Research Radar — 2026-07-03
Methods & AI
Computational
BulkFormer: A large-scale foundation model for bulk transcriptomes
Cell Systems Published 2026-07-02 research article DOI: 10.1016/j.cels.2026.06.001
foundation model deep learning transcriptomics multi-omics AI
Summary: BulkFormer is a transformer-based foundation model pre-trained on over 100,000 bulk RNA-seq samples spanning diverse tissues, diseases, and species. It learns universal representations of transcriptional programs that capture regulatory mechanisms, disease states, and drug responses, significantly outperforming existing methods on gene function prediction, disease classification, biomarker discovery, and phenotype prediction.
Why it matters: Establishes that bulk RNA-seq pretraining can outperform single-cell pretraining for many downstream transcriptomic tasks, challenging the prevailing single-cell-centric paradigm for biological foundation models.
Why for Yiru: Directly applicable to my multi-omics integration and biomarker discovery work — BulkFormer's approach to modeling gene interactions could enhance analysis of bulk transcriptomic data from tumor microenvironments.
Spatial transcriptomic analyses highlight distinct erythroid niches in mice and humans
Nature Genetics Published 2026-06-30 research article DOI: 10.1038/s41588-026-02671-2
spatial transcriptomics single-cell cell atlas imaging
Summary: Using an integrated spatial transcriptomics approach combining MERFISH and Visium HD with scRNA-seq, this study identifies distinct erythroblastic islands in both murine and human bone marrow. These niches exhibit unique spatial organization around specific stromal cell types and macrophage subsets with discrete signaling environments that regulate erythroid differentiation and enucleation.
Why it matters: Provides the first high-resolution spatial map of erythroid niches across species, revealing conserved principles of hematopoietic tissue organization that may inform bone marrow disease modeling.
Why for Yiru: The spatial transcriptomics integration framework is relevant to my tumor microenvironment analysis, particularly for understanding how niche organization governs cell fate decisions in immune and cancer cells.
Reshaping biomolecular structure prediction through strategic conformational exploration with HelixFold-S1
Nature Machine Intelligence Published 2026-07-01 research article DOI: 10.1038/s42256-026-01264-2
deep learning AI foundation model protein structure
Summary: HelixFold-S1 integrates AlphaFold2-based structure prediction with a strategic conformational exploration module that samples diverse structural states through targeted perturbations of the inter-residue distance map. By incorporating Markov state models and evolutionary coupling analysis, it systematically expands the protein folding landscape, revealing cryptic pockets, alternative functional states, and conformational transitions.
Why it matters: Advances protein structure prediction beyond static snapshots to dynamic conformational ensembles, which is essential for understanding protein function and designing targeted therapeutics.
Why for Yiru: While protein structure prediction is not my primary focus, the conformational exploration methodology could be relevant for understanding immune receptor flexibility in CAR design and antigen recognition.
Connecting single-cell transcriptomes to projectomes in the mouse visual cortex
Nature Published 2026-07-02 research article DOI: 10.1038/s41586-026-10424-8
single-cell spatial transcriptomics cell atlas imaging
Summary: This study combines single-cell and spatial transcriptomics with high-resolution axonal projection mapping using the STARmap PLUS platform and anterograde viral tracing across the mouse visual cortex. The resulting integrated transcriptomic-projectomic atlas reveals precise relationships between gene expression profiles and long-range connectivity, identifying transcriptomic signatures that predict projection targets.
Why it matters: Establishes a foundational framework for linking transcriptomic cell types to their long-range connectivity patterns, bridging molecular and systems neuroscience at unprecedented resolution.
Why for Yiru: The integrative spatial transcriptomics methodology is relevant to my approaches for linking molecular profiles to functional tissue architecture in the tumor microenvironment.
Towards the construction of a virtual yeast
Nature Published 2026-07-03 research article DOI: 10.1038/s41586-026-10574-9
digital twin AI deep learning systems biology whole-cell modeling
Summary: This study presents a whole-cell computational model of Saccharomyces cerevisiae that integrates 15 distinct submodels of cellular processes, including metabolism, gene expression, signaling, cell cycle, and stress response. The virtual yeast represents a major milestone toward predictive biology and in silico cell engineering.
