Abstract Liquid biopsy enables noninvasive assessment of cancer severity and prognosis, informing clinical decisions across the cancer care spectrum. Estimates of the fraction of tumor-derived DNA shed into circulation (tumor content; TC) reflect disease severity, with higher TC more frequently observed in advanced stages and linked to poorer outcomes. However, TC is a product of a complex pathology and is impacted by many factors including tumor type, size, shedding rate, aggressiveness, vascularization, genotype, and metastatic state. We aim to predict two of them, size and metastatic state, using a targeted cell-free DNA (cfDNA) methylation assay. To assess metastatic potential, we trained models for two binary prediction tasks in lung cancer: distant metastasis (Stage I/II vs IV) and late stage (Stage I/II vs III/IV). Features quantifying region-level methylation patterns were residualized with respect to sample-level TC to capture signals orthogonal to size. TC was evaluated as a single predictor in parallel. We trained on a dataset of 54 true and 324 synthetic lung cancer cfDNA samples to further emphasize size-independent, stage-related features. All models were assessed at 90% target specificity on an independent test set of 30 true cancer samples classified as lung cancer by a multiclass tissue-of-origin model to represent the end-to-end performance of the assay. For size prediction, we developed a lung-specific TC estimator, fit log-linear models linking TC to radiology-derived tumor size metrics, and generated predictions for 57 held-out test samples. All samples were from the CORE-HH clinical study (NCT05435066). The residual methylation models identified distant metastasis with 87.5% sensitivity (14/16 Stage IV) at 83% specificity (5/6 Stage I/II) and late stage with 67% sensitivity (16/24 Stage III/IV) at 83% specificity (5/6 Stage I/II). These results substantially improved on the TC-only model (19% and 17% sensitivity, respectively, at the same specificities), indicating methylation metrics capture tumor progression-associated signal orthogonal to TC. For size prediction, restricting to Stage I-III cases with PET metrics (N = 19) yielded the strongest fits (log10(total volume) R2 > 0.5, p < 5x10-4). Test set predictions showed moderate explanatory power (R2 = 0.24, p = 1.3x10-4), consistent with factors beyond size (e.g., visceral metastasis) influencing shedding. These findings highlight the potential for stage and size prediction models to deliver clinically actionable insights from a single blood draw. Late-stage prediction could guide workup prioritization, treatment intensity, and surveillance strategies, while size prediction may support prognosis and therapy selection. These capabilities motivate further model development for complementary prediction tasks (e.g., aggressiveness, genotype) toward a suite of tools for precision oncology. Citation Format: Kade P. Pettie, Shiva Farashahi, Jackson Killian, Andrew Wong, Yifan Wu, Dorna Kashef, Franziska Michor, Jocelyn Charlton, Kieran Chacko. Liquid biopsy cfDNA methylation predicts lung tumor size and metastatic potential in a single assay [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 2718.

