Research Radar — 2026-07-22
Methods & AI
Computational
Expanding the scope of protein language modeling to protein-protein interactions with MSA Pairformer
Cell Published 2026-07-21 AI_model DOI: 10.1016/j.cell.2026.06.029
deep learning protein language model protein-protein interactions AI computational biology
Summary: MSA Pairformer expands the scope of protein language modeling to protein-protein interactions, enabling prediction and design of protein complexes. By extending the Pairformer architecture to handle multiple sequence alignments from interacting protein families, the model captures co-evolutionary signals across interaction interfaces, achieving state-of-the-art performance on protein complex structure prediction and design tasks.
Why it matters: Bridges the gap between single-chain protein language models and the reality that most proteins function through interactions, enabling AI-driven design of protein complexes with therapeutic and biotechnological applications.
Why for Yiru: The extension of protein language models to pairwise interactions demonstrates a principled approach that could inspire similar advances in modeling ligand-receptor interactions in spatial transcriptomics and cell-cell communication networks in the tumor microenvironment.
Accurate, sensitive, and efficient chromatin accessibility quantification at target loci using UNIChro-seq
Nature Communications Published 2026-07-20 computational_method DOI: 10.1038/s41467-026-75767-2
chromatin accessibility epigenomics fine-mapping disease risk variants method development
Summary: UNIChro-seq digitally counts accessible chromatin molecules at target loci, enabling accurate and sensitive quantification of allelic effects. Applied to 57 autoimmunity risk alleles and 20 genome-edited variants, the method provides a targeted approach for dissecting the regulatory impact of non-coding disease-associated variants at base-pair resolution.
Why it matters: Provides a targeted, high-resolution approach for quantifying chromatin accessibility at specific regulatory loci, addressing a critical need in functional genomics for validating the regulatory impact of non-coding risk variants identified through GWAS.
Why for Yiru: The targeted chromatin accessibility quantification approach could complement my single-cell and spatial analyses by providing locus-specific regulatory information — useful for validating how non-coding risk variants influence gene regulation in immune cells within the tumor microenvironment.
Design and optimization of a kinase-controlled allosteric switch
Nature Methods Published 2026-07-21 synthetic_biology DOI: 10.1038/s41592-026-03163-1
synthetic biology allostery kinase phosphorylation protein engineering
Summary: Post-translational control enables rapid and precise regulation of cell behavior. Using an allosterically controllable Gal4 transcription factor scaffold, researchers developed a classic kinase FRET biosensor architecture as a phospho-switch, achieving a 20-fold phosphorylation-dependent transcriptional change. The design establishes a generalizable platform for engineering kinase-controlled gene expression.
Why it matters: Offers a modular synthetic biology platform for engineering post-translational control of gene expression, with potential applications in cell therapy, biosensing, and basic research where rapid, reversible control of cellular behavior is needed.
Why for Yiru: The engineered phospho-switch concept is relevant to my interests in synthetic biology approaches for immunotherapy — similar kinase-controlled switches could be designed to regulate CAR-T cell activity or control therapeutic gene expression in engineered immune cells.
Neural sampling from cognitive maps enables goal-directed imagination and planning
Nature Machine Intelligence Published 2026-07-21 AI_research DOI: 10.1038/s42256-026-01254-4
neural networks cognitive maps planning AI stochastic computing compositional coding
Summary: The authors integrate cognitive maps, stochastic computing, and compositional coding to build neural network models that enable goal-directed imagination and planning, addressing how brains achieve flexible problem-solving with minimal energy. The model demonstrates compositional generalization and efficient planning in grid-world environments.
Why it matters: Advances our understanding of how neural systems may accomplish flexible, energy-efficient planning through compositional coding and stochastic sampling — principles that could inform the next generation of more efficient AI architectures.
Why for Yiru: The compositional coding and stochastic sampling principles are broadly relevant to my AI interests — similar approaches could inspire more efficient methods for representation learning in spatial transcriptomics and biomedical foundation models.
PepCL: A replay-based continual learning framework for updating peptide-MHC models
bioRxiv (bioinformatics) Published 2026-07-20 AI_model DOI: 10.64898/2026.07.20.733558
deep learning continual learning peptide-MHC immunoinformatics AI
Summary: PepCL is a replay-based continual learning framework for updating peptide-MHC binding prediction models as new data arrives without catastrophic forgetting. By selectively replaying representative training examples from previous tasks, the model maintains performance on older peptide-MHC alleles while adapting to newly characterized ones, addressing a critical practical challenge in immunoinformatics.
Why it matters: Solves the practical problem of continuously updating peptide-MHC binding predictors as new allele-specific data becomes available, which is essential for keeping immunoinformatics tools relevant as our knowledge of the immunopeptidome expands.
