8533 Background: Response to first-line standard-of-care (SoC) chemo-immunotherapy (IO) for patients with NSCLC without targetable mutations is heterogeneous, highlighting the need for predictive biomarkers. GEMINI (NCT05236114) integrates real-world outcomes, whole exome sequencing (WES), single-cell spatial transcriptomics (SpTx), and AI-pathology to establish a benchmarking resource and patient-level digital twins, enabling back-translation into testable hypotheses. With >4 million cells from 53 biopsies, GEMINI is one of the largest single-cell spatial datasets linked to IO outcomes. Methods: Patients with metastatic NSCLC were analyzed for outcome associations. Progression-free survival (PFS) was defined from IO start to progression, next regimen, last follow-up, or 2 years. Patients were classified as fast progressors (<3 months PFS) or slow progressors (>3 months PFS). Baseline biopsies (n=53) underwent WES and SpTx. Neural networks traced single-cell boundaries on H&E to quantify gene expression; cells were annotated via clustering and LLM-assisted labeling. AI-, manual-, and digital-pathology (DSP) defined tumor, immune, and stroma regions. Cohort-level benchmarking was integrated into patient-level digital twins to back-translate spatial-genomic features into individualized risk and mechanism hypotheses. Results: WES revealed expected mutation frequencies: STK11 15%, TP53 73%, KEAP1 21%, KRAS 46%, supporting cohort representativeness. Stroma-associated TIL counts were higher in slow versus fast progressors by AI-path (p=0.034) and manual-path (p=0.014). DSP showed immune aggregates in slow progressors were lymphocyte-diverse, whereas fast progressors were enriched for five macrophage subtypes consistent with immunosuppressive niches. Spatial proximity of lymphocytes and stroma to a tumor subcluster (C2) predicted progression (p<0.01). Immunoglobulin light-chain expression localized to the tumor core in slow progressors, suggesting tumor–B cell interactions with disease arrest. Differential expression identified 14 EMT/ECM genes overexpressed in fast-progressor stroma, implicating stromal barrier/ECM remodeling in IO resistance. Digital twins captured these spatial-omic signatures to forecast risk and generate patient-specific, testable hypotheses. Conclusions: GEMINI provides a large single-cell spatial transcriptomic benchmark linked to IO outcomes for patients with NSCLC and enables AI-driven digital twins for clinical decision support. Fast progressors show stromal EMT/ECM programs and immunosuppressive myeloid niches; slow progressors exhibit lymphocyte diversity and tumor–B cell interactions near subcluster C2. Findings support multimodal risk stratification and nominate stromal EMT/ECM targeting to overcome IO resistance, with prospective validation via digital-twin biomarkers. Clinical trial information: NCT05236114 .

1023 Background: The antibody-drug conjugate (ADC), trastuzumab deruxtecan (T-DXd), produces meaningful responses in metastatic breast cancer (mBC), yet current biomarkers inadequately identify likely responders or elucidate resistance mechanisms. We developed a biology-informed survival model to predict T-DXd benefit in real-world data, assess treatment specificity, and characterize resistance evolution at time of progression. Methods: We analyzed RNA-seq and clinical outcomes from 150 de-identified T-DXd–treated mBC patients in the Tempus real-world multimodal database. Baseline expression profiles were embedded into biology-structured biomodules using a large molecular foundation model trained on over 1-million transcriptomes, capturing processes relevant to ADC activity: DNA damage response (DDR), intracellular trafficking, cellular stress adaptation, and tumor microenvironment signaling. These biomodules informed a survival model trained to predict T-DXd–specific clinical benefit. Predictive specificity was evaluated in a matched cohort of T-DXd–eligible mBC patients treated with taxane-based chemotherapy and HER2-targeting therapy. Longitudinal tumor profiling from 15 patients with progressive disease were analyzed to monitor resistance evolution. Results: The survival model predicted and identified biologically distinct benefit groups within the T-DXd cohort, with minimal separation observed in a control standard-of-care cohort, confirming treatment-specific predictive value (HR 2.22 [95% CI 1.14–4.35], p = 0.017). Predicted-benefit patients had longer rwTTNT (median=345 vs 245 days), and no prognostic signal appeared in control cohorts (C-index = 0.51), suggesting predictive specificity. Longitudinal analysis of progressed patients revealed reproducible biological shifts at progression, including attenuation of programs associated with effective ADC engagement and upregulation of adaptive stress-response pathways. These molecular transitions correlated with clinical deterioration and earlier transition to subsequent therapy. Conclusions: This study delineates the biology associated with durable T-DXd benefit and captures resistance evolution at progression in real-world patients. These findings highlight opportunities to refine patient selection and identify therapeutically actionable biology driving acquired resistance to T-DXd in the mBC setting.