10543 Background: Radiographic findings concerning for malignancy frequently prompt extensive diagnostic evaluations that fail to yield a definitive diagnosis, resulting in delayed treatment, persistent uncertainty, and unnecessary interventions. Multi-cancer early detection (MCED) tests based on circulating cell-free DNA (cfDNA) methylation have demonstrated high specificity in asymptomatic screening populations; however, performance in individuals with clinical suspicion of malignancy remains unclear. We compared the performance of a cfDNA methylation–based MCED test in asymptomatic participants and in participants with imaging-based suspicion of malignancy. Methods: Adults aged 45–79 years enrolled in CORE-HH (NCT05435066) were evaluated in (1) a case–control cohort (n = 4,137) comprising asymptomatic participants who reported no history of cancer (n = 2,523) and newly diagnosed cancer participants (n = 1,614); and (2) a prospective cohort (n = 209) of individuals with imaging-based suspicion of malignancy undergoing clinical evaluation (cancer confirmed, n = 155; no cancer, n = 54). Cancer types included lung, breast, colorectal, prostate, pancreatic/hepatobiliary, head and neck, gynecologic, and hematologic malignancies. Samples were analyzed using a cfDNA methylation–based MCED test incorporating an AI/ML model, at a pre-specified specificity of 98.5%. Test performance was compared between cohorts among non-cancer participants and among early-stage (stage I–II) and late-stage (stage III–IV) cancer cases using Fisher’s exact test (two-sided). Results: Among non-cancer participants (n = 2,577), empirical specificity was 98.1% (53/54) in the clinical suspicion cohort and 98.5% (2,485/2,523) in the asymptomatic case–control cohort (OR, 1.23; 95% CI, 0.03–7.6; p = 0.56). Overall sensitivity was 54.8% (85/155) in the clinical suspicion cohort and 53.9% (870/1,614) in newly diagnosed cancer cases from the case–control cohort (OR, 1.04; 95% CI, 0.74–1.47; p = 0.87). Stage I–II sensitivity was 27.7% (18/65) vs 28.7% (204/711) (p = 1.00), and stage III–IV sensitivity was 82.9% (58/70) vs 81.5% (573/703) (p = 0.87), for the clinical suspicion and case–control cohorts, respectively. Overall cancer stage distribution was similar between cohorts (p = 0.89). Conclusions: At the prespecified 98.5% specificity threshold, the cfDNA methylation–based MCED test demonstrated comparable specificity and sensitivity in prospectively enrolled participants undergoing evaluation for imaging-based suspicion of malignancy and in a case–control cohort of newly diagnosed cancer cases versus asymptomatic controls. Recognizing the tendency of case–control designs to overestimate test performance, these results describe MCED test performance in a cohort with clinical suspicion of malignancy, extending characterization beyond asymptomatic screening populations. Clinical trial information: NCT05435066 .

Aneuploidy is a hallmark of human tumors. While patient-level copy number alteration (CNA) differences have been investigated extensively in large cohorts, their intratumoral heterogeneity remains understudied. Here, we conducted a pan-cancer analysis of 94 human tumors at single cell resolution, representing seven cancer types: bladder, breast, colon, glioblastoma, kidney, lung, and ovarian. Single-cell copy number profiling was used to analyze 62,646 aneuploid cells, in addition to bulk exome sequencing of most patients and single-nucleus RNA-seq of 6 samples. In many cancer types, increased subclonal diversity was associated with higher CNA burden, whole genome doubling, TP53 mutations, and increased geographic diversity. Cancer cells from each patient shared a set of truncal CNAs, suggesting evolution from a single ancestral cell. Many tumors accumulated CNAs in bursts of evolution, suggesting that punctuated evolution is common in diverse cancer types. This study greatly improves our knowledge of intratumoral chromosome diversity across human cancers.

