Research Radar — 2026-05-03
Methods & AI
Computational
Resolving sensitivity, specificity and signal contamination in Xenium spatial transcriptomics
Nature Methods Published 2026-04-30 research article DOI: 10.1038/s41592-026-03089-8
spatial transcriptomics computational biology signal deconvolution Xenium
Summary: Investigates technical noise in Xenium spatial transcriptomics data, including transcript spillover, and introduces SPLIT to resolve mixed signals and enhance cell-type specificity.
Why it matters: As Xenium becomes a widely adopted spatial platform, understanding and correcting its technical artifacts is critical for reliable biological conclusions.
Why for Yiru: Directly relevant to spatial omics methods evaluation — you work with spatial transcriptomics data and need to understand platform-specific noise characteristics.
eSIG-Net: an interaction language model that decodes the protein code of single mutations
Nature Methods Published 2026-04-29 research article DOI: 10.1038/s41592-026-03086-x
protein language model mutation effect prediction deep learning protein interactions
Summary: An interaction language model that predicts the effects of mutations on protein interaction, decoding how single amino acid changes alter binding interfaces.
Why it matters: Protein language models are evolving beyond structure prediction to functional interpretation — this represents a new capability class for variant effect prediction.
Why for Yiru: Relevant to computational immunology and protein engineering applications — predicting mutation effects on immune receptor interactions.
Interpretable, flexible and spatially aware integration of multiple spatial transcriptomics datasets from diverse sources
Nature Genetics Published 2026-04-27 research article DOI: 10.1038/s41588-026-02579-x
spatial transcriptomics data integration deep learning non-negative matrix factorization
Summary: INSPIRE combines deep learning with non-negative matrix factorization to integrate diverse spatial transcriptomics datasets, revealing shared and context-specific spatial gene programs across scales.
Why it matters: Cross-platform spatial data integration is a major unsolved problem — INSPIRE addresses the fragmentation of spatial transcriptomics data across technologies.
Why for Yiru: Directly applicable to multi-study spatial transcriptomics analyses and cross-platform benchmarking efforts.
Dango: Predicting higher-order genetic interactions
Cell Systems Published 2026-04-24 research article DOI: 10.1016/j.cels.2026.101593
genetic interactions hypergraph neural network self-attention genotype-to-phenotype
Summary: DANGO uses a self-attention hypergraph neural network to model higher-order genetic interactions, expanding the yeast trigenic interaction landscape to over 400 million interactions.
Why it matters: Moving beyond pairwise epistasis to higher-order interactions is essential for understanding complex genetic architectures — hypergraph neural networks offer a principled approach.
Why for Yiru: Methodological inspiration for modeling higher-order interactions in immune gene networks and combinatorial perturbations.
Modeling chimeric antigen receptor response at the single-cell level with conditional optimal transport
Cell Systems Published 2026-04-22 research article DOI: 10.1016/j.cels.2026.101591
CAR-T optimal transport single-cell computational immunology
Summary: CAROT is an optimal transport-based framework that predicts heterogeneous single-cell gene expression responses to CAR designs, enabling systematic exploration of CAR architecture-function relationships.
Why it matters: CAR-T design space is vast — computational frameworks that predict single-cell responses to design variations could accelerate rational engineering.
Why for Yiru: Directly intersects CAR-T biology with computational modeling using optimal transport — a mathematical framework you have expertise in.
Fusing imaging and metabolic modeling via multimodal deep learning in ovarian cancer
Cell Systems Published 2026-04-22 research article DOI: 10.1016/j.cels.2026.101594
multimodal deep learning metabolic modeling cancer imaging
Summary: Integrates patient-specific metabolic models, CT imaging, and transcriptomics in a multimodal deep-learning framework for ovarian cancer survival prediction with mechanistic interpretability.
Why it matters: Demonstrates how metabolic modeling can be fused with imaging and transcriptomics in a single deep learning pipeline — a paradigm for multi-omics integration.
