Research Radar — 2026-05-14
Methods & AI
Computational
D-SPIN constructs regulatory network models from scRNA-seq that reveal organizing principles of perturbation response
Cell Published 2026-05-12 research article DOI:
gene regulatory networks single-cell RNA-seq perturbation modeling network inference
Summary: D-SPIN is a computational framework for constructing mechanistically interpretable, generative models of cellular regulatory networks from single-cell mRNA-seq data. The framework models how perturbations change cell states by reconfiguring the underlying regulatory interactions, revealing global organization and key regulators of perturbation responses, as well as mechanisms of combinatorial drug responses.
Why it matters: Bridges scRNA-seq data with causal regulatory network models, enabling mechanistic understanding of perturbation responses beyond differential expression.
Why for Yiru: Generative network models from single-cell data directly relevant to spatial omics interpretation and perturbation-response mapping in tumor microenvironments.
The illusion of interpretability in biologically informed neural networks
bioRxiv Published 2026-05-07 preprint DOI: 10.64898/2026.05.07.723544
interpretable ML neural networks identifiability bioinformatics
Summary: Demonstrates that biologically informed neural networks (BINNs), despite architectures mirroring gene-to-pathway relationships, suffer from fundamental nonidentifiability: even under ideal conditions the model recovers input-output mapping without identifying true internal gene-to-pathway weights or pathway activations. This failure persists across classification, regression, and survival tasks.
Why it matters: Critical cautionary finding for the growing field of interpretable AI in biology — architectural transparency does not guarantee mechanistic interpretability.
Why for Yiru: Directly relevant to any work using visible neural networks or pathway-structured models for single-cell and spatial omics data.
Mechanisms Matter: Transportability of Cellular Perturbation Effects
bioRxiv Published 2026-05-08 preprint DOI: 10.64898/2026.05.08.723625
causal inference perturbation prediction deep learning transportability
Summary: Uses causal transportability theory to show cross-context generalization of perturbation effects is governed by shared causal mechanisms, not distributional similarity. Extensive benchmarking reveals deep learning models fail to generalize across contexts with different causal mechanisms, often performing no better than simple baselines.
Why it matters: Establishes that current perturbation prediction models lack mechanistic grounding for cross-context generalization — fundamental limitation for drug response prediction.
Why for Yiru: Causal framework for understanding when and why perturbation models fail to transfer, essential for designing robust computational methods in cancer biology.
StabCell: Stability selection for clustering and marker detection in single-cell RNA sequencing
bioRxiv Published 2026-05-07 preprint DOI: 10.64898/2026.05.07.720061
single-cell RNA-seq differential expression stability selection clustering
Summary: Introduces StabCell, a stability selection framework integrating clustering and marker gene detection via repeated subsampling. Provides approximate empirical per-family error rate control, selecting fewer false positive markers than conventional pseudobulk approaches, especially under low signal-to-noise conditions. Validated on iPSC-to-cardiomyocyte differentiation data.
Why it matters: Addresses the circularity problem in scRNA-seq analysis where the same data is used for both clustering and DE testing, producing overconfident results.
Why for Yiru: Stability-based marker detection offers more robust cell-type characterization for spatial transcriptomics and TME cell state annotation.
Temporal-deviation-driven community detection uncovers early-warning signals for critical transitions in complex diseases
bioRxiv Published 2026-05-08 preprint DOI: 10.64898/2026.05.08.723925
early-warning signals dynamical systems community detection disease progression
Summary: Presents TD-COM, a framework detecting personalized early-warning signals of critical disease transitions from single-snapshot data. Constructs temporal perturbation maps capturing latent dynamics from static profiles, validated on hour-scale to multi-decade transcriptomic data, outperforming existing methods in accuracy and robustness.
Why it matters: Enables early detection of impending clinical deterioration from a single time point, critical when longitudinal sampling is infeasible.
Why for Yiru: Dynamical systems approach to disease progression complements spatial omics analysis of tumor evolution and treatment response timing.