Abstract Diffuse large B cell lymphoma (DLBCL) is a heterogeneous disease comprised of at least 5 molecular subtypes (Clusters 1-5, C1-5). C1 DLBCLs share genetic features with transformed marginal zone lymphomas such as BCL6 translocations and genetic bases of immune evasion including inactivating mutations of CD70, the CD27 costimulatory ligand. We generated a murine model of Bcl6-driven lymphomagenesis in the setting of Cd70 deficiency and perturbed CD70/CD27 costimulation. We observed earlier onset and increased penetrance of the disease in Cd70 -/-;Bcl6 tg/+ compared to Bcl6 tg/+ animals. Almost all Cd70 -/-;Bcl6 tg/+ and Bcl6 tg/+ mice that were euthanized for symptoms (sick) exhibited splenomegaly. Histomorphologic analyses of these spleens revealed architectural disruption, diffuse infiltration of small-to-large highly proliferative monoclonal B cells intermixed with T cells, consistent with a diagnosis of DLBCL. To investigate the process of Bcl6-driven lymphomagenesis in the presence or absence of CD70, we harvested spleens from asymptomatic wild-type (WT), Cd70 -/-, Bcl6 tg/+ and Cd70 -/-;Bcl6 tg/+ animals at 6, 14 and 18 months and performed scRNA-seq of splenic cell suspensions. By serially characterizing asymptomatic and sick Bcl6 tg/+ and Cd70 -/-;Bcl6 tg/+ mice, we identified cell subsets with low-to-absent expression of the marginal zone (Cd21) and follicular (Cd23) B-cell markers. These Cd21 lo/- /Cd23 lo/- cells exhibited transcriptional features of extrafollicular innate-like B cells, including Bhlhe41 and Fcrl5 expression and regulatory features, including Ebi3 and Il10 expression. The earliest detected Cd21 lo/- /Cd23 lo/- aberrant B cells were progressively replaced by clonally related aged/autoimmune and MHC-IIlo cycling B cells. Clonal relationships between lymphoma precursors and malignant DLBCLs were determined by tracking shared BCR clonotypes and analyzing the acquisition of additional genomic alterations (with InferCNV). Transcriptional analysis of 409 human DLBCLs with C1-5 designations confirmed that human C1 tumors selectively expressed the innate-like B-cell markers, BHLHE41 and EBI3. We further characterized the T-cell anti-tumor immune responses in Bcl6 tg/+ and Cd70 -/-;Bcl6 tg/+ mice, which were predominantly mediated by CD4+, rather than CD8+, cytotoxic T cells (CTLs). CD4+ CTL clonal expansion was seen earlier (14 months) in Bcl6 tg/+ mice and was significant at 18 months in both Bcl6 tg/+ and Cd70 -/-; Bcl6 tg/+ cohorts, compared to WT controls. In vitro analyses confirmed that CD4+ CTLs from 14-month-old Bcl6 tg/+ animals killed Bcl6-driven DLBCLs in a MHC-II-dependent manner. However, in both tumor-bearing cohorts, sick animals had decreased CD4+ CTL and increased Treg clonal expansion. Together, these results highlight CD70/CD27 axis disruption as a central mechanism of accelerating lymphoma development, implicate innate-like B cells as the cells-of-origin in C1 DLBCLs and point to MHC-II downregulation and Treg clonal expansion as mechanisms of immune evasion from CD4+ CTL-mediated immunity. Citation Format: Elisa Mandato, Eleonora Calabretta, Gali Bai, Li Song, Tianfang Ma, Filip Garbicz, Marianna Palazzo, Julia Paczkowska, Il-Kyu Choi, Donna Neuberg, Scott Rodig, Ruben Carrasco, Baochun Zhang, Franziska Michor, Margaret A. Shipp. Bcl6-driven Cd70-deficient Diffuse large B-cell lymphomas originate from innate-like cells with blunted CD4+ cytotoxic T-cell immune surveillance [abstract]. In: Proceedings of the Fifth AACR International Meeting on Advances in Malignant Lymphoma: From Discovery to Clinical Impact; 2026 Jun 24-27; Philadelphia, PA. Philadelphia (PA): AACR; Blood Cancer Discov 2026;7(3_Suppl):Abstract nr PR002.

Abstract Targeted therapies are designed to eliminate cancer cells by directly inhibiting oncogenic driver proteins. In addition to their primary inhibitory effects on oncogenic signaling, these agents frequently impose collateral cellular stresses, such as DNA damage. KRAS-targeted therapies, particularly KRAS G12C inhibitors (G12Ci), represent a major therapeutic advance but remain limited in efficacy. Previous reports of targeted therapy-induced DNA damage, including studies of TKIs and MAPK inhibitors, have primarily been based on cytotoxic dosing conditions. Far less is known about whether DNA damage can also be induced by targeted therapies in less sensitive cancer models, particularly under sublethal doses that better mimic clinical responses. Failure to repair DNA damage can lead to chromosomal instability (CIN) and chromosomal aberrations. CIN is widely recognized to promote tumor evolution by enhancing cellular plasticity and adaptability, thereby contributing to therapeutic resistance and metastatic progression. However, it remains unknown how KRAS G12C inhibition influences CIN and whether G12Ci-induced CIN might generate unique, exploitable vulnerabilities.In this study, we profiled 15 KRAS G12C-mutant NSCLC cell lines representing diverse mutational backgrounds. We treated these models with the KRAS G12Ci LY3499446 and comprehensively assessed their DNA damage responses, CIN phenotypes, and sensitivity screening to combination therapies with agents that perturb chromosomal stability. We observed heterogeneous induction of DNA damage and CIN across these cell lines. Notably, we identified the strongest correlation between G12Ci-induced CIN and synergistic interaction with the Aurora kinase A inhibitor (AURKAi) LSN3321213. Machine learning-based single-cell image tracking and DNA barcoding analyses revealed that AURKA inhibition alone causes mitotic arrest followed by mitotic slippage, allowing cells to evade death, whereas combined G12Ci and AURKAi treatment triggers catastrophic mitotic cell death. Mechanistically, we found that G12Ci stabilizes Cyclin B1 through mitotic activation of ATR/ATM DNA repair signaling, thereby prolonging mitotic arrest. Under conditions of combined inhibition of KRAS G12C and AURKA, in which Cyclin B1 degradation is impaired, cells fail to exit mitosis and undergo catastrophic cell death. Together, our findings identify CIN as a predictive marker of response to combined KRAS G12C and AURKA inhibition, providing mechanistic rationale to enhance the therapeutic window of AURKA inhibitors when used with targeted therapies. Citation Format: Chendi Li, Varuna Nangia, Melissa Vieira, Anahita Nimbalkar, Christopher Graser, Jeremy Chang, Mohammad Syed, Yi Shen, Radhika Koranne, Lee Zou, Franziska Michor, Sabrina L. Spencer, Aaron N. Hata. Targeted therapy-induced chromosomal instability dictates mitotic dependency on Aurora kinase A [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 6773.

