Research Radar — 2026-08-04
Methods & AI
Computational
SpatialFormer: universal spatial representation learning from subcellular molecular to multicellular landscapes
Nature Computational Science Published 2026-07-29 Research Article DOI: 10.1038/s43588-026-01016-7
spatial transcriptomics representation learning cell-cell relationships cell communication
Summary: SpatialFormer is a transformer-based model for spatial transcriptomics that learns cell–cell relationships to predict cell co-localization, annotate cell types and niches, and identify cell communication gene pairs in pulmonary fibrosis.
Why it matters: It treats spatial organization as a learnable representation rather than only as a downstream visualization, linking cellular neighborhoods with communication signals in diseased tissue.
Why for Yiru: This is directly aligned with your spatial omics and tumor microenvironment interests, especially the goal of learning spatially organized cell states and communication patterns.
Continual integration of single-cell multimodal data with MIRACLE
Nature Computational Science Published 2026-07-30 Research Article DOI: 10.1038/s43588-026-01030-9
single-cell multimodal integration continual learning batch correction
Summary: MIRACLE is an online continual-learning framework that integrates single-cell multimodal data across batches, modalities, tissues and diseases without requiring the full collection to be processed as one static dataset.
Why it matters: Continual integration addresses a practical scaling problem for growing single-cell atlases, where new tissues, modalities and disease cohorts arrive over time.
Why for Yiru: The method is relevant to your computational biology and spatial multi-omics work because it offers a way to keep heterogeneous reference atlases usable as new experiments are added.
Inference of secreted protein signaling activities in intercellular communication
Nature Methods Published 2026-07-29 Research Article DOI: 10.1038/s41592-026-03172-0
cell-cell communication secreted proteins spatial transcriptomics single-cell transcriptomics
Summary: SecAct is a computational framework that leverages transcriptomic data to infer secreted-protein signaling activities and quantify intercellular communication across tissues and disease contexts.
Why it matters: It extends cell–cell communication analysis beyond a narrow set of ligand–receptor pairs by making secreted-protein signaling activities systematically inferable from transcriptomic data.
Why for Yiru: This directly supports your spatial omics and tumor microenvironment interests, where inferred secreted-factor activity can connect local cell states with immune suppression and therapy resistance.
Unify learns cellular evolution with universal multimodal embeddings
Nature Communications Published 2026-07-30 Research Article DOI: 10.1038/s41467-026-76230-y
single-cell cross-species integration multimodal embeddings transfer learning
Summary: Unify integrates single-cell RNA-sequencing data across species by combining RNA expression with embeddings from protein language models and general-purpose language models.
Why it matters: Cross-species integration is essential for transferring cell-state knowledge between model organisms and human disease, but expression-only alignment can miss conserved biological context.
Why for Yiru: The multimodal embedding strategy is useful for your computational immunology and spatial atlas interests, particularly when comparing immune and stromal states across experimental systems.
End-to-end multimodal pathology foundation model with clinical dialogue
Nature Medicine Published 2026-07-30 Research Article DOI: 10.1038/s41591-026-04521-4
computational pathology foundation model whole-slide imaging clinical AI
Summary: PRISM2 was trained on 2.3 million whole-slide images and 14 million clinical question–answer pairs. It matched clinical-grade cancer-detection performance without task-specific training, highlighting the value of clinical dialogue supervision.
Why it matters: The study tests how clinical language supervision can improve a pathology model beyond image-only pretraining and reduce dependence on task-specific fine-tuning.
Why for Yiru: This is highly relevant to your biomedical AI and spatial biology interests because it connects large-scale tissue imaging with clinically grounded multimodal representation learning.
Mapping enhancer–gene regulatory interactions from single-cell data
Nature Genetics Published 2026-08-02 Research Article DOI: 10.1038/s41588-026-02695-8
single-cell genomics enhancer–gene regulation gene regulation cell types
Summary: scE2G is a family of models that predicts enhancer–gene regulatory interactions from single-cell datasets and maps those interactions across diverse cell types and tissues.
Why it matters: It provides a route from cell-resolved regulatory data to candidate enhancer–gene links, helping connect noncoding regulation with cellular identity and tissue context.
Why for Yiru: The method complements your single-cell and spatial omics interests by offering a regulatory layer for interpreting cell states and their tissue-specific programs.
Biomedical discoveries
Biomedicine
CRISPR screens identify targets to rescue age-related T cell dysfunction in cancer
Cell Published 2026-07-28 Research Article DOI: 10.1016/j.cell.2026.07.016
CRISPR screen T cell dysfunction aging tumor microenvironment
Summary: The study identifies Dusp5 and Zfp219 as regulators of age-related T cell dysfunction. Loss of either gene drives T cells toward an effector-like state and improves tumor control through distinct mechanisms in the aged tumor microenvironment.