Dogcatcher2: Improved statistical detection of transcriptional readthrough and repetitive element analysis across sequencing platforms
bioRxiv Published 2026-05-07 preprint DOI: 10.64898/2026.05.07.723642
transcriptional readthrough repetitive elements bioinformatics RNA-seq
Summary: Dogcatcher2 improves detection of downstream-of-gene (DoG) transcription using statistical methods on gene body-normalized coverage. Outperforms existing tools across platforms including pseudobulk scRNA-seq, and reveals enrichment of inverted Alu pairs in DoG regions connecting readthrough to dsRNA generation and innate immune signaling.
Why it matters: Connects transcriptional dysregulation to innate immune activation via double-stranded RNA, with implications for cancer and neurodegeneration.
Why for Yiru: Transcriptional readthrough as a source of immunogenic dsRNA is relevant to understanding tumor immunogenicity and immunotherapy response mechanisms.
Biomedical discoveries
Biomedicine
Tumor-associated tissue-resident macrophages drive pancreatic cancer progression through IGF1-IGF1R signaling
bioRxiv Published 2026-05-08 preprint DOI: 10.64898/2026.05.08.723816
tumor-associated macrophages pancreatic cancer immunotherapy tumor microenvironment
Summary: Identifies tissue-resident macrophage-derived TAMs (TRM-TAMs) as independent poor prognostic indicators in PDAC. Using an iPSC-derived macrophage organoid co-culture platform, demonstrates TRM-TAMs drive cancer cell proliferation and chemoresistance via IGF1 signaling, providing rationale for why unselected IGF1R inhibitor trials failed and suggesting biomarker-stratified revival.
Why it matters: Resolves the paradox of IGF1R inhibitor failure in PDAC by identifying TRM-TAM abundance as a stratification biomarker, potentially reviving a targeted therapy approach.
Why for Yiru: Tissue-resident macrophage biology in the TME, organoid-immune co-culture models, and macrophage-mediated chemoresistance all intersect with core interests in tumor immunology.
Targeting an RNA Editor to Impede H3K27M+ Pediatric Gliomas
bioRxiv Published 2026-05-08 preprint DOI: 10.64898/2026.05.08.723800
glioma ADAR RNA editing immunotherapy dsRNA sensing
Summary: Identifies ADAR, an RNA-editing enzyme suppressing endogenous dsRNA sensing, as overexpressed in H3K27M-mutant diffuse midline glioma. ADAR depletion or pharmacological degradation via ATRA activates type I interferon signaling, increases tumor immunogenicity, enhances CD8+ T cell infiltration, and synergizes with immune checkpoint blockade and radiation in orthotopic models.
Why it matters: Pharmacological ADAR degradation via clinically available ATRA represents a readily translatable strategy for sensitizing immunologically cold pediatric gliomas to immunotherapy.
Why for Yiru: RNA-editing-mediated immune evasion, tumor immunogenicity reprogramming, and immunotherapy combination strategies directly relevant to cancer immunology interests.
Tumor Protein D54 (TPD54) regulates intracellular protein trafficking, cellular function and disease progression in melanoma
bioRxiv Published 2026-05-07 preprint DOI: 10.64898/2026.05.07.721771
melanoma protein trafficking immune evasion tumor microenvironment
Summary: Establishes TPD54 as a central regulator of intracellular protein transport exploited by melanoma cells. TPD54 maintains Golgi integrity, orchestrates vesicular trafficking, augments pro-cancerous cytokine secretion, and increases surface expression of adhesion receptors. Targeting TPD54 in mouse models attenuated tumor growth, disrupted tumor vasculature, enhanced CD8+ T cell infiltration, and reduced metastasis.
Why it matters: Reveals protein trafficking as a druggable vulnerability in melanoma that simultaneously impairs multiple pro-tumorigenic pathways including immune evasion.
Why for Yiru: Protein trafficking control of the cancer cell surface landscape and secretome represents an underexplored axis of immune evasion with translational potential.
