{
  "abstract": "Background High costs and technical demands continue to limit the routine use of RNA-seq in clinical oncology, restricting access to transcriptomic insights in standard care. DeepPT, our deep learning approach, 1 addresses this limitation by inferring genome-wide gene expression directly from readily available H&E-stained histopathology slide scans. By leveraging slides already used in clinical workflows, DeepPT can be used for pathway analysis without requiring molecular profiling. Here, we demonstrate how expression inference by DeepPT can reveal clinically relevant tumor characteristics, supporting its potential role in diagnosis and treatment guidance.Methods To evaluate the ability of DeepPT to derive clinical insights from pathway-level analysis, we analyzed 119 cancer hallmark pathways 2 in TCGA-BRCA and TCGA-LUAD. To estimate the pathway activity in a sample from its H&E slides, we calculate the Normalized Enrichment Scores by applying Gene Set Enrichment Analysis3–6 on the gene expression inferred by DeepPT (denoted inferred-NES). This enabled integrated analysis linking pathway activity to key clinical features related to immune response, including immune infiltration, tumor mutational burden (TMB), and TP53 mutation status.Results We first examined whether pathway analysis from H&E slides can capture tumor ‘hotness’. Reassuringly, pathways related to immune infiltration had markedly higher inferred-NES in ‘hot’ relative to ‘cold’ tumors (using tumor ‘hotness’ classifications from). 4 Among these, the ‘Downstream Signaling in Naive CD8+ T-cells’ pathway stood out (figure 1), showing significant differences between ‘hot’ and ‘cold’ tumors in TCGA-BRCA and TCGA-LUAD, and strongly correlating with the abundance of immune infiltration estimates in the TransNEO cohort.5 This consistency across cohorts provides strong evidence that DeepPT can identify the key upregulated immune pathways in the TME.The inferred-NES of pathways from the ‘Genome Instability and Mutation’ - a key cancer driver - showed strong correlations with measured TMB and deleterious TP53 mutations. Figure 2 shows the strong associations between the inferred-NES of the ‘FOXM1 Transcription Factor Network’ pathway with TMB and TP53 mutation status across three separate cohorts. Importantly, the strength of the association between the inferred-NES and the clinically-relevant features, were on par with the strength of these associations based on NES calculated from RNA-seq data.Conclusions We demonstrated that DeepPT-inferred NES scores, calculated from routine H&E slides, are associated with clinically-relevant tumor characteristics, including immune infiltration, TMB levels, and TP53 mutation. Elucidating these properties currently requires much more expensive and demanding techniques. Our methods could support accessible immune characterization and treatment decision-making.Acknowledgements Eytan Ruppin Cancer Data Science Laboratory, Center for Cancer Research, National Cancer Institute, National Institutes of Health, Bethesda, USAReferences Hoang DT, Dinstag G, Shulman ED, Hermida LC, Ben-Zvi DS, Elis E, Ruppin E. A deep-learning framework to predict cancer treatment response from histopathology images through imputed transcriptomics. Nature Cancer. 2024;5(9):1305–1317.Iorio F, Garcia-Alonso L, Brammeld JS, Martincorena I, Wille DR, McDermott U, Saez-Rodriguez J. Pathway-based dissection of the genomic heterogeneity of cancer hallmarks’ acquisition with SLAPenrich. Scientific Reports. 2018;8(1):6713.Mootha VK, Lindgren CM, Eriksson KF, Subramanian A, Sihag S, Lehar J, Groop LC. PGC-1α-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nature Genetics. 2003;34(3):267–273.Saltz J, Gupta R, Hou L, Kurc T, Singh P, Nguyen V, Danilova L. Spatial organization and molecular correlation of tumor-infiltrating lymphocytes using deep learning on pathology images. Cell Reports. 2018;23(1),181–193.Sammut, et al. Nature. 2022;601:623–629. Sammut SJ, Crispin-Ortuzar M, Chin SF, Provenzano E, Bardwell HA, Ma W, Caldas C. Multi-omic machine learning predictor of breast cancer therapy response. Nature. 2022;601(7894):623–629.Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Mesirov JP. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proceedings of the National Academy of Sciences. 2005;102(43):15545–15550.Abstract 1080 Figure 1NES based on DeepPT-inferred expression of the ‘Downstream signaling in Naive CD8+ T-cells’ pathway are associated with tumor hotness. Left and middle: DeepPT-inferred NES for the CD8+ T-cell pathway in TCGA-LUAD and TCGA-BRCA, showing higher values in ‘hot’ tumors. Right: Inferred-NES in TransNEO correlates with independent immune cell abundance estimatesAbstract 1080 Figure 2DeepPT-inferred NES of genome instability pathway is associated with TP53 status and TMB. Top: Inferred-NES of the FOXM1 pathway is positively correlated with TMB and elevated in TP53-mutant TCGA-LUAD samples. Bottom: Similar patterns observed in TCGA-BRCA and TransNEO, showing TP53-associated differences in inferred-NES",
  "authors": [
    {
      "affiliations": [
        "Pangea Biomed, Tel Aviv, Israel"
      ],
      "name": "Rachel E Bell"
    },
    {
      "affiliations": [
        "Pangea Biomed, Tel Aviv, Israel"
      ],
      "name": "Gal Dinstag"
    },
    {
      "affiliations": [
        "Pangea Biomed, Tel Aviv, Israel"
      ],
      "name": "Yaron E Kinar"
    },
    {
      "affiliations": [
        "Pangea Biomed, Tel Aviv, Israel"
      ],
      "name": "Omer Tirosh"
    },
    {
      "affiliations": [
        "Pangea Biomed, Tel Aviv, Israel"
      ],
      "name": "Maayan Chechik"
    },
    {
      "affiliations": [
        "Pangea Biomed, Tel Aviv, Israel"
      ],
      "name": "Doreen S Ben-Zvi"
    },
    {
      "affiliations": [
        "Pangea Biomed, Tel Aviv, Israel"
      ],
      "name": "Tuvik Beker"
    },
    {
      "affiliations": [
        "Pangea Biomed, Tel Aviv, Israel"
      ],
      "name": "Ranit Aharonov"
    }
  ],
  "title": "1080 DeepPT enables fast and low-cost pathway-based tumor immune characterization from histopathological slides",
  "uid": "fef714a5-3d8a-5307-91d7-9786ef025d46"
}