Background Immune checkpoint blockade (ICB) has revolutionized cancer therapy, yet resistance-both primary and acquired-remains a significant obstacle, affecting the majority of patients. Methods Here, we leverage a large-scale, real-world clinicogenomic dataset to systematically explore the molecular underpinnings of ICB resistance in the post-progression setting. We analyze over 5,000 pan-cancer patients with clinical and pre-/post-treatment genomic and transcriptomic data and systematically compare the clinical and molecular features of acquired versus primary ICB resistance. Results Post-ICB progression, acquired resistance showed extended survival compared to primary resistance across all cancer types. This clinical phenotype was paralleled by a universally immune-inflamed, albeit dysfunctional, tumor microenvironment (TME) at the onset of acquired resistance, with sustained or ICB-induced inflammatory and interferon responses. We confirm previously described mechanisms of acquired resistance, including B2M loss-of-function (LoF) in non-small cell lung cancer (NSCLC), and identify novel potential mediators, including LoF of TGFBR2 in NSCLC, CYLD in head and neck cancer, and RUNX1 in triple-negative breast cancer. Further supporting their involvement in resistance, these acquired ICB alterations associated with immune-escaped TMEs, characterized by active immunomodulatory oncogenic signaling, hyperproliferation and invasiveness, or altered tumor metabolism. Conclusions These findings emphasize the heterogeneity of molecular drivers of acquired resistance to ICB within and across cancers, and highlight the potential for personalized therapeutic interventions post-progression to improve patient outcomes.

The phenomenon of mixed/heterogenous treatment responses to cancer therapies within an individual patient presents a challenging clinical scenario. Furthermore, the molecular basis of mixed intra-patient tumor responses remains unclear. Here, we show that patients with metastatic lung adenocarcinoma harbouring co-mutations of EGFR and TP53, aremore likely to have mixed intra-patient tumor responses to EGFR tyrosine kinase inhibition (TKI), compared to those with an EGFR mutation alone. The combined presence of whole genome doubling (WGD) and TP53 co-mutations leads to increased genome instability and genomic copy number aberrations in genes implicated in EGFR TKI resistance. Using mouse models and an in vitro isogenic p53-mutant model system, we provide evidence thatWGD provides diverse routes to drug resistance by increasing the probability of acquiring copy-number gains or losses relative to non-WGD cells. These data provide amolecular basis for mixed tumor responses to targeted therapy, within an individual patient, with implications for therapeutic strategies.

Inhibition of CDK4/6 kinases has led to improved outcomes in breast cancer. Nevertheless, only a minority of patients experience long-term disease control. Using a large, clinically annotated cohort of patients with metastatic hormone receptor-positive (HR+) breast cancer, we identify TP53 loss (27.6%) and MDM2 amplification (6.4%) to be associated with lack of long-term disease control. Human breast cancer models reveal that p53 loss does not alter CDK4/6 activity or G1 blockade but instead promotes drug-insensitive p130 phosphorylation by CDK2. The persistence of phospho-p130 prevents DREAM complex assembly, enabling cell-cycle re-entry and tumor progression. Inhibitors of CDK2 can overcome p53 loss, leading to geroconversion and manifestation of senescence phenotypes. Complete inhibition of both CDK4/6 and CDK2 kinases appears to be necessary to facilitate long-term response across genomically diverse HR+ breast cancers.

