Research Radar — 2026-05-09
Methods & AI
Computational
AI-predicted spatial transcriptomics unlocks breast cancer biomarkers from pathology
Cell Published 2026-05-08 research article DOI:
spatial transcriptomics computational pathology breast cancer AI tumor microenvironment
Summary: Path2Space predicts spatial gene expression directly from histopathology, enabling low-cost, large-scale characterization of the tumor microenvironment. The inferred spatial breast cancer landscapes reveal clinically relevant subgroups and improve the prediction of treatment response from routine H&E slides alone.
Why it matters: Spatial transcriptomics is powerful but expensive and low-throughput. AI-based inference from routine pathology slides could democratize spatial TME analysis, making it feasible for large clinical cohorts and retrospective studies.
Why for Yiru: Core spatial omics paper with direct AI methodology. Predicting spatial gene expression from H&E images is a transformative approach that bridges computational pathology, spatial biology, and clinical translation — all central interests.
Rapid directed evolution guided by protein language models and epistatic interactions
Science Published 2026-05-08 research article DOI: 10.1126/science.aea1820
protein language models directed evolution protein engineering epistasis machine learning
Summary: Develops a protein language model-guided approach for rapid directed evolution that explicitly models epistatic interactions between mutations, enabling efficient navigation of high-dimensional sequence space to find synergistic mutation combinations that outperform stepwise stacking and traditional ML methods.
Why it matters: Protein engineering's central challenge is the combinatorial explosion of sequence space. PLM-guided evolution that accounts for epistasis could dramatically accelerate enzyme design, therapeutic protein optimization, and biosensor development.
Why for Yiru: Protein language models applied to a core biotechnology problem. The epistasis modeling angle connects to broader interests in representation learning for biological sequences and structure-function relationships.
PromptBio-Bench: Benchmarking LLM-based Bioinformatics Agents for End-to-End Data Analysis
bioRxiv Published 2026-05-05 preprint DOI: 10.1101/2026.05.05.723092
LLM agents bioinformatics benchmarking automated analysis
Summary: Introduces a comprehensive evaluation suite of 194 expert-curated bioinformatics and data science tasks at varied difficulty levels, with structured file comparison against expert reference answers. Three state-of-the-art agents show comparable performance, with accuracy declining markedly at higher difficulty levels.
Why it matters: LLM-based agents are increasingly proposed for automating bioinformatics workflows, but systematic evaluation has been lacking. A rigorous benchmark is essential for tracking progress and identifying failure modes before clinical or research deployment.
Why for Yiru: Directly relevant to how Boss might use AI agents in bioinformatics pipelines. Understanding agent capabilities and limitations informs tool selection and workflow design for spatial and single-cell analysis.
BART-spatial unravels biologically significant transcriptional regulators from spatial omics data
bioRxiv Published 2026-05-05 preprint DOI: 10.1101/2026.05.05.723027
spatial omics transcriptional regulators gene regulation computational method
Summary: BART-spatial integrates spatial variability and pseudo-temporal information with publicly available TR binding profiles to infer functional transcriptional regulators from spatial omics data. Outperforms existing methods across multiple platforms including 10X Visium, Visium HD, Atera, and spatial RNA-ATAC-seq, identifying regulators undetectable by expression alone.
Why it matters: TR activity often doesn't correlate with mRNA levels, and existing tools ignore spatial heterogeneity. BART-spatial fills a critical gap for decoding gene regulatory programs in their tissue context.
Why for Yiru: Spatial gene regulation is a core interest. Identifying active TRs from spatial omics data directly supports research into tissue organization, TME architecture, and cell state regulation.
FILM: mapping organellar metabolism by mid-infrared photothermal-modulated fluorescence
Nature Methods Published 2026-05-07 research article DOI: 10.1038/s41592-026-03090-1
metabolic imaging organelles mid-infrared photothermal microscopy single-cell
Summary: FILM is a mid-infrared photothermal microscopy variant using optical boxcar demodulation-based illumination, denoising, and spectral deconvolution. Its gentle nature allows imaging of metabolic processes inside organelles in cell culture and in C. elegans.
Why it matters: Organelle-level metabolism has been largely invisible to conventional imaging. FILM opens a window into subcellular metabolic dynamics, potentially revealing how metabolic states differ between organelles in health and disease.
Why for Yiru: Spatial biology at unprecedented resolution — subcellular metabolic imaging. This technology could complement spatial transcriptomics by adding a metabolic layer to tissue architecture analysis.
