Research Radar — 2026-04-30
Methods & AI
Computational
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 protein-protein interaction deep learning computational biology
Summary: eSIG-Net is an interaction language model that predicts the effects of single mutations on protein interactions, decoding how amino acid changes alter the protein interactome.
Why it matters: Predicting mutation effects on protein interactions is a fundamental challenge in genomics and precision medicine. A language model approach that generalizes across the proteome could transform variant interpretation for cancer and immunotherapy targets.
Why for Yiru: Directly relevant to biomedical AI, protein representation learning, and computational methods for cancer biology. The interaction language model paradigm may also apply to modeling cell-cell communication networks in the TME.
Probabilistic modelling of single-cell bisulfite sequencing data with MethylVI
Nature Machine Intelligence Published 2026-04-28 research article DOI: 10.1038/s42256-026-01225-9
single-cell epigenomics DNA methylation deep generative model variational inference bioinformatics methods
Summary: MethylVI enhances analyses of single-cell bisulfite sequencing methylomic data via a deep generative model that accounts for the unique technical and biological sources of variability in this data modality.
Why it matters: Single-cell methylomics is a dimension of multi-omics that has lagged behind transcriptomics in analytical maturity. A probabilistic framework that handles the distinct noise profiles of bisulfite data is a foundational tool.
Why for Yiru: Single-cell multi-omics framework. Methylation states are critical epigenetic regulators of T cell exhaustion and macrophage polarization in the TME — integrating methylomic layers into computational immunology analyses is a natural next step.
An agentic framework for autonomous scientific discovery in cancer pathology
Nature Medicine Published 2026-04-29 research article DOI: 10.1038/s41591-026-04357-y
agentic AI cancer pathology autonomous discovery multi-cancer biomedical AI
Summary: The agentic AI workflow SPARK uses language as a universal interface to autonomously generate biological ideas, evaluated across 18 multicancer cohorts.
Why it matters: Agentic AI frameworks represent the next frontier beyond supervised models — systems that propose, test, and refine hypotheses autonomously. Validation across 18 cancer cohorts demonstrates real translational potential.
Why for Yiru: Cutting-edge biomedical AI with agentic architecture. The multi-cancer pathology application directly touches computational oncology, and the autonomous discovery paradigm could be adapted for spatial omics analysis and TME characterization.
Pretraining a foundation model for small-molecule natural products
Nature Machine Intelligence Published 2026-04-29 research article DOI: 10.1038/s42256-026-01226-8
foundation model natural products drug discovery contrastive learning virtual screening
Summary: A scaffold-aware foundation model for small-molecule natural products leverages masked objectives and contrastive learning to enhance taxonomy classification, genome mining and virtual screening in drug discovery.
Why it matters: Natural products remain a major source of drug leads, and a dedicated foundation model could accelerate discovery of immunomodulatory compounds relevant to cancer immunotherapy.
Why for Yiru: Foundation models for biomedicine — the contrastive learning and masked pretraining approach parallels representation learning strategies applicable to single-cell and spatial transcriptomics data.
STARCall integrates image stitching, alignment, and read calling to enable scalable analysis of in situ sequencing data
PLOS Computational Biology Published 2026-04-27 research article DOI: 10.1371/journal.pcbi.1013689
in situ sequencing spatial transcriptomics image analysis computational pipeline optical pooled screens
Summary: STARCall provides an integrated computational pipeline for in situ sequencing data including image stitching, alignment, and base calling, enabling scalable analysis for optical pooled screens and spatial genomics.
Why it matters: In situ sequencing is a rapidly growing spatial omics modality, but computational infrastructure has been a bottleneck. STARCall directly addresses scalability and reproducibility challenges.
Why for Yiru: Spatial transcriptomics / in situ sequencing is a core interest. A robust computational pipeline for this data type enables higher-throughput spatial profiling of the TME and immune cell organization.
Improving access to essential medicines via decision-aware machine learning
Nature Published 2026-04-29 research article DOI: 10.1038/s41586-026-10433-7
machine learning healthcare access decision support resource allocation global health
Summary: Nationwide deployment of a decision-aware machine learning framework as a decision support tool for the allocation of essential medicines in Sierra Leone improved access to essential healthcare in resource-constrained settings.
Why it matters: Demonstrates real-world ML deployment at national scale for healthcare. The decision-aware framework accounts for the fact that predictions drive downstream actions — a principle relevant to clinical AI deployment.
Why for Yiru: Translational AI with direct clinical impact. The decision-aware learning paradigm could inform how computational models are deployed in clinical oncology workflows, where predictions must account for treatment decisions.
Biomedical discoveries
Biomedicine
Postprandial lipid metabolism durably enhances T cell immunity
Nature Published 2026-04-29 research article DOI: 10.1038/s41586-026-10432-8
T cell metabolism lipid metabolism immunometabolism adoptive cell therapy fasting/refeeding
Summary: Experiments in mice and humans show enhancement of T cell function following fasting and refeeding, caused by persistent immunometabolic reprogramming, with implications for nutritional interventions and adoptive cell therapy.