The Radiation Oncology-Biology Integration Network (ROBIN) initiative addresses critical gaps in radiation oncology by integrating advanced biological research, technologic innovation, and clinical practice. ROBIN leverages "omics" technologies, data science, and integrative analyses to elucidate the mechanisms governing tumor and normal tissue responses to radiotherapy (RT). Through five specialized centers-OligoMET, ImmunoRad, GenRad, METEOR, and KIDSROBIN-the network covers a broad spectrum of cancer and radiation biology research. Each center conducts translational programs linked to clinical trials, targeting key domains, including metastasis biology, RT-immune system interactions, and genomic determinants of treatment response. KIDSROBIN assures the invaluable inclusion of pediatric cancers to the consortium. By collecting clinically annotated human biospecimens and applying single-cell and spatially resolved omics, ROBIN enables mechanistic insights into radiation effects directly in patients. A central pillar of the initiative is its commitment to data standardization and sharing, using cloud-based platforms to generate accessible and interoperable datasets. ROBIN also prioritizes education and cross-disciplinary training to cultivate the next generation of scientists in radiation biology and oncology. This integrated approach positions ROBIN to drive transformative advances in radiation oncology and multimodal cancer therapy, informing personalized treatment strategies and improving patient outcomes. This review provides an overview of the ROBIN program and its key strategies, research activities, and contributions to advancing radiation biology and oncology. The vision and leadership of Dr. Norman Coleman have been foundational to the development of the ROBIN initiative, inspiring a collaborative ecosystem that bridges science and clinical practice to drive meaningful impact in patient care.

Each excel spreadsheet lists markers tested in each of the three cell line models.

Diffuse midline gliomas (DMGs) are driven by the H3K27M oncohistone-a challenging therapeutic target. However, conventional therapeutic modalities are never curative. Against this backdrop, we address an important unresolved question--are there H3K27M-induced oncogenic vulnerabilities that can be exploited for therapeutic benefit. We show that H3K27M induces hypertranscription, thus identifying hypertranscription as a new molecular feature of H3K27M-driven DMGs. We demonstrate this finding in genetic mouse models, human DMG cells, and primary tumor specimens. We further demonstrate that H3K27M-induced hypertranscription perturbs replication, heightens basal replication stress, and enhances sensitivity to ATR inhibition. In exploring therapeutic implications of these findings, we document brain penetrance, target engagement, and therapeutic efficacy of a clinical-stage ATR inhibitor (alnodesertib) in vitro and in intracranial DMG xenografts. We further demonstrate synergistic activity of alnodesertib with radiotherapy-the current standard of care for DMGs. These findings provide the mechanistic underpinning and preclinical rationale for including alnodesertib as monotherapy and in combination with radiation in clinical trials for children with H3K27M DMGs. The broad implications of our studies highlight ATR inhibition as a therapy for aggressive human cancers displaying hypertranscription.

Lists of differentially expressed genes between HER2hi and HER2lo cells in each model.

