods. To assess slide-based prediction, we evaluated Path2Omics performance on TIME_ACT genes. Strikingly, Path2Omics achieved strong predictive fidelity, with a median correlation of 0.61 between actual and imputed gene expression values across 24 cancer types. Using the slide-inferred TIME_ACT scores, Path2Omics-TIME_ACT achieved performance nearly equivalent to that obtained with measured gene expression, with AUCs above 0.94 in all 8 evaluated cohorts for hot/cold classification, and a mean AUC of 0.73 and a mean odds ratio of 5.07 across nine histopathology cohorts for predicting ICB response. Notably, Path2Omics was trained on TCGA data and applied to external treatment cohorts without further fine-tuning, and TIME_ACT is unsupervised learning approach, alleviating concerns about overfitting. Conclusions: Path2Omics-TIME_ACT is a robust pan-cancer predictor of tumor immune hotness and ICB response directly from routine pathology slides. With further prospective testing and validation, this framework could significantly broaden access to precision immunotherapy. Citation Format: Sumit Mukherjee, Sumeet Patiyal, Lipika R. Pal, Tian-Gen Chang, Sumona Biswas, Saugato Rahman Dhruba, Amos Stemmer, Arashdeep Singh, Tien-Hua Chen, Binbin Wang, Denis Marino, Wonwoo Shon, Yuan Yuan, Mark Faries, Omid Hamid, Karen Reckamp, Keluo Yao, Sabina Signoretti, David A. Braun, Joo Sang Lee, Hyunjeong Joo, Hyungsoo Kim, Jair Bar, Iris Barshack, Chi-Ping Day, Sridhar Hannenhalli, Karine Sargsyan, Andrea B. Apolo, Kenneth Aldape, Muh-Hwa Yang, Michael B. Atkins, Ze’ev A. Ronai, Danh-Tai Hoang, Eytan Ruppin. Path2Omics–TIME_ACT: Pan-cancer prediction of tumor hotness and immune checkpoint blockade response from pathology slides [abstract]. In: Proceedings of the AACR Immuno-Oncology Conference (AACR IO): Discovery and Innovation in Cancer Immunology: Revolutionizing Treatment through Immunotherapy; 2026 Feb 18-21; Los Angeles, CA. Philadelphia (PA): AACR; Cancer Immunol Res 2026;14(2 Suppl):Abstract nr B061.