Why for Yiru: Directly relevant to my deep learning and immunoinformatics work — the continual learning framework could be applied to spatial transcriptomics models that need to adapt to new tissue types or cancer indications without retraining from scratch, and the peptide-MHC focus is directly relevant to tumor neoantigen prediction.
Robustly enhancing crop genomic prediction accuracy through ensemble learning and iterative optimization
Nature Communications Published 2026-07-20 machine_learning_method DOI: 10.1038/s41467-026-75788-x
ensemble learning genomic prediction AI crop breeding computational biology
Summary: GEG2P is a genetic algorithm-based ensemble learning method for genotype-to-phenotype prediction integrating 20 base learners, improving accuracy by 4.02% on average across diverse crops. The approach uses iterative optimization to select and weight individual predictors, addressing instability and limited accuracy challenges in genomic prediction.
Why it matters: Demonstrates that systematic ensemble learning with iterative optimization can materially improve genomic prediction accuracy across diverse species, a principle that could have broad applications in both agricultural and biomedical predictive modeling.
Why for Yiru: The ensemble learning strategy for genomic prediction offers methodological insights applicable to my work — similar ensemble approaches could improve predictive models for patient stratification and treatment response in cancer immunotherapy by integrating diverse omics data types.
Biomedical discoveries
Biomedicine
Localized PD-1 CAR T therapy reprograms neuroinflammation
Cell Published 2026-07-21 immuno_therapy DOI: 10.1016/j.cell.2026.06.036
CAR-T immunotherapy neuroinflammation PD-1 T cell
Summary: Localized delivery of PD-1-targeted CAR T cells reprograms the neuroinflammatory microenvironment. The study demonstrates that regional administration of CAR T cells directed against PD-1 effectively attenuates neuroinflammation, revealing a new therapeutic paradigm for applying CAR T cell technology beyond oncology to inflammatory and autoimmune conditions of the central nervous system.
Why it matters: Expands the therapeutic scope of CAR T cells beyond cancer to neuroinflammatory disease, demonstrating that localized delivery can effectively reprogram tissue microenvironments without the systemic toxicity concerns associated with traditional immunosuppressive therapies.
Why for Yiru: Highly relevant to my CAR-T and immunotherapy interests — the concept of localized CAR T delivery to reprogram tissue microenvironments could be adapted for solid tumors, where regional administration may overcome trafficking barriers and improve therapeutic index in the tumor microenvironment.
Quiescent tumor cells shape the immunosuppressive microenvironment in pancreatic cancer
Nature Communications Published 2026-07-21 cancer_immunology DOI: 10.1038/s41467-026-75883-z
pancreatic cancer tumor microenvironment CAR-T quiescent cells Epiregulin macrophage immunotherapy resistance
Summary: A rare population of quiescent PDAC cells increases after CAR-T therapy, expressing high levels of Epiregulin (EREG) which induces an immunosuppressive tumor microenvironment via ErbB4-expressing macrophages. Targeting EREG enhances CAR-T sensitivity, identifying a resistance mechanism driven by therapy-induced quiescence and macrophage reprogramming in pancreatic cancer.
Why it matters: Reveals a therapy-induced resistance mechanism in which quiescent tumor cells actively remodel the microenvironment through EREG-ErbB4 macrophage signaling, providing a targetable axis to improve CAR-T efficacy in pancreatic cancer — one of the most immunotherapy-resistant solid tumors.
Why for Yiru: Directly relevant to my interests in CAR-T, macrophage biology, and the tumor microenvironment — the finding that quiescent tumor cells communicate with macrophages to drive immunosuppression identifies a targetable vulnerability that could be exploited in combination with CAR-T therapy.
A second wind for leukemic lungs
Nature Immunology Published 2026-07-20 cancer_immunology DOI: 10.1038/s41590-026-02598-0
AML leukemia lung tumor microenvironment single-cell spatial T cell galectin-9 inflammation
Summary: Single-cell and spatial analyses of acute myeloid leukemia (AML) in the lung reveal an inflamed and remodeled microenvironment characterized by altered T cell populations and elevated galectin-9 and IL-33 signaling. Corticosteroids or antibodies targeting galectin-9 or IL-33 reduce leukemic infiltration and restore respiratory function, identifying potential therapeutic targets for AML-associated lung pathology.
Why it matters: Applies cutting-edge single-cell and spatial transcriptomics to characterize the leukemic niche in the lung, revealing targetable inflammatory pathways that could improve respiratory outcomes in AML patients — a clinically important but understudied aspect of the disease.
Why for Yiru: Directly relevant to my spatial transcriptomics and tumor microenvironment research — the application of single-cell and spatial approaches to characterize AML lung infiltration provides a template for how these technologies can reveal niche-specific therapeutic vulnerabilities in hematologic malignancies.