Intratumor heterogeneity for human epidermal growth factor receptor 2 (HER2) in HER2-positive breast cancer is a driver of resistance to HER2-targeted therapies. The advancement of treatments for HER2 heterogeneous (HET) tumors has been hindered by the lack of preclinical models that accurately mimic the human disease. In this study, we describe human HER2 HET breast cancer models composed of ERBB2-amplified (HER2hi) and nonamplified (HER2lo) cell populations derived from the same tumor. Utilizing these models, together with cellular barcoding, we demonstrate subclonal cooperation between HER2hi and HER2lo subpopulations. Furthermore, HER2lo cells drive resistance to HER2-targeting antibody-drug conjugates (ADC) like trastuzumab deruxtecan (T-DXd) but are sensitive to HER2 kinase inhibitors. CRISPR screens in HET cocultures identified sensitizers of HER2lo cells to T-DXd, including ATP-binding cassette subfamily C member 1 and ubiquitin-specific peptidase 9 X (USP9X). USP9X inhibition enhances the lysosomal targeting of HER2, thereby potentiating ADC payload release and reducing tumor recurrence after T-DXd treatment. Our results elucidate the functional relevance of HER2 heterogeneity and propose improved therapies for these tumors. SIGNIFICANCE:Studies of HER2 HET breast cancer models demonstrated that HER2lo cells drive HER2-targeting ADC resistance and accelerate recurrence by cooperating with HER2hi cells. We identified novel therapeutic strategies to sensitize HER2lo cells to T-DXd, providing mechanistic insight and offering promising avenues to overcome resistance and improve patient outcomes.

Simulated values and estimates of single-type BESTD.

In estrogen receptor-positive (ER +) breast cancer, CDK4/6 inhibitors (CDK4/6is) combined with endocrine therapy (ET) are standard first-line treatment for metastatic disease. However, most patients eventually develop resistance. Activating ESR1 mutations are a prevalent mechanism of acquired resistance to ET and are enriched after ET plus a CDK4/6 inhibitor (CDK4/6i), but their role in the clonal evolution and adaptive mechanisms of acquired resistance to CDK4/6 inhibition, independent of ET, is unknown. In addition, whether different CDK4/6is impose distinct selective pressures and divergent resistance states remains elusive. To investigate the clonal dynamics, cell states and cellular plasticity during acquired CDK4/6i resistance in mutant versus wild-type (WT) ESR1, we performed high-complexity DNA barcoding (ClonTracer library) with longitudinal sampling and multi-omic profiling in an isogeneic MCF7 model expressing WT ER or Y537S mutant ER. We also evaluated the clonality of the ESR1 mutations in clinical samples with CDK4/6i resistance. We showed that ESR1 mutations are enriched in clinical tumors with acquired resistance to CDK4/6is, and in paired biopsies expanded to near clonality after treatment. We demonstrated progressive clonal selection with both divergent and partially convergent evolutionary trajectories. The ESR1 mutation substantially reshapes clonal and epigenetic evolution during palbociclib resistance but had a weaker impact under abemaciclib selection. Overall, clonal evolution and cell states in palbociclib and abemaciclib resistance were distinct. Single-cell RNA-seq revealed transcriptional heterogeneity highlighting cellular plasticity during passaging of cells and selection. Finally, in vivo barcoding of mammary xenograft, local recurrences, and distant metastases demonstrated site-specific clonal outgrowth in mutant ER metastases, and partial overlap between metastatic and CDK4/6i-resistant subclones, supporting the dual role of specific populations in therapeutic resistance and metastatic colonization. High-resolution lineage tracing and multi-omic studies demonstrate that CDK4/6i resistance is shaped by clonal selection and adaptive remodeling of cell states, with the ESR1 mutation status and the specific inhibitor acting as key determinants of evolutionary trajectories. These findings suggest that both variables should be considered when designing sequential and combination treatment strategies to overcome CDK4/6i resistance.