Integrative Genomic, Single-Cell, and Functional Profiling of the CD48-CD244 Axis and NK-Cell Dysfunction in Multiple Myeloma
bioRxiv Published 2026-05-08 preprint DOI: 10.64898/2026.05.08.723909
multiple myeloma NK cells immunotherapy immune evasion
Summary: Integrates multi-omics, CRISPR screens, and machine learning to dissect CD48-CD244 axis in multiple myeloma. CD48 overexpression initially enhances NK cytotoxicity but chronic exposure drives NK exhaustion. In vivo, CD48-overexpressing tumors progress more slowly, with NK cell depletion accelerating disease, revealing a context-dependent role for this axis.
Why it matters: Resolves the paradoxical association of CD48 with both high-risk disease and NK activation, suggesting therapeutic windows for modulating this axis.
Why for Yiru: NK cell dysfunction in the bone marrow TME, multi-omics integration, and context-dependent immune receptor biology all align with core interests.
Engineering a pacemaker-driven human mini-heart guided by spatial multi-omics of sinoatrial node development
bioRxiv Published 2026-05-07 preprint DOI: 10.64898/2026.05.07.723626
spatial multi-omics organoids cardiac biology AI-guided modeling
Summary: Integrates spatial transcriptomics and single-nucleus multiomics of human fetal sinoatrial node with stem cell engineering to generate pacemaker organoids and assemble a pacemaker-driven human mini-heart. AI-guided perturbation modeling identifies conserved regulatory pathways including YAP-TEAD and NRG-ERBB signaling controlling pacemaker specification.
Why it matters: Spatial multi-omics-guided tissue engineering represents a paradigm for reconstructing complex human tissue architecture and function in vitro.
Why for Yiru: Spatial multi-omics integration with AI-guided perturbation modeling for tissue engineering — methodological convergence of core technical interests.
The gut microbiota metabolite Urolithin A mitigates JAK signaling to suppress cytokine-mediated autoimmune diseases
bioRxiv Published 2026-05-08 preprint DOI: 10.64898/2026.05.08.723914
autoimmunity JAK signaling microbiome metabolite
Summary: Characterizes Urolithin A, a gut-derived metabolite, as a direct JAK1 inhibitor that broadly dampens cytokine signaling (IFN-I, IFN-II, IL-6). UA binds the JAK1 JH1 domain and attenuates autoimmune pathogenesis in Trex1-KO mice and lupus/psoriasis models.
Why it matters: Natural metabolite-based JAK inhibition offers a potential alternative to synthetic JAK inhibitors with distinct safety profiles for autoimmune disease.
Why for Yiru: Metabolite-mediated immune regulation and JAK-STAT signaling intersect with interests in immunomodulation, though less directly cancer-focused.
Cross-disciplinary watchlist
Other Fields
Peripheral control enabled by distributed sensing in an octopus-inspired soft robotic arm for autonomous underwater grasping
Nature Machine Intelligence Published 2026-05-12 research article DOI: 10.1038/s42256-026-01230-y
soft robotics distributed sensing embodied intelligence bio-inspired design
Summary: Presents an octopus-inspired soft robotic arm using optoelectronic mechanosensors in suction cups to detect contact forces and infer object positions, enabling autonomous underwater grasping through peripheral distributed control without centralized processing.
Why it matters: Demonstrates how distributed sensing and peripheral control can achieve autonomous behavior in unstructured environments, relevant to embodied AI and soft robotics.
Why for Yiru: Bio-inspired distributed intelligence and sensorimotor integration at the physical layer — conceptual parallels to decentralized biological computation.
Chemistry in the AI era
Nature Published 2026-05-12 editorial DOI: 10.1038/d41586-026-01521-9
AI in chemistry scientific AI materials discovery drug design
Summary: Nature editorial examining how artificial intelligence is transforming chemistry research, from reaction prediction and materials discovery to drug design and automated laboratories.
Why it matters: Captures the breadth of AI's impact across chemistry subfields, with implications for how computational methods are reshaping experimental sciences.
Why for Yiru: AI-driven molecular design and automated experimentation paradigms may inform computational approaches in biomedical research.