Tumor-derived stearic acid induces macrophage Egr2 signaling to suppress anti-tumor immunity in breast cancer
bioRxiv (immunology) Published 2026-07-15 cancer_immunology DOI: 10.64898/2026.07.15.738781
macrophage breast cancer tumor microenvironment stearic acid lipid metabolism T cell immunosuppression
Summary: Tumor-derived stearic acid induces macrophage Egr2 signaling to suppress anti-tumor immunity in breast cancer. The study identifies a lipid-mediated immunosuppressive axis in which fatty acids from the tumor microenvironment reprogram macrophage function through the transcription factor Egr2, leading to impaired T cell activation and anti-tumor immunity.
Why it matters: Uncovers a previously unrecognized lipid-dependent immunosuppressive mechanism through which tumor-derived metabolites reprogram macrophage function, opening new avenues for targeting metabolic-immune crosstalk in the tumor microenvironment.
Why for Yiru: Highly relevant to my macrophage and tumor microenvironment interests — the identification of stearic acid as a tumor-derived signal that reprograms macrophage immunosuppressive function reveals a metabolite-immune axis that could be targeted in combination with immunotherapies.
Mapping Functional Tumor Suppressor Networks in Esophageal Adenocarcinoma Using In Vivo CRISPR Screening and Perturb-sequencing
bioRxiv (cancer_biology) Published 2026-07-19 functional_genomics DOI: 10.64898/2026.07.19.739473
CRISPR screen perturb-seq cancer esophageal adenocarcinoma tumor suppressor functional genomics
Summary: In vivo CRISPR screening and Perturb-sequencing map functional tumor suppressor networks in esophageal adenocarcinoma. The study combines CRISPR-based genetic perturbation with single-cell transcriptomic readouts to systematically characterize how tumor suppressor loss remodels the cellular landscape and identifies network-level vulnerabilities.
Why it matters: Demonstrates the power of combining in vivo CRISPR screening with Perturb-sequencing to dissect tumor suppressor networks at single-cell resolution, providing a roadmap for functional genomics approaches to study gene function in physiologically relevant cancer models.
Why for Yiru: Directly relevant to my CRISPR screen and perturbation biology interests — the integration of in vivo CRISPR screening with single-cell transcriptomic readouts (Perturb-seq) is exactly the kind of multi-omics functional approach that can reveal how genetic perturbations reshape the tumor microenvironment and immune landscape.
Engineering Oncolytic Measles Virus with MG53 Couples Pyroptotic Tumor Killing with Immune Microenvironment Remodeling to Enhance Checkpoint Immunotherapy in Non-small Cell Lung Cancer
bioRxiv (immunology) Published 2026-07-15 oncolytic_virotherapy DOI: 10.64898/2026.07.15.738733
oncolytic virus immunotherapy tumor microenvironment pyroptosis lung cancer checkpoint blockade MG53
Summary: Engineering oncolytic measles virus with MG53 couples pyroptotic tumor killing with immune microenvironment remodeling to enhance checkpoint immunotherapy in NSCLC. The engineered virus simultaneously induces immunogenic cell death through pyroptosis and remodels the tumor microenvironment to overcome resistance to PD-1/PD-L1 checkpoint blockade.
Why it matters: Presents a multimodal oncolytic virus strategy that combines direct tumor killing via pyroptosis with immune microenvironment remodeling, offering a rational combinatorial approach to enhance checkpoint immunotherapy efficacy in immunologically cold tumors like NSCLC.
Why for Yiru: Relevant to my immunotherapy and tumor microenvironment interests — the concept of engineering oncolytic viruses to simultaneously kill tumor cells and remodel the immune microenvironment addresses key challenges in converting immunologically cold tumors to hot, which is central to improving checkpoint immunotherapy responses.
Cross-disciplinary watchlist
Other Fields
Ten years of mapping gene expression in tissues
Nature Published 2026-07-21 news_and_views DOI: 10.1038/d41586-026-02212-1
spatial transcriptomics gene expression tissue mapping technology review
Summary: Spatially resolved transcriptomic technologies continue to enhance understanding of development and disease through a feedback loop involving academia and industry. This retrospective marks a decade of progress in spatial transcriptomics, from early proof-of-concept studies to commercial platforms enabling routine spatial profiling at unprecedented resolution and throughput.
Why it matters: Provides a timely retrospective on the explosive growth of spatial transcriptomics over the past decade, highlighting how the field has matured from niche technology to a mainstream approach with transformative potential across development, neuroscience, and cancer research.
Why for Yiru: Directly aligned with my core interest in spatial transcriptomics — this review offers a valuable perspective on where the field has been, current capabilities, and emerging directions that will shape the next wave of spatial methods and applications in cancer biology.