Abstract Introduction: Blood-based liquid biopsies offer potential for non-invasive cancer screening. However, detecting early-stage disease is complicated by low levels of circulating tumor biomarkers and background noise from normal cells. To address this, we developed a novel deep-learning framework to detect cancer signal at the resolution of single DNA reads. Applied to bisulfite-converted cell-free DNA (cfDNA) samples, our method significantly improves early-stage cancer sensitivity. Methods: We designed a massively parallel 2-D convolutional neural network architecture that differentiates cancer and non-cancer signal in cfDNA by learning local methylation patterns at thousands of genomic regions. The model takes next generation sequencing (NGS) data as input, encodes aligned sequences within genomic windows as images, and outputs informative feature vectors for classification. However, training is complicated by two real-world data limitations: (1) disease samples contain a mix of unlabeled fragments from normal and diseased cells, and (2) acquiring sufficient early-stage disease data is costly, burdensome, and time-intensive. To address these, we designed a novel data generation technique that (1) assigns positive labels for groups of reads via in silico spike-in of tumor biopsy reads and (2) generates large, diverse datasets via fine-tuned in silico mixing of non-cancer cfDNA reads. Our model’s architecture has key advantages: compact input encoding, interpretable saliency maps, and scalable parallel architecture: we trained on 720TB of data across 10 million genomic bases in a single day, highlighting our framework’s efficiency. Results: We validated our method with targeted bisulfite sequencing data from the CORE-HH clinical study (NCT05435066, N=1229 non-cancers, N=1118 cancers, including N=599 Stage I/II). We pretrained the model on 1.3 billion training examples generated using a held-out set of non-cancer plasma (N=174) and tumor tissue biopsies (N=505). Predictions on data from clinical samples yielded feature vectors, with saliency maps confirming the model highlights biopsy-learned patterns. In a 10x5 cross-validation, classifiers trained on these feature vectors improved overall sensitivity by 9.9 points (Stage I: +6.5 pts, II: +17.6 pts, III: +14.3 pts, IV: +9.5 pts) at 98.5% specificity, compared to classifiers trained on region-wide average methylation values. These performance improvements, coupled with the scalability of the framework, underscore its potential as a transformative tool in the early diagnosis of cancer and establish a foundation for training models on NGS data in other liquid biopsy assays. Future work will investigate the potential to incorporate per-read embeddings from DNA-based large language models, without sacrificing scalability. Citation Format: Jackson A. Killian, Kade Pettie, Kyle Gowen, Shiva Farashahi, Esther Brown, Feras Hantash, Jocelyn Charlton, Franziska Michor, Kieran I. Chacko, Dorna Kashef. Improving early cancer detection by training scalable deep neural networks to extract tumor signal from cell-free DNA [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 5465.

Abstract Triple-negative breast cancer (TNBC) is the most aggressive form of breast cancer and is characterized by a very high recurrence rate. The primary treatment for this cancer type is systemic chemotherapy, often combined with immunotherapy. However, these treatments typically result in only short-term responses. Recent experimental studies have highlighted the critical role of chromatin modifications, such as DNA methylation and histone modifications, in TNBC development and progression. Consistent with these findings, epigenetic therapies targeting these modifications have shown promise in preclinical and early clinical studies. Building on these insights, we developed a predictive computational modeling platform to gain a deeper understanding of how different chromatin modifications influence TNBC and to identify optimal therapeutic strategies targeting these modifications. The platform integrates experimental data, pharmacokinetics, and mechanistic insights into the role of chromatin regulation in TNBC. We focused on therapies combining an EZH2 inhibitor, which reduces the establishment of repressive H3K27me3 marks and promotes chromatin opening, and an AKT inhibitor, which enhances the expression of GATA3 and BMF. Using this framework, we confirmed the experimentally observed synergy between the two inhibitors and uncovered its mechanistic basis: EZH2 inhibition quickly reaches a plateau once chromatin becomes fully accessible, whereas the effect of AKT inhibition increases more gradually across a broader concentration range. In-silico clinical trials involving 1,500 virtual patients further showed that optimized combination schedules markedly outperform monotherapies. Based on these predictions, we conducted in-vitro experiments using live-cell imaging, which validated the model’s predictions on how different inhibitor doses and combinations shape treatment response, and confirmed the combination regimen identified by the platform as optimal. Beyond TNBC, the generalizable framework developed here can potentially be adapted to other types of cancer in which chromatin modifications play a similar role. This research can then also help establish the basis for the discovery of common therapeutic targets, contributing to a broader range of cancer treatment strategies. Overall, this research could lead to the development of novel, more effective, and more durable treatment strategies for TNBC, thereby improving patient outcomes and providing new possibilities for cancer therapy research. Citation Format: Simone Bruno, Alexandra Indeglia, Sophia Lichterfeld, Karen M. Cichowski, Franziska Michor. Optimal epigenetic therapies in triple-negative breast cancer [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 1838.