CroCoDeEL: accurate control-free detection of cross-sample contamination in metagenomic data
Nature Communications Published 2026-05-09 research article DOI: 10.1038/s41467-026-72637-9
metagenomics contamination detection bioinformatics quality control microbiome
Summary: Presents CroCoDeEL, a control-free computational method for detecting cross-sample contamination in metagenomic studies. The method identifies contamination without requiring negative controls, making it applicable to existing datasets and large-scale meta-analyses.
Why it matters: Cross-sample contamination is a pervasive problem in metagenomics that can produce spurious findings. A control-free detection method enables retrospective quality assessment of published studies and improves reliability of meta-analyses.
Why for Yiru: Computational quality control for high-throughput sequencing data. The control-free approach is methodologically interesting and applicable to other omics contexts where contamination is a concern.
Biomedical discoveries
Biomedicine
Multimodal clocks of human aging
Cell Published 2026-05-08 research article DOI:
aging multimodal biomarkers organ aging coagulation
Summary: A multidimensional framework for quantifying human aging reveals highly conserved aging biomarkers and trajectories across multiple centers. The study demonstrates that accumulation of liver-derived coagulation factors acts as a mechanistic driver of vascular and systemic aging, with organs aging asynchronously.
Why it matters: Understanding aging as a multisystem process with organ-specific clocks could enable targeted anti-aging interventions. The liver-coagulation-vascular axis provides a mechanistic link between organ function and systemic aging.
Why for Yiru: Multimodal systems biology of aging with translational implications. The organ asynchrony concept and coagulation-driven vascular aging are novel mechanistic insights with potential clinical relevance.
Interleukin-17-Producing γδ T Cells Originate from SOX13+ Progenitors that Are Independent of γδTCR Signaling
Immunity Published 2026-05-08 research article DOI:
γδ T cells IL-17 T cell development SOX13 innate immunity
Summary: Identifies SOX13+ progenitors as the developmental origin of IL-17-producing γδ T cells, demonstrating that this lineage commitment occurs independently of γδTCR signaling — revising the paradigm that TCR signals are universally required for γδ T cell functional programming.
Why it matters: IL-17-producing γδ T cells are critical in barrier immunity, autoimmunity, and cancer. Understanding their developmental origin could enable therapeutic manipulation of this lineage for inflammatory disease and tumor immunity.
Why for Yiru: T cell biology and development. γδ T cells are an understudied but important immune population in the TME, and their developmental wiring has implications for understanding tissue-resident immunity.
Ferroptosis inhibition enhances liver and lung graft function
Cell Published 2026-05-08 research article DOI:
ferroptosis transplantation ischemia-reperfusion injury liver lung
Summary: Veeckmans et al. demonstrate that ischemia-reperfusion injury triggers an early lipid peroxidation wave in human transplants. Pharmacological ferroptosis inhibition with FXT-001 suppresses lipid radical propagation, modulates iron homeostasis, and improves liver and lung graft function in porcine and human perfusion models.
Why it matters: Ischemia-reperfusion injury limits transplant success and organ availability. Targeting ferroptosis — a regulated cell death pathway driven by lipid peroxidation — could expand the donor organ pool and improve transplant outcomes.
Why for Yiru: Ferroptosis is emerging as a key cell death mechanism in cancer and tissue injury. The translational success in large animal models and human tissue is notable, and ferroptosis biology intersects with tumor immunology.
Complex I protein NDUFB9 is a metabolic vulnerability in triple negative breast cancer brain metastases
Nature Communications Published 2026-05-09 research article DOI: 10.1038/s41467-026-72927-2
breast cancer brain metastasis metabolic vulnerability NDUFB9 Complex I
Summary: Identifies the mitochondrial Complex I subunit NDUFB9 as a selective metabolic dependency in triple-negative breast cancer brain metastases. Targeting NDUFB9 disrupts oxidative phosphorylation specifically in the brain metastatic niche, revealing a therapeutically exploitable metabolic adaptation.
Why it matters: Brain metastases are a devastating complication of breast cancer with few treatment options. Organ-specific metabolic vulnerabilities represent a precision medicine approach to treating metastasis based on the target organ microenvironment.
Why for Yiru: Cancer metabolism in the metastatic niche. The concept of organ-specific metabolic adaptations is relevant to understanding how the TME varies across metastatic sites and how to target these differences therapeutically.