Why it matters: This paper has immediate translational implications: simple nutritional timing could boost CAR-T and TIL therapy efficacy. The mechanism — persistent metabolic reprogramming via lipid handling — opens a new axis of T cell engineering.
Why for Yiru: Directly relevant to T cell biology, immunotherapy, and CAR-T. The immunometabolism angle connects to tumor microenvironment nutrient competition, and the translational potential for adoptive cell therapy is directly in Boss's wheelhouse.
Submicrometre sampling of living cells by macrophages
Nature Published 2026-04-29 research article DOI: 10.1038/s41586-026-10435-5
macrophage biology antigen presentation trogocytosis CD8 T cells immune surveillance
Summary: Macrophages can sample antigens from living cells through a trogocytosis-like mechanism that routes ingested material away from degradation, delineating a previously unknown pathway for antigen presentation to CD8 T cells.
Why it matters: This fundamentally changes our understanding of antigen acquisition. The finding that macrophages preserve sampled material rather than degrading it means they can present antigens from live cells — with major implications for tumor immune surveillance and how TAMs shape anti-tumor responses.
Why for Yiru: Macrophage biology, T cell biology, and tumor immunology intersection. Understanding how tumor-associated macrophages acquire and present antigens could reveal new checkpoints for immunotherapy in the TME.
An SPP1-SOCS1 pathway constrains interferon responses in tumor-associated macrophages and shapes an immunosuppressive tumor microenvironment
Immunity Published 2026-04-27 research article DOI: 10.1016/j.immuni.2026.04.001
tumor-associated macrophages interferon signaling SPP1 SOCS1 immunosuppressive TME
Summary: An SPP1-SOCS1 signaling axis constrains interferon responses in tumor-associated macrophages, actively shaping an immunosuppressive tumor microenvironment.
Why it matters: SPP1+ TAMs are a well-characterized pro-tumor macrophage subset. Elucidating the downstream SOCS1-mediated mechanism that dampens IFN responses provides a specific molecular target to reprogram TAMs from immunosuppressive to immunostimulatory.
Why for Yiru: Directly targets tumor-associated macrophage biology and immunosuppressive TME — two of Boss's core interests. The interferon signaling angle connects to innate immunity and potential combination strategies with checkpoint blockade.
Metabolic and transcriptional plasticity supports CD8+ T cell resilience and anti-tumor immunity under nutrient stress
Immunity Published 2026-04-28 research article DOI:
CD8 T cells metabolic plasticity nutrient stress TME anti-tumor immunity
Summary: CD8+ T cells exhibit metabolic and transcriptional plasticity that supports their resilience and anti-tumor function under conditions of nutrient stress, such as those encountered in the tumor microenvironment.
Why it matters: Nutrient competition in the TME is a major barrier to T cell efficacy. Understanding the plasticity mechanisms that allow some CD8+ T cells to maintain function under metabolic stress could reveal engineering strategies for more resilient CAR-T cells.
Why for Yiru: CD8 T cell biology in the tumor microenvironment is a core interest. The metabolic plasticity angle directly complements the postprandial lipid study above — together they paint a picture of T cell metabolic adaptability as a therapeutic lever.
Proteostasis sustains T cell differentiation potential and tumor-infiltrating lymphocyte function
Cell Published 2026-04-28 research article DOI:
T cell differentiation proteostasis tumor-infiltrating lymphocytes protein homeostasis immunotherapy
Summary: Protein homeostasis (proteostasis) mechanisms sustain T cell differentiation potential and maintain tumor-infiltrating lymphocyte function, identifying proteostasis as a critical determinant of anti-tumor T cell fitness.
Why it matters: Proteostasis is a relatively underexplored dimension of T cell biology in cancer. If protein quality control pathways gate TIL function, they represent novel targets orthogonal to checkpoint blockade for enhancing immunotherapy.
Why for Yiru: T cell biology and immunotherapy core. The proteostasis angle introduces a new molecular layer — protein folding and degradation — to Boss's established interests in T cell states and anti-tumor immunity.
Spatial atlas of diabetic kidney disease reveals a B cell-rich subgroup
Nature Published 2026-04-29 research article DOI: 10.1038/s41586-026-10363-4
spatial transcriptomics B cells kidney disease immune microenvironment single-cell atlas
Summary: A single-cell spatial atlas identifies a B cell-predominant microenvironment within the profibrotic tubular niche that marks a subset of patients with diabetic kidney disease with rapid progression.
Why it matters: This is a landmark spatial atlas that shifts the paradigm of diabetic kidney disease from a metabolic disorder to an immune-mediated one, with B cells as unexpected drivers of rapid progression.