Why it matters: It links genetic perturbations to age-dependent T cell state and tumor control, providing experimentally testable targets for overcoming immune dysfunction in older hosts.
Why for Yiru: This directly fits your computational immunology, tumor microenvironment and CAR-T interests by connecting CRISPR perturbation with T cell state engineering.
Subclinical cholestasis is a hallmark of gut dysbiosis causing resistance to cancer immunotherapy
Cancer Cell Published 2026-07-28 Research Article DOI: 10.1016/j.ccell.2026.06.022
gut microbiome gut–liver axis immunotherapy resistance tumor immunosurveillance
Summary: The study shows that gut dysbiosis impairs tumor immunosurveillance through the gut–liver axis, causing subclinical cholestasis and resistance to immunotherapy in tumors distant from gastrointestinal and hepatic tissues.
Why it matters: The associated changes in plasma γ-glutamyl transferase, bile acids and sMAdCAM-1 suggest a systemic metabolic and vascular route by which gut dysbiosis can shape responses to cancer therapy.
Why for Yiru: This is relevant to your tumor microenvironment and computational immunology interests because it connects systemic host state, tissue crosstalk and immunotherapy response.
BCAT2 links branched-chain amino acids metabolism to interferon signaling to sustain macrophage inflammation
Nature Immunology Published 2026-07-30 Research Article DOI: 10.1038/s41590-026-02604-5
macrophages branched-chain amino acid metabolism interferon signaling inflammation
Summary: The study finds that BCAT2-dependent branched-chain amino acid catabolism sustains interferon-driven macrophage activation and chronic inflammation during autoimmune arthritis.
Why it matters: It identifies a metabolic dependency that couples nutrient processing to persistent interferon signaling and inflammatory macrophage states.
Why for Yiru: The metabolism–macrophage connection is useful for your tumor microenvironment and computational immunology interests, where myeloid state and nutrient context often co-vary.
Progranulin deficiency-induced lysosomal dysfunction drives maladaptive myeloid cell states through the MITF/TFE transcription factors
Immunity Published 2026-08-02 Research Article DOI: 10.1016/j.immuni.2026.07.009
progranulin lysosomes microglia macrophages
Summary: The study shows that lysosomal dysfunction is sufficient to remodel microglia and macrophages epigenetically, transcriptionally and functionally, including in progranulin deficiency, and connects lysosomal function with microglial state.
Why it matters: It places lysosomal function upstream of maladaptive myeloid-state remodeling and links a cellular maintenance pathway to neurodegenerative disease biology.
Why for Yiru: This is relevant to your interest in macrophage and myeloid states, and the state-remodeling framework may inform how you interpret immune-cell heterogeneity in tissue data.
Tumor-hepatocyte crosstalk drives a hepatic lactate-TGF-β axis of CD8⁺ T cell exhaustion and immunotherapy resistance in small-cell lung cancer liver metastases
bioRxiv: cancer biology Published 2026-07-30 preprint DOI: 10.64898/2026.07.30.741857
liver metastasis spatial transcriptomics CD8⁺ T cell exhaustion lactate TGF-β
Summary: Using clinical data, multi-region single-cell RNA sequencing, spatial transcriptomics and metabolic assays, the study links SCLC–hepatocyte crosstalk to a lactate- and TGF-β-rich niche that suppresses CD8⁺ T cell function. TGF-β receptor inhibition restored CD8⁺ T cell proliferation in the reported models.
Why it matters: It proposes convergent metabolic and cytokine signals, including H3K18 lactylation at regulatory loci, as a mechanism for liver-metastasis-associated T cell dysfunction and immunotherapy resistance.
Why for Yiru: This is exceptionally aligned with your spatial omics, tumor microenvironment and computational immunology interests because it integrates spatial cell states with a mechanistic resistance axis.
MLL4/KMT2D mutations increase immune activity and predict therapy efficacy in colorectal cancer
bioRxiv: cancer biology Published 2026-08-02 preprint DOI: 10.64898/2026.07.31.742076
KMT2D/MLL4 colorectal cancer tumor immunity immune checkpoint therapy
Summary: Across genomic and clinical datasets, KMT2D-mutated colorectal cancers showed higher immune-checkpoint and CD8 expression, stronger T-effector and interferon-γ signatures, and more CD8⁺ T cell, NK cell and macrophage infiltration. The mutation was associated with better immune-checkpoint-inhibitor outcomes and greater cisplatin sensitivity in the reported analyses.