Why it matters: Establishes the first comprehensive whole-cell digital twin of a eukaryotic organism, providing a blueprint for predictive modeling of cellular behavior that could transform drug discovery and synthetic biology.
Why for Yiru: Directly relevant to my interest in digital twin AI for biology — this framework provides a template for building predictive models of immune cell behavior and informs my vision of patient-specific immune digital twins.
Whole-cell particle-based digital twin simulations from 4D lattice light-sheet microscopy data
Cell Published 2026-07-02 research article DOI: 10.1016/j.cell.2026.07.002
digital twin imaging AI cell biology
Summary: This work presents a computational framework for constructing whole-cell digital twins directly from 4D lattice light-sheet microscopy data. The method segments cellular structures and initializes particle-based reaction-diffusion simulations that recapitulate dynamic processes such as organelle transport, cell division, and drug-induced morphological transitions.
Why it matters: Bridges live-cell imaging and computational simulation to create data-driven whole-cell models, enabling in silico experimentation on complex cellular dynamics that were previously inaccessible.
Why for Yiru: The particle-based digital twin framework is highly relevant to my interest in building predictive cell models — combining microscopy-driven digital twins with transcriptomic data could enable multi-scale immune cell modeling.
Biomedical discoveries
Biomedicine
Dual tumour–myeloid targeting of glioblastoma with GPNMB CAR-T cells
Nature Published 2026-07-01 research article DOI: 10.1038/s41586-026-10641-1
CAR-T immunotherapy tumor microenvironment T cell cancer macrophage
Summary: CAR-T cells engineered to target GPNMB, a cell surface glycoprotein overexpressed on both GBM tumor cells and immunosuppressive myeloid cells in the tumor microenvironment, simultaneously eliminate tumor cells and reprogram the myeloid compartment. This dual targeting strategy converts suppressive macrophages to a pro-inflammatory state, demonstrating potent anti-tumor activity.
Why it matters: Pioneers a dual tumour–myeloid targeting strategy for CAR-T therapy in solid tumors, directly addressing the immunosuppressive tumor microenvironment that limits current adoptive cell therapies.
Why for Yiru: Directly relevant to my work on CAR-T cell therapy and tumor microenvironment biology — this dual-targeting paradigm could be extended to other solid tumors and engineered into next-generation CAR designs.
TROP2 targeting reveals therapy-driven cell state dynamics in colorectal cancer
Nature Published 2026-07-02 research article DOI: 10.1038/s41586-026-10705-2
cancer tumor microenvironment single-cell spatial transcriptomics cell state
Summary: Longitudinal single-cell and spatial profiling of patient-derived xenografts and clinical biopsies uncovers therapy-driven cell state transitions following TROP2-ADC treatment in colorectal cancer. These include a drug-tolerant persister state characterized by EMT and a reversible resistant state associated with dedifferentiation.
Why it matters: Reveals the dynamic nature of therapy-driven cell plasticity in CRC and identifies TROP2 as both a marker of resistant states and a therapeutic target, with direct implications for ADC-based combination strategies.
Why for Yiru: Cell state dynamics and plasticity in cancer are central to understanding immunotherapy resistance — the single-cell and spatial profiling approach here provides a template for my own studies of therapy-induced TME remodeling.
LEF1 and niche factors determine T cell stemness across chronic diseases
Cell Published 2026-07-02 research article DOI: 10.1016/j.cell.2026.07.001
T cell immune immunotherapy single-cell CRISPR screen
Summary: Integrative single-cell transcriptomic, epigenomic, and functional CRISPR screening across chronic infection and tumor models identifies LEF1 as a master regulator of T cell stemness programs. LEF1 deletion drives terminal exhaustion, while its enforced expression enhances the persistence and anti-tumor efficacy of adoptively transferred T cells.
Why it matters: Identifies a druggable master regulator of T cell stemness with direct implications for improving adoptive T cell therapies, including CAR-T and TIL therapy, by preventing terminal exhaustion.