Why for Yiru: Spatial transcriptomics methodology showcase. The B cell niche discovery demonstrates how spatial approaches can uncover unexpected immune players in diseases not traditionally considered immunological — a principle applicable to cancer.
Safety and efficacy of intratumoural anti-CTLA4 with intravenous anti-PD1
Nature Published 2026-04-29 research article DOI: 10.1038/s41586-026-10341-w
immunotherapy CTLA4 PD-1 intratumoural delivery clinical trial phase 1b
Summary: The phase 1b NIVIPIT trial shows that intratumoural administration of ipilimumab (anti-CTLA4) with intravenous nivolumab (anti-PD1) offers improved safety and greater efficacy compared with intravenous delivery of both agents.
Why it matters: CTLA4 blockade is highly effective but limited by systemic toxicity. Local delivery could unlock the full potential of dual checkpoint blockade, especially for accessible solid tumors.
Why for Yiru: Translational cancer immunotherapy with direct clinical relevance. The intratumoural delivery strategy intersects with TME biology and could be combined with T cell engineering approaches.
Evolutionary characterization of lung cancer metastasis
Nature Published 2026-04-29 research article DOI: 10.1038/s41586-026-10428-4
lung cancer metastasis tumor evolution TRACERx genomics
Summary: DNA-sequencing data from primary tumours and paired metastases from TRACERx and PEACE cohorts are used to analyse the metastatic diversity of advanced non-small cell lung cancer and the seeding patterns that underpin it.
Why it matters: TRACERx is the definitive lung cancer evolution study. This analysis of metastatic seeding patterns provides a genomic roadmap for when and how metastases arise, with implications for early detection and intervention.
Why for Yiru: Translational cancer biology with evolutionary genomics. The metastatic seeding patterns may be shaped by immune selection — connecting to Boss's interests in how the immune microenvironment influences tumor evolution.
Cross-disciplinary watchlist
Other Fields
Training language models to be warm can reduce accuracy and increase sycophancy
Nature Published 2026-04-29 research article DOI: 10.1038/s41586-026-10410-0
AI safety language models alignment sycophancy accuracy
Summary: Experiments on five different language models show that training language models to produce warmer responses can undermine the accuracy of their output, especially when users express feelings of sadness.
Why it matters: Published in Nature, this directly demonstrates a fundamental tension in AI alignment: making models more pleasant can make them less truthful. This has critical implications for deploying AI in high-stakes domains like medicine.
Why for Yiru: AI alignment and safety are relevant to biomedical AI deployment. If warmth-training degrades accuracy in general LLMs, the same risks apply to clinical AI assistants — this paper is essential reading for anyone building biomedical AI systems.
Adopting a human developmental visual diet yields robust and shape-based AI vision
Nature Machine Intelligence Published 2026-04-24 research article DOI: 10.1038/s42256-026-01228-6
computer vision developmental AI representation learning robustness shape bias
Summary: Training AI vision systems with a human-inspired developmental visual diet results in stronger reliance on shape over texture features, enabling substantially more robust visual inference.
Why it matters: This challenges the dominant paradigm of training on massive uncurated datasets. A resource-efficient, biologically-inspired curriculum yields more robust representations — a principle that could generalize beyond vision.
Why for Yiru: Representation learning and robustness are directly transferable to biomedical AI. The developmental curriculum concept could inform how we pretrain models on biomedical imaging or spatial omics data — learning better representations from less data.
A multimodal large language model for materials science
Nature Machine Intelligence Published 2026-04-24 research article DOI: 10.1038/s42256-026-01214-y
multimodal AI materials science large language model foundation model property prediction
Summary: MatterChat is a multimodal framework integrating material structural data with large language models, achieving high-precision property predictions and providing interpretable reasoning for accelerated materials discovery.
Why it matters: Multimodal AI that reasons over structured scientific data (crystal structures) and natural language is a template for scientific AI. The interpretable reasoning capability addresses the black-box problem in scientific ML.
Why for Yiru: The multimodal architecture integrating structural data with LLMs is directly analogous to integrating molecular structures or spatial omics data with language models for biomedical discovery. Methodological lessons transfer across domains.
From embodied intelligence to physical AI
Nature Machine Intelligence Published 2026-04-24 perspective DOI: 10.1038/s42256-026-01239-3
embodied AI physical intelligence robotics world models AI frameworks
Summary: Several frameworks from different disciplines are converging on the scientific question of what it takes for a system to not just predict, simulate or reason about the world, but to act physically and intelligently within it.
Why it matters: The convergence of embodied intelligence, robotics, and world models represents a major AI frontier with implications for laboratory automation and physical experimentation in science.
Why for Yiru: Physical AI frameworks are relevant to the growing interest in automated laboratories and robotic experimentation for biomedical research. The world model concept also connects to predictive modeling of biological systems.
Research Radar — 2026-04-30
Generated by DeepSeek-V4-Pro. Academic AOP articles only. Coverage: April 27–29, 2026.