Anti-PD-1 plus nab-paclitaxel and bevacizumab for second-line treatment of cancer of unknown primary (Fudan CUP-002): a phase II trial
Nature Communications Published 2026-05-09 research article DOI: 10.1038/s41467-026-72745-6
cancer of unknown primary immunotherapy anti-PD-1 clinical trial phase II
Summary: Reports results of a phase II trial combining anti-PD-1 with nab-paclitaxel and bevacizumab as second-line therapy for cancer of unknown primary. The triplet regimen shows promising efficacy in this difficult-to-treat population where tissue-of-origin remains unknown and treatment options are limited.
Why it matters: Cancer of unknown primary accounts for 3-5% of malignancies and has a dismal prognosis. An immunotherapy-chemotherapy-antiangiogenic combination showing efficacy could establish a new standard of care in a disease with few evidence-based options.
Why for Yiru: Clinical immunotherapy in a challenging cancer setting. CUP is an interesting model for studying how immunotherapy performs when the tumor's tissue of origin — and thus its immune contexture — is unknown.
Transposable elements shape stemness in normal and leukemic hematopoiesis
Nature Genetics Published 2026-05-04 research article DOI: 10.1038/s41588-026-02585-z
transposable elements stemness leukemia hematopoiesis epigenetics
Summary: Identifies distinct transposable element subfamilies as genetic determinants of stemness properties in normal and leukemic stem populations. Specific TE families are differentially activated in hematopoietic stem cells versus leukemic stem cells, with clinical implications for acute myeloid leukemia.
Why it matters: Transposable elements have long been considered genomic parasites, but this study positions them as active regulators of stem cell identity. TE-mediated stemness regulation could reveal new therapeutic targets in leukemia.
Why for Yiru: Epigenetic regulation of cell state with a genomics angle. Transposable elements as functional regulators connects to broader interests in how non-coding genome elements shape cellular identity in normal and malignant contexts.
Cross-disciplinary watchlist
Other Fields
Large language models exhibit speciesist bias against animals
Nature Communications Published 2026-05-09 research article DOI: 10.1038/s41467-026-72297-9
LLM AI ethics speciesism bias moral reasoning
Summary: Demonstrates that large language models detect speciesist statements but often reproduce mainstream moral reasoning that treats harm toward animals as acceptable. LLMs reflect human-like trade-offs in moral decision-making, highlighting the need to extend AI fairness frameworks beyond anthropocentric boundaries.
Why it matters: AI ethics has focused primarily on human-centric biases. Expanding fairness frameworks to include non-human animals represents a frontier in AI alignment research and challenges assumptions about what constitutes ethical AI behavior.
Why for Yiru: AI ethics and bias from a novel angle. The finding that LLMs reproduce human moral trade-offs rather than correcting them has implications for how AI systems are deployed in value-laden domains including biomedical research.
Proactive collaboration via autonomous interaction
Nature Communications Published 2026-05-09 research article DOI: 10.1038/s41467-026-72797-8
multi-agent systems robotics autonomous collaboration AI team coordination
Summary: Shows that proactive collaboration — driven by team initiatives rather than centralized control — enables robot teams to anticipate needs and reorganize themselves by recruiting or releasing teammates. The approach improves performance in both real-world and simulated tasks.
Why it matters: Autonomous multi-agent coordination is a grand challenge in robotics and AI. Proactive collaboration that anticipates team needs rather than merely reacting to commands could enable more robust robot teams for disaster response, manufacturing, and space exploration.
Why for Yiru: Multi-agent AI systems with emergent coordination behavior. The principles of autonomous team reorganization have conceptual parallels to cellular collective behaviors in tissue biology and immune responses.
Using satellite imagery to map rural marketplaces and monitor their activity at high frequency
Nature Communications Published 2026-05-09 research article DOI: 10.1038/s41467-026-72865-z
satellite imagery computer vision development economics remote sensing AI
Summary: Uses distinctive signatures in globally available satellite imagery to detect rural weekly markets and track their activity at high frequency, addressing a critical data gap in development economics where systematic market data have been sparse.
Why it matters: Rural markets are economic lifelines for billions of people, yet data on their location and activity are nearly nonexistent. Satellite-based monitoring could transform economic measurement, humanitarian response, and infrastructure planning in developing regions.
Why for Yiru: AI-powered computer vision applied to a non-biomedical domain with significant societal impact. The approach of extracting economic signals from imagery has conceptual parallels to extracting biological signals from tissue images.