Why for Yiru: Multimodal integration framework that could inspire approaches for combining spatial omics with metabolic and imaging readouts in tumor microenvironment studies.
Biomedical discoveries
Biomedicine
Distinct in vivo dynamics of donor-derived stem cell memory CAR T cells post-allogeneic HSCT relapse
Cell Published 2026-04-30 research article DOI: 10.1016/j.cell.2026.03.047
CAR-T TSCM stem cell memory allogeneic HSCT cancer immunotherapy
Summary: Donor-derived CD8+ CAR TSCM cells exhibit enhanced expansion and a favorable safety profile, inducing complete responses at low doses without lymphodepletion. Their distinctive in vivo behavior establishes TSCM cells as a robust platform for next-generation CAR T cell therapy.
Why it matters: Landmark clinical demonstration that TSCM-based CAR-T can achieve complete responses without lymphodepletion — potentially paradigm-shifting for allogeneic CAR-T.
Why for Yiru: Directly relevant to CAR-T biology and immunotherapy — TSCM differentiation state is a key axis for engineering more effective cellular therapies.
Proteostasis sustains T cell differentiation potential and tumor-infiltrating lymphocyte function
Cell Published 2026-04-29 research article DOI: 10.1016/j.cell.2026.02.019
T cell exhaustion proteostasis tumor-infiltrating lymphocytes immunotherapy E3 ubiquitin ligase
Summary: Exhausted T cells experience a breakdown of proteostasis characterized by unfolded protein accumulation. Targeting T cell proteostasis via E3 ubiquitin ligase activity rescues T cell function, highlighting a new axis for cancer immunotherapy.
Why it matters: Reveals proteostasis as a previously underappreciated determinant of T cell exhaustion — opens a new therapeutic axis beyond checkpoint blockade.
Why for Yiru: Connects T cell biology with proteostasis — relevant to understanding T cell dysfunction mechanisms in the tumor microenvironment.
Unbiased niche labeling maps immune-excluded niche in bone metastasis
Cell Published 2026-04-28 research article DOI: 10.1016/j.cell.2026.04.009
tumor microenvironment bone metastasis immune exclusion macrophages spatial biology
Summary: An unbiased niche-labeling method, SAMENT, maps cellular and molecular features of metastatic niches. Highlights the role of ERα+ niche macrophages in preventing T cell infiltration and promoting bone metastasis.
Why it matters: Novel method for unbiased niche mapping reveals a specific macrophage population driving immune exclusion in bone metastases — a prevalent and treatment-resistant metastatic site.
Why for Yiru: Combines spatial biology with immune exclusion mechanisms in metastasis — directly relevant to tumor microenvironment and macrophage biology interests.
Kupffer cell calibration of T cell responses via VSIG4–CD5 interaction promotes tumor evasion
Nature Immunology Published 2026-04-29 research article DOI: 10.1038/s41590-026-02510-w
Kupffer cells T cell liver metastasis immune checkpoint VSIG4 nanobody
Summary: Identifies a direct binding interaction between Kupffer cell VSIG4 and T cell CD5 that calibrates CD8+ T cell responses to liver metastasis. An anti-VSIG4 nanobody enhances immune checkpoint blockade in mice.
Why it matters: Discovers a new immune checkpoint axis (VSIG4-CD5) specific to the liver microenvironment — liver metastases are notoriously resistant to current immunotherapies.
Why for Yiru: Reveals organ-specific immune evasion mechanisms — relevant to understanding how tissue-resident macrophages shape anti-tumor immunity across metastatic sites.
Lymphoid tissue chemokines limit priming duration to preserve CD8+ T cell functionality
Science Published 2026-04-30 research article DOI: 10.1126/science.adq2080
CD8+ T cell T cell priming lymphoid tissue chemokines immunology
Summary: Shows that lymphoid tissue chemokines limit the duration of naive T cell-dendritic cell interactions during priming. This temporal constraint is essential for preserving CD8+ T cell effector functionality.