Lists of antibodies, target, application, company and catalogue number, and dilution.

Abstract Integrating diverse molecular modalities to obtain a comprehensive view of cellular identity remains a major challenge in single-cell biology. A fundamental but underappreciated obstacle is structural mismatch — the phenomenon in which the neighborhood structure of a cell differs depending on which molecular modality is used to define it. Existing approaches typically embed modalities into a shared latent space, which actively erases the structural differences between modalities that make multimodal measurements scientifically valuable. Here we introduce Move BeTween modAlities (MBTA), the first framework explicitly designed to address structural mismatch. Rather than forcing modalities into a shared representation, MBTA maintains modality-specific latent spaces and connects them via flow matching, preserving the structural integrity of each modality while enabling accurate cross-modal translation. Across extensive benchmarks on multi-modal single-cell datasets, MBTA consistently outperformed existing methods, with the largest gains observed in datasets with pronounced structural mismatch. Applied to joint genomic and transcriptomic profiles of breast cancer patients, MBTA identified transcriptomic lineage relationships corroborated by genomic variation and outperformed state-of-the-art transcriptomics-based copy number inference methods. Extending this framework to mouse embryonic development, we reconstructed temporal trajectories jointly defined by gene expression and seven complementary epigenetic modalities. MBTA can connect any number of molecular readouts without erasing their individual character, serving as the computational foundation for assembling multi-layered portraits of cells.

Lists of differentially expressed genes between HER2hi and HER2lo cells in the HCC1954 and 21PT models growing in monoculture or in co-culture.

Single-cell RNA sequencing (scRNA-seq) profiles cellular heterogeneity but captures only static snapshots, limiting inference of gene expression dynamics. We developed PROFET (particle-based reconstruction of generative force-matched expression trajectories), a framework that reconstructs continuous, nonlinear single-cell trajectories from sparsely sampled scRNA-seq time series. PROFET combines a particle-based gradient-flow algorithm with simulation-free force matching to accurately infer cellular dynamics. Across mouse and human in vitro datasets and an in vivo axolotl regeneration dataset, PROFET achieved 2.6-12.5× lower prediction error than ten state-of-the-art trajectory inference methods. Applying PROFET to newly generated scRNA-seq data from a palbociclib-treated MCF7 cell line and three published breast cancer patient datasets, we reconstructed treatment-response trajectories and identified a resistant cell subpopulation exhibiting large phenotypic shifts and enrichment of the surface markers UNC5B, TLR3, PCDH19, PROCR, SLITRK6, and SEMA6B. PROFET provides a biologically grounded framework for reconstructing cell-state dynamics from static single-cell data across development, regeneration, and therapeutic response. A record of this paper's transparent peer review process is included in the supplemental information.

Understanding drug responses at the cellular level is essential for elucidating mechanisms of action and advancing preclinical drug development. Traditional dose-response models rely on simplified metrics, limiting their ability to quantify parameters like cell division, death, and transition rates between cell states. To address these limitations, we developed Bayesian Estimation of STochastic processes for Dose-Response (BESTDR), a framework modeling cell growth and treatment response dynamics to estimate concentration-response relationships using longitudinal cell count data. BESTDR enables quantification of rates in multistate systems across multiple cell lines using hierarchical modeling to support high-throughput screening. Validation of BESTDR with synthetic and experimental datasets demonstrates its accuracy and robustness in estimating drug response. By integrating mechanistic modeling of cytotoxic, cytostatic, and other drug effects, BESTDR enhances dose-response studies, facilitating robust drug comparisons and mechanism-specific analyses. BESTDR offers a versatile tool for early-stage preclinical research, paving the way for drug discovery and informed experimental design. SIGNIFICANCE:BESTDR leverages time-course cell count data to provide mechanistic insights into drug actions, distinguishing cytostatic, cytotoxic, and state transitions, thus advancing dose-response modeling crucial for preclinical research and development of targeted therapies. This article is part of a special series: Driving Cancer Discoveries with Computational Research, Data Science, and Machine Learning/AI .