Combination therapy is well established as a key intervention strategy for cancer treatment, with the potential to overcome monotherapy resistance and deliver a more durable efficacy. However, given the scale of unexplored potential target space and the resulting combinatorial explosion, identifying efficacious drug combinations is a critical unmet need that is still evolving. In this paper, we demonstrate a network biology-driven, simulation-based solution, the Simulated Cell™. Integration of omics data with a curated signaling network enables the accurate and interpretable prediction of 66,348 combination-cell line pairs obtained from a large-scale combinatorial drug sensitivity screen of 684 combinations across 97 cancer cell lines (BAC = 0.62, AUC = 0.7). We highlight drug combination pairs that interact with DNA Damage Response pathways and are predicted to be synergistic, and deep network insight to identify biomarkers driving combination synergy. We demonstrate that the cancer cell ‘avatars’ capture the biological complexity of their in vitro counterparts, enabling the identification of pathway-level mechanisms of combination benefit to guide clinical translatability.

Abstract Accurately predicting drug sensitivity and understanding what is driving it are major challenges in drug discovery. Graphs are a natural framework for capturing diverse pharmacological data for efficacy predictions, thanks to their ability to integrate multimodal data and represent relationships such as gene-gene or drug-target interactions as edges. They have also had proven success across a range of other drug discovery tasks including repositioning and target identification. In this study, we sought to address the explainability challenges of drug response predictions. Recent developments in the field of Graph AI have led to improvements in interpretability mechanisms that highlight parts of a graph which are driving predictions. We have conducted a comprehensive review of multiple major approaches for tackling drug efficacy prediction using graph methods, benchmarking the performance and interpretability of these algorithms across indications. Methods: We assembled a combined dataset of GDSC1 and GDSC2 drug response data in cell lines, with multiomic cell line data and drug target and chemical structure data. We then applied graph-based approaches for the prediction of binarized IC50 on an indication-by-indication basis. Approach 1 involved the creation of a ‘GDSC knowledge graph’, where drug response and cell line ‘omic information is represented in an unweighted knowledge graph: cell lines are connected to genes expressed in them, drugs are connected to genes they target, and so on. We then used state-of-the-art graph embedding techniques to predict IC50 using paired drug and cell line embeddings. In Approach 2 we used a weighted knowledge graph instead, and generated embeddings using heterogeneous graph neural networks (HGNNs). In Approach 3, we modelled response prediction as a graph classification task, where one single graph captures one drug-cell line interaction. The graph classifier and HGNN models both have in-built interpretability mechanisms, including graph attention, that can signify the genes in the cell line which were most important for the eventual prediction. We can also integrate biomedical prior knowledge with all these models by capturing gene-pathway and gene-gene data in the graphs. Results: Our models outperformed benchmark models including DNNs and GBMs, and identified both established and novel response biomarkers in NSCLC cell lines (AUC = 0.94, Accuracy = 89%). We have also applied our models to Breast Cancer, Pancreatic Cancer, Colorectal Cancer and Haematological malignancies with similar predictive performance and explainability. Conclusions:Our graph analytical framework for response predictions showed better performance than benchmarking models and provided insights from explainability. This framework is easily extendable to response and ‘omic data from any disease model and patient studies. Citation Format: Jake Cohen-Setton, Krishna Bulusu, Jonathan Dry, Ben Sidders. Explainable AI: Graph machine learning for response prediction and biomarker discovery [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 902.