Why for Yiru: Directly relevant to my interests in T cell biology and immunotherapy — LEF1 modulation could be a strategy to enhance persistence of engineered T cells in my CAR-T and T cell therapy research.
Lactate binds and inhibits the innate immune sensor STING to promote tumor immune evasion
Immunity Published 2026-07-02 research article DOI: 10.1016/j.immuni.2026.07.001
tumor microenvironment immunotherapy immune T cell cancer
Summary: Lactate is shown to directly bind to the innate immune sensor STING, inhibiting its oligomerization and downstream activation of IRF3 and NF-κB. This metabolite-mediated inhibition suppresses type I interferon production in dendritic cells, dampening anti-tumor T cell responses. Genetic ablation of lactate dehydrogenase restores STING-dependent antitumor immunity.
Why it matters: Reveals a direct mechanistic link between tumor aerobic glycolysis and STING inhibition via lactate, identifying a key metabolic node for overcoming innate immune evasion in the tumor microenvironment.
Why for Yiru: The lactate-STING axis is a fascinating intersection of metabolism and innate immunity — highly relevant to my interests in TME remodeling and identifying combinatorial immunotherapy strategies targeting metabolic immune evasion.
Steatosis shapes prognosis-defining liver metastasis heterogeneity in CRC
Nature Published 2026-07-01 research article DOI: 10.1038/s41586-026-10686-2
cancer tumor microenvironment spatial transcriptomics multi-omics metastasis
Summary: Using spatial transcriptomics, metabolomics, and functional assays, this study demonstrates that the steatotic liver microenvironment creates a unique metabolic niche that drives profound intra-tumoral heterogeneity in CRC liver metastases, including lipid-adapted persister cells and altered immune evasion programs.
Why it matters: Uncovers how host metabolic state (liver steatosis) directly shapes metastatic heterogeneity and immune evasion in colorectal cancer liver metastases, with implications for prognosis and therapeutic targeting.
Why for Yiru: The integration of spatial transcriptomics with metabolomics to characterize metastatic niche heterogeneity is directly relevant to my multi-omics approaches for studying the tumor microenvironment.
A genome-scale CRISPRi perturbation atlas of human induced pluripotent stem cells
Nature Biotechnology Published 2026-06-29 research article DOI: 10.1038/s41587-026-03199-w
CRISPR screen perturbation single-cell cell atlas stem cell
Summary: A genome-scale CRISPRi screen in human iPSCs systematically represses ~18,000 genes with a library of ~200,000 sgRNAs. The resulting perturbation atlas maps genes essential for pluripotency maintenance, cell cycle regulation, and stress responses, establishing a foundational resource for functional genomics in human stem cells.
Why it matters: Provides the first genome-scale perturbation atlas in human iPSCs, setting a new benchmark for functional genomics and establishing a reference for interpreting gene function in stem cell biology and disease modeling.
Why for Yiru: The CRISPRi screening approach and perturbation atlas methodology are directly relevant to my work on functional genomics screens in immune cells — this resource provides a framework for designing and interpreting perturbation experiments.
Cross-disciplinary watchlist
Other Fields
Multi-antigen-targeting T cells in pediatric central nervous system tumors: a phase 1 trial
Nature Medicine Published 2026-07-01 clinical trial DOI: 10.1038/s41591-026-04449-9
T cell immunotherapy cancer clinical trial
Summary: This phase 1 trial evaluates multi-antigen-targeted T cells recognizing HER2, IL13Rα2, and EPHA2 in children with recurrent CNS tumors. The therapy demonstrates acceptable safety with objective radiographic responses and prolonged disease stabilization in a heavily pretreated pediatric population.
Why it matters: Provides clinical proof-of-concept for multi-antigen-targeting T cell therapy in pediatric CNS tumors, addressing antigen heterogeneity as a key barrier to adoptive cell therapy in solid tumors.
Why for Yiru: Directly relevant to my interest in T cell therapy and immunotherapy resistance — the multi-targeting strategy to overcome antigen escape is applicable to designing more durable CAR-T therapies.