A genomic catalog of the mouse gut virome reveals features associated with ageing
Nature Communications Published 2026-07-21 genomics_resource DOI: 10.1038/s41467-026-75836-6
virome microbiome ageing mouse gut metagenomics genomics
Summary: The Mouse Reference Gut Virome (MRGV) comprises 109,778 viral genomes and 28,824 species, expanding known mouse gut viral diversity by approximately 67.6%. Age-associated features of the virome are identified, revealing shifts in viral community composition and functional potential linked to host ageing.
Why it matters: Provides a comprehensive reference resource for the mouse gut virome, substantially expanding known viral diversity and enabling future studies of how the viral component of the microbiome influences host physiology and ageing.
Why for Yiru: While outside my primary focus areas, the metagenomic cataloging approach and the integration of age-related metadata demonstrate principles of large-scale reference resource construction that are relevant to building comprehensive atlases in spatial transcriptomics and single-cell biology.
Pangenome-resolved structural variation drives adaptation and trait evolution in cucumber
Nature Genetics Published 2026-07-21 plant_genomics DOI: 10.1038/s41588-026-02682-z
pangenome structural variation cucumber adaptation crop genomics trait evolution
Summary: High-quality genome assemblies for 125 cucumber accessions with 135,597 structural variations identified, including the discovery of the CsCcu R gene for scab resistance and a CsFT tandem duplication associated with flowering adaptation. The pangenome resource reveals how structural variation drives adaptation and trait diversification.
Why it matters: Demonstrates the power of pangenome-resolved analysis at scale, revealing how structural variation shapes trait evolution and adaptation in a major crop species — a framework applicable to understanding genomic diversity across plant and animal species.
Why for Yiru: The pangenome and structural variation analysis framework offers methodological insights for my genomics work — similar approaches could be applied to characterize structural variation in immune gene families and its impact on immunotherapy responses.
Daily briefing: CRISPR gets an AI-designed upgrade
Nature Published 2026-07-17 research_highlight DOI: 10.1038/d41586-026-02272-3
CRISPR AI genome editing protein design synthetic biology
Summary: Synthetic CRISPR proteins designed by AI edit the genome more efficiently than natural counterparts. This research highlight summarizes recent advances in AI-guided protein design applied to genome editing, where machine learning models generate novel CRISPR-associated nucleases with enhanced editing efficiency and specificity compared to naturally occurring variants.
Why it matters: Demonstrates that AI-designed proteins can outperform evolutionarily optimized natural proteins for genome editing, marking a milestone in computational protein design and opening the door to custom-designed molecular tools for biotechnology and therapeutics.
Why for Yiru: Relevant to my AI and genome editing interests — the AI-guided design of improved CRISPR systems showcases the power of foundation models for protein engineering, a paradigm that could extend to designing improved CAR constructs or other therapeutic proteins for immunotherapy.
High-throughput antigen discovery using Functional Genomic Vaccinology (FGV) identifies protective Streptococcus pneumoniae vaccine candidates
Nature Communications Published 2026-07-21 vaccine_development DOI: 10.1038/s41467-026-75848-2
vaccinology antigen discovery Streptococcus pneumoniae functional genomics bacterial vaccine
Summary: FGV is a high-throughput antigen discovery platform integrating genome-wide prediction, proteome-scale screening, and immunogenicity validation for bacterial vaccine development. Applied to S. pneumoniae, the platform identified protective vaccine candidates that elicited robust immune responses and protection in preclinical models.
Why it matters: Establishes a systematic, high-throughput platform for rational antigen discovery that could accelerate vaccine development for bacterial pathogens, addressing a critical need for novel vaccines against antimicrobial-resistant organisms.
Why for Yiru: The high-throughput functional genomics approach to antigen discovery offers methodological parallels to my interests in systematic screening — similar platforms could be adapted for tumor antigen discovery in cancer vaccine development.
A bivalent molecular glue linking lysine acetyltransferases to oncogene-induced cell death
Cell Published 2026-07-20 chemical_biology DOI: 10.1016/j.cell.2026.06.037
molecular glue targeted therapy lymphoma DLBCL protein degradation cancer therapy
Summary: A bivalent small molecule modality that kills DLBCL cells at sub-nanomolar potency through induced proximity, linking lysine acetyltransferases to oncogene-induced cell death. Structural and mechanistic studies demonstrate that fortuitous protein-protein contacts convert proximity induction into targeted changes in chromatin, revealing a key mechanism by which small molecules can co-opt oncogenic transcriptional regulators to elicit malignant cell death.
Why it matters: Establishes a new paradigm for targeted cancer therapy using bivalent molecular glues that reprogram oncogenic transcription factors through induced proximity, opening a new chemical space for drug discovery beyond traditional inhibition and degradation strategies.
Why for Yiru: The molecular glue mechanism for redirecting epigenetic regulators is conceptually relevant to my interests in protein-protein interactions and immunotherapy — similar proximity-inducing strategies could potentially be applied to redirect immune cell signaling in the tumor microenvironment or engineer synthetic receptor systems.