Abstract Background: Why some patients fail or have short lived response to immune checkpoint blockade (ICB) immunotherapy remains largely unknown. While baseline molecular assessments have provided clues to prognostic factors, insights into resistance drivers remains elusive. This is partially due to the difficulty in getting access to post progression samples from patients that were either primary resistant or developed acquired resistance after an initial response to ICB. Thus, the tumor-intrinsic and -extrinsic features that are selected for during progression and potentially drive primary and acquired resistance to immunotherapy remain underexplored. Methods: To compare clinical features and immunogenomic drivers of acquired and primary resistance to ICB across major cancers, we analysed and annotated de-identified patient records in the Tempus real-world database. We built an immuno-oncology cohort consisting of >2500 multimodal (DNA, RNA and clinical outcome data) pre-treatment baseline with >1500 post-treatment tumour biopsy samples from mainly NSCLC, TNBC, HNC and Bladder cancer patients. We used bulk RNA-seq data to estimate activation of the hallmark oncogenic pathways and immune cell composition and used panel DNA-seq data (>500 genes) to quantify mutation selection at the gene and pathway levels using dndscv. Results: Compared to acquired, primary resistant patients tended to have a higher observation of liver lesions at progression. Post-ICB, acquired resistant NSCLC and HNC patients showed a significantly inflamed tumour microenvironment (TME) characterised by higher estimation of infiltration of T cells and myeloid cells and higher activation of interferon gamma (IFNg) signalling as compared to primary resistant patients. In addition, in post-ICB acquired resistance in NSCLC we observed selection for mutations in genes involved in known immunomodulatory pathways, including loss-of-function mutations in B2M. Consistently, acquired resistance patients showed stronger selection for mutations in Hedgehog and Notch pathways as compared to primary resistance patients across NSCLC, HNC and TNBC post-ICB. Conclusions: Acquired and primary ICB resistant patients have distinct clinical and molecular features at progression. Their tumours’ TME is fundamentally different with acquired resistance TMEs being infiltrated with immune cells albeit escaped post progression. In addition, ICB selects mutations that potentially activate pathways such as Notch and Hedgehog. This multi-modal Real-World Data with post therapy biopsies has given insights for patient selection strategies and provides rational into combination treatment options for acquired resistant patients. Citation Format: Mohamed Reda Keddar, Sebastian Carrasco Pro, Kathleen Burke, Ana Camelo Stewart, Mark Cobbold, Ross Stewart, Sajan Khosla, Ben Sidders, Scott Hammond, Douglas C. Palmer, Jonathan Dry, Martin L. Miller. Multimodal real world data reveals immunogenomic drivers of acquired and primary resistance to immune checkpoint blockade [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 5100.

Abstract Introduction: Patient-derived organoids (PDOs) enable the ex vivo study of tumor heterogeneity and its effect on response to treatment. Our previous work demonstrated that our robust pan-cancer organoid platform shows genetic and transcriptomic concordance with clinical tumor samples and can be deployed for high-throughput drug response screening. Here, we ran a standard of care (SOC) panel on PDOs with paired de-identified clinically curated data to enable a comparison of patient response to the same drugs ex vivo and exploration of potentially more effective treatment alternatives. Methods: Patient tumor samples were cultivated into tumor PDOs in extracellular matrix + chemically defined media. PDOs underwent pathological review to ensure development of malignant tissue. PDOs were identified by clusters of Hoechst-positive cells and live and dead cell counts for each PDO were determined using fluorescent stains. The median number of cells per PDO in the well was calculated, excluding PDOs with <3 cells and the top 1% of PDOs by area. Drug screens were run with 6 doses of the compound and the inverse area under the curve of ToPro3 live cell measurements was calculated to quantify response. Tempus xT DNA-seq and xR whole-transcriptome assays were used to perform NGS on organoids and patient samples where available. Data was processed through our standard pipeline to identify targetable mutations, neoantigens, copy number variants (CNVs) and fusions. Results: We analyzed 38 PDOs with paired patient data across 9 different cancer types comprising mostly lung (n=14) and colorectal (n=12); 17 PDOs had responses associated with treatment given close to the biopsy collection date. Across the full PDO cohort, we observed a range of responses to SOC compounds, providing a platform to better understand biomarkers and mechanisms of response. Treatment response in PDOs correlated with patient response in many cases. In the 3 patients with progression-free survival >1 year and outcomes of either complete response or stable disease, all showed strong responses to the corresponding drugs in the SOC panel. For most patients with progressive disease recorded after treatment the corresponding PDO cluster showed limited response to treatment. The SOC panel includes several targeted therapies that provide insight into how patients respond to treatment beyond chemotherapy. For two patients with limited response to treatment in both patient and PDO, we identified targeted therapies relevant to their clinico-genomic landscape. One patient showed improved clinical outcomes to a later line of therapy with a similar mechanism of action to the drug that showed response in the organoid. Conclusion: These results suggest that PDOs may serve as a powerful tool for predicting patient response to treatment and aid the development of new therapies. Citation Format: Kathleen A. Burke, Brian Larsen, Yi-Hung Carol Tan, Jessica Barbeau, Curtis Brinkman, Elle Moore, Nick Callamaras, Jonathan R. Dry, Iker Huerga, Kate Sasser, Jeffrey A. Borgia. Integration of patient-derived tumor organoids and patient clinical multimodal data to investigate the role of organoids in predicting treatment response [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 225.