Why it matters: The findings connect chromatin-regulator loss to an immunologically hot colorectal cancer phenotype and suggest that mutation status may help stratify therapy response.
Why for Yiru: This connects tumor genotype to immune-state features, which is useful for your computational immunology and tumor microenvironment interests and for thinking about response-predictive biomarkers.
Cross-disciplinary watchlist
Other Fields
Spatial atlas of the human brain vasculature reveals specialized cell ensembles
Cell Published 2026-07-30 Resource DOI: 10.1016/j.cell.2026.07.007
spatial transcriptomics brain vasculature cell atlas vascular cell ensembles
Summary: An integrative atlas of 314,535 transcriptomes and spatial profiling of 1,529,740 cells in human temporal cortex and hippocampus reveals vascular cell ensembles spanning endothelial, mural, fibroblast and perivascular macrophage populations.
Why it matters: The resource connects arteriovenous architecture with specialized cellular communities and disease vulnerability, providing a detailed reference for human cerebrovascular organization.
Why for Yiru: The atlas is a strong methodological reference for your spatial omics interests, especially for studying how spatially organized vascular and immune niches shape tissue function.
Quantifying protein unfolding kinetics with a high-throughput microfluidic platform
Cell Systems Published 2026-07-28 Research Article DOI: 10.1016/j.cels.2026.101681
protein biophysics microfluidics protein unfolding high-throughput assay
Summary: The microfluidic platform expresses and immobilizes EGFP-tagged protein variants in device chambers, uses thermolysin cleavage as proteins unfold, and quantifies unfolding rates by microscopy while using minimal reagents.
Why it matters: It turns protein unfolding kinetics into a parallelizable measurement, making comparative biophysical characterization more scalable and resource-efficient.
Why for Yiru: This is relevant to your biomedical AI and protein-engineering interests as an example of a compact, quantitative assay that could generate data for predictive modeling.
Longitudinal dynamics of the maternal gut virome associate with metabolic features of preterm birth
Nature Communications Published 2026-07-31 Research Article DOI: 10.1038/s41467-026-76220-0
maternal gut virome preterm birth longitudinal multi-omics metabolism
Summary: Longitudinal multi-omics analyses indicate that disruption of the maternal gut virome precedes preterm birth and is associated with metabolic alterations; viral signatures improved prediction of preterm birth in the reported analyses.
Why it matters: The temporal association supports the virome as a potentially informative layer for risk assessment rather than treating microbial composition as a static snapshot.
Why for Yiru: The longitudinal multi-omics design is relevant to your interest in integrating molecular states over time and could inform broader computational approaches to host–microbe biology.
A massively parallel CRISPR-based screening platform for modifiers of neuronal depolarization
Nature Communications Published 2026-07-31 Research Article DOI: 10.1038/s41467-026-75882-0
CRISPR screening neuronal excitability functional genomics high-throughput screening
Summary: The study develops a scalable CRISPR screening platform to uncover genes that modify neuronal depolarization and thereby reveal genetic determinants of neuronal excitability in health and disease.
Why it matters: A massively parallel functional screen can move neuronal biology from candidate testing toward systematic discovery of regulators of cell-state behavior.
Why for Yiru: The platform is relevant to your perturbational biology and computational modeling interests, and its state-based screening logic may translate conceptually to immune-cell function studies.
Structure-alignment-driven cross-graph modeling for functional RNA design
Nature Computational Science Published 2026-07-29 Research Article DOI: 10.1038/s43588-026-01029-2
RNA design structure alignment graph modeling aptamers and ribozymes
Summary: AlignIF uses multiple structure alignments to capture evolutionary rules and conservation patterns across RNA families, enabling structure-guided design of functional aptamers and ribozymes.
Why it matters: The approach brings evolutionary structural information into graph modeling, providing a route to design diverse functional RNAs rather than optimizing sequence in isolation.
Why for Yiru: This is a useful biomedical AI example for your interest in structure-aware modeling and could inspire analogous ways to encode conservation and context in biological design tasks.
An AI-enabled structural atlas decodes kinase specificity across the human proteome
Nature Biotechnology Published 2026-07-28 Research Article DOI: 10.1038/s41587-026-03239-5
kinase specificity structural atlas AI human proteome
Summary: The study uses an AI-enabled structural atlas to assign phosphorylation potential and kinase specificity across the human proteome.
Why it matters: A proteome-scale view of kinase specificity can provide a systematic map for interpreting signaling regulation and prioritizing experimentally testable interactions.
Why for Yiru: The atlas-scale use of AI is relevant to your biomedical AI interests and offers a compact example of turning structural predictions into a proteome-wide signaling resource.