https://CHENyiru3.github.io/research-radar/ Daily ranked academic article recommendations from Research Radar. en Tue, 28 Apr 2026 14:03:48 +0000 https://www.nature.com/articles/s41588-026-02579-x 2026-04-28-comp-1-10.1038/s41588-026-02579-x Tue, 28 Apr 2026 13:45:00 +0000 Section: Computational | Source: Nature Genetics | Published: 2026-04-27 | DOI: 10.1038/s41588-026-02579-x | Recommendation: READ FULL | Summary: INSPIRE combines deep learning with non-negative matrix factorization to integrate diverse spatial transcriptomics datasets, recovering shared and context-specific spatial gene programs and tissue organization across scales. | Why it matters: Cross-platform spatial integration is a central bottleneck. INSPIRE's interpretable factorization is a strong alternative to black-box embeddings when biological programs must be compared across cohorts. spatial transcriptomics deep learning NMF cross-dataset integration https://www.cell.com/cell-systems/fulltext/S2405-4712(26)00075-X?rss=yes 2026-04-28-comp-2-https://www.cell.com/cell-systems/fulltext/S2405-4712(26)00075-X?rss=yes Tue, 28 Apr 2026 13:45:00 +0000 Section: Computational | Source: Cell Systems | Published: 2026-04-24 | DOI: | Recommendation: READ FULL | Summary: DANGO uses a self-attention hypergraph neural network integrating multimodal molecular networks and protein sequence embeddings to predict higher-order genetic interactions, expanding the yeast trigenic interaction landscape to over 400 million interactions. | Why it matters: Higher-order interactions are combinatorially intractable to measure; predictive graph models make perturbation landscapes computationally navigable. genetic interactions hypergraph neural networks self-attention genotype-to-phenotype https://www.cell.com/cell-systems/fulltext/S2405-4712(26)00052-9?rss=yes 2026-04-28-comp-3-https://www.cell.com/cell-systems/fulltext/S2405-4712(26)00052-9?rss=yes Tue, 28 Apr 2026 13:45:00 +0000 Section: Computational | Source: Cell Systems | Published: 2026-03-31 | DOI: | Recommendation: READ FULL | Summary: PaSCient learns sample-level representations from single-cell RNA-seq across disease and tissue contexts, highlighting cells and genes that predict patient-level phenotypes such as disease severity and drug response. | Why it matters: Single-cell foundation models operate at cell level, but clinical translation requires patient-level representations. This directly addresses that scale mismatch. single-cell foundation models patient-level phenotypes representation learning https://www.nature.com/articles/s41592-026-03079-w 2026-04-28-biomed-1-10.1038/s41592-026-03079-w Tue, 28 Apr 2026 13:45:00 +0000 Section: Biomedicine | Source: Nature Methods | Published: 2026-04-27 | DOI: 10.1038/s41592-026-03079-w | Recommendation: READ FULL | Summary: Spatial 5mC-seq enables unbiased spatiotemporal genome-wide methylome profiling at single-cell resolution in mammalian embryos and decidua. | Why it matters: Expands spatial omics beyond transcript measurements into the methylome, opening a new axis for linking cell state with spatial context. spatial epigenomics DNA methylation single-cell resolution development https://www.nature.com/articles/s41587-026-03084-6 2026-04-28-biomed-2-10.1038/s41587-026-03084-6 Tue, 28 Apr 2026 13:45:00 +0000 Section: Biomedicine | Source: Nature Biotechnology | Published: 2026-04-22 | DOI: 10.1038/s41587-026-03084-6 | Recommendation: READ FULL | Summary: sc-SPLASH extends reference-free analysis to barcoded single-cell and spatial transcriptomics data, reducing dependence on fixed annotations and prior marker sets. | Why it matters: Reference-free discovery is essential when annotations are incomplete or biased — a recurring limitation in exploratory single-cell and spatial analyses. single-cell spatial transcriptomics reference-free bioinformatics https://www.science.org/doi/10.1126/science.ady5196 2026-04-28-biomed-3-10.1126/science.ady5196 Tue, 28 Apr 2026 13:45:00 +0000 Section: Biomedicine | Source: Science | Published: 2026-04-09 | DOI: 10.1126/science.ady5196 | Recommendation: READ FULL | Summary: CRISPR knockout screens in human AML cells identify sialylated CD43 as a glyco-immune barrier that restrains macrophage-mediated antileukemic immunity. | Why it matters: Macrophage-directed immunotherapy has struggled clinically; a specific tumor glycosylation barrier provides a concrete mechanism and intervention axis. tumor immunology macrophage phagocytosis CRISPR screen glycosylation https://www.cell.com/cell/fulltext/S0092-8674(26)00269-2?rss=yes 2026-04-28-biomed-4-https://www.cell.com/cell/fulltext/S0092-8674(26)00269-2?rss=yes Tue, 28 Apr 2026 13:45:00 +0000 Section: Biomedicine | Source: Cell | Published: 2026-03-30 | DOI: | Recommendation: READ FULL | Summary: uPAR CAR T cells synergize with senescence-inducing therapies to eradicate primary and metastatic solid tumors by targeting both tumor cells and supportive stroma, with minimal myelodepletion. | Why it matters: Solid-tumor CAR T needs targets that capture tumor and stromal compartments; a convergent ecosystem vulnerability suggests broader efficacy than tumor-cell-only antigens. CAR T solid tumors tumor ecosystem stroma senescence https://www.cell.com/cell/fulltext/S0092-8674(26)00345-4?rss=yes 2026-04-28-biomed-5-https://www.cell.com/cell/fulltext/S0092-8674(26)00345-4?rss=yes Tue, 28 Apr 2026 13:45:00 +0000 Section: Biomedicine | Source: Cell | Published: 2026-04-27 | DOI: | Recommendation: READ FULL | Summary: CRISPR perturbation of ~4 million gene pairs in human HAP1 cells maps ~89,000 genetic interactions, revealing a hierarchical network linking genes to complexes, pathways, and cellular processes while