Why it matters: Reveals a temporal control mechanism for T cell priming that balances activation with preservation of function — fundamental insight for vaccine and immunotherapy design.
Why for Yiru: Basic T cell immunology insight with implications for understanding how priming duration affects CAR-T and adoptive cell therapy efficacy.
Disordered protein LAT encodes relative levels of signaling pathways in T cell activation
Science Published 2026-04-30 research article DOI: 10.1126/science.ads6847
T cell signaling LAT disordered protein single-cell screening
Summary: Develops a single-cell screening approach to interrogate how the disordered adapter protein LAT coordinates multiple downstream T cell signaling pathways, identifying widespread functional encoding in its disordered regions.
Why it matters: Shows how intrinsically disordered proteins encode signaling specificity — fundamental insight for T cell biology with implications for synthetic receptor design.
Why for Yiru: T cell signaling mechanisms are directly relevant to CAR-T design and understanding how synthetic receptors interface with endogenous signaling machinery.
Cross-disciplinary watchlist
Other Fields
Toward life with a 19–amino acid alphabet through generative artificial intelligence design
Science Published 2026-04-30 research article DOI: 10.1126/science.aeb5171
generative AI protein design synthetic biology amino acid alphabet
Summary: Leverages computational design and synthetic biology to explore building cells using only 19 canonical amino acids, demonstrating that generative AI can design functional proteins with a simplified building block alphabet.
Why it matters: Fundamental question about the minimal chemical requirements for life addressed through AI-driven protein design — pushes the boundaries of what generative models can achieve in molecular engineering.
Why for Yiru: Demonstrates the power of generative AI for biological design — a paradigm that extends beyond biomedicine into fundamental questions about the chemical basis of life.
Deeper detection limits in astronomical imaging using self-supervised spatiotemporal denoising
Science Published 2026-02-19 research article DOI: 10.1126/science.ady9404
self-supervised learning denoising astronomy computer vision spatiotemporal
Summary: Presents Astronomical Self-supervised Spatiotemporal Denoising, using self-supervised learning to correct correlated noise across pixels and exposures in astronomical imaging, substantially improving detection limits.
Why it matters: Self-supervised denoising that leverages spatiotemporal structure without requiring clean ground truth — a methodology transferable across imaging domains.
Why for Yiru: Self-supervised denoising approaches developed for astronomy could inform methods for spatial transcriptomics image processing and signal extraction.
Swimming with robots: investigating fish locomotion, sensing, and schooling behavior with robotic swimmers
Nature Communications Published 2026-05-02 perspective DOI: 10.1038/s41467-026-72478-6
robotics bio-inspired AI locomotion collective behavior closed-loop systems
Summary: Perspective on how adaptive, closed-loop robotic fish enable controlled tests of neuromechanical, sensorimotor and social feedback underlying fish behavior, shifting from engineering curiosities to experimental partners in biology.
Why it matters: Illustrates how robotics and AI are becoming experimental tools for biological discovery — the closed-loop paradigm is relevant across scientific domains.
Why for Yiru: The concept of using AI-driven closed-loop systems as experimental probes — analogous to how computational models can serve as experimental partners in biomedical research.
Deepfakes are everywhere. The godfather of digital forensics is fighting back
Science Published 2026-04-30 news article DOI:
deepfakes digital forensics AI safety computer vision
Summary: Profiles Hany Farid's career building tools to detect fake images and his ongoing battle against AI-generated deepfakes, highlighting the escalating arms race between generative AI and forensic detection.
Why it matters: The deepfake detection arms race is a microcosm of broader AI safety challenges — understanding detection limits informs how we think about AI reliability in scientific contexts.
Why for Yiru: AI reliability and detection of AI-generated content are increasingly relevant to scientific publishing and data integrity — a concern that extends to biomedical research.