A video describing PDX mice, minimal information, and why minimal information is needed in the PDX research field.

PDF file - 37K, Characteristics of the 4 CRC datasets used in training (KFSYSCC) and validation of the RAS model.

PDF file - 35K, Data sets used for evaluation of the colorectal RAS model, and reported AUC in classification of KRAS mutant from wild-type samples.

supplementary figures and tables legend

Figure S1. A schematic representation for the method of transforming the log2 linear RPPA values to a 5 point categorical scale suitable for BMA modeling. Figure S2. Differential AML signaling responses to single agent PIM inhibition with AZD1208 and combination with FLT3 inhibitor AC220 Figure S3. Synergic combinations of drugs are sought to increase sensitivity. Figure S4. Susceptibility identified in AZD1208 Resistant Cells. Figure S5. Generation of a Boolean model to predict phosphorylation events, responses to single treatment and synergistic combinations of treatments. Figure S6. Generation of an AND/OR model to predict phosphorylation events, responses to single treatment and synergistic combinations of treatments. Table S1. Categorized RPPA. RPPA results for MOLM16, MV411, KG1A and EOL1 cells, categorized to 5 point scale from 0-4 Table S2. PhosphoScan MassSpec. LC-MS/MS phosphorylation proteomic in MOLM16 cells treated with 2 uM AZD1208 for 3 hours Table S3. QMN RPPA. Quadrant median normalized (QMN) calculated for RPPA protein levels for MOLM16, MV411, KG1A and EOL1 cells. Table S4. RPPA Stats. Calling statistically significant total protein and phosphorylation changes (determined by log2 QMN differences greater than or equal to 0.5 and Wilcoxon Rank Sum Tests p-values less than or equal to 0.1) Table S5. Exome. Whole exome DNA sequencing

Supplementary Figures 8-12 from Transcriptional Pathway Signatures Predict MEK Addiction and Response to Selumetinib (AZD6244)

Supplementary Table and Figure Legends from Transcriptional Pathway Signatures Predict MEK Addiction and Response to Selumetinib (AZD6244)

PDF file - 35K, Characteristics of the 2 metastatic CRC datasets used for evaluation of prediction of PFS.

Supplementary Figure 6 from Transcriptional Pathway Signatures Predict MEK Addiction and Response to Selumetinib (AZD6244)

<p>Supplementary Table S5. RPPA analysis of phosphorylated and total protein levels in PC9 and NCI-H1975 AZD9291 resistant populations compared to respective parental cells.</p>

55 FGFR2 modified dynamic biomarker after treatment by AZD4547