Casdatifan shows durable response linked to HIF-2α biology in kidney cancer
Nature Published 2026-07-02 research article DOI: 10.1038/s41586-026-10718-x
cancer biomarker clinical trial
Summary: Casdatifan, a next-generation HIF-2α inhibitor, demonstrates durable objective responses in advanced clear cell renal cell carcinoma. Responses are enriched in tumors with specific biomarkers of HIF-2α pathway activation, including high CA9 expression and loss of VHL.
Why it matters: Validates HIF-2α as a therapeutic target in kidney cancer and identifies predictive biomarkers for patient stratification, advancing precision oncology in RCC.
Why for Yiru: The biomarker-driven patient stratification approach and HIF-2α biology are relevant to my interests in identifying predictive biomarkers for cancer therapy response.
Clinical decision support in hematological malignancies using a case-grounded AI agent
Nature Medicine Published 2026-07-02 research article DOI: 10.1038/s41591-026-04494-4
AI deep learning cancer clinical decision support
Summary: This study develops an LLM-based AI agent that provides clinical decision support in hematological malignancies. Grounded in a curated database of real-world clinical cases, the agent interprets complex patient data to recommend diagnoses, prognostic stratifications, and treatment options.
Why it matters: Demonstrates the translational potential of LLM-based AI agents for clinical decision support in oncology, with a rigorous case-grounded approach that addresses the hallucination and reliability challenges of general-purpose LLMs.
Why for Yiru: Highly relevant to my interest in AI applications in medicine — the case-grounded approach for clinical decision support provides a template for building reliable AI tools for immunotherapy response prediction.
Innate immune responsiveness predicts enhanced cellular immunity after controlled human influenza infection
Nature Medicine Published 2026-07-01 research article DOI: 10.1038/s41591-026-04483-7
immune biomarker single-cell multi-omics
Summary: Using single-cell multi-omics (scRNA-seq, CITE-seq, Olink proteomics), this study finds that baseline innate immune signatures strongly predict the magnitude of influenza-specific T cell responses and symptomatic disease after controlled H1N1 challenge in humans.
Why it matters: Establishes that pre-existing innate immune state is a major determinant of adaptive T cell immunity, with implications for vaccine response prediction and personalized immunization strategies.
Why for Yiru: The multi-omics approach for predicting immune responses from baseline signatures is directly applicable to my work on predicting immunotherapy response using pre-treatment tumor and blood profiling.
An agentic artificially intelligent X-ray scientist
Nature Machine Intelligence Published 2026-07-01 research article DOI: 10.1038/s42256-026-01261-5
AI deep learning laboratory automation
Summary: An autonomous AI agent is designed to fully operate an X-ray diffractometer for materials discovery. The agent autonomously designs experiments, collects diffraction patterns, and decides on next experiments without human intervention, demonstrating a new paradigm for AI-driven laboratory automation.
Why it matters: Demonstrates the feasibility of fully autonomous AI-driven experimental science, setting a precedent that could be extended to biology and drug discovery laboratories.
Why for Yiru: The autonomous AI agent concept is interesting but tangential to my primary research focus; however, similar agentic approaches could eventually be applied to automated experimental biology and screening.
Lineage tracing from cellular heritage to disease destiny
Nature Genetics Published 2026-07-01 review DOI: 10.1038/s41588-026-02628-5
single-cell CRISPR screen spatial transcriptomics cancer cell atlas
Summary: This Review synthesizes how CRISPR-based barcoding, somatic mutation phylodynamics, and integrated spatial transcriptomics are elucidating the origins of cellular heterogeneity in development, aging, and cancer. It provides a comprehensive framework for understanding how lineage history shapes disease outcomes.
Why it matters: Offers a unified perspective on how emerging lineage tracing technologies are transforming our understanding of cellular heterogeneity and disease progression, serving as a valuable reference for the field.
Why for Yiru: The synthesis of lineage tracing with single-cell and spatial technologies is highly relevant to my research on tumor heterogeneity and clonal evolution — this review provides a roadmap for integrating lineage information into my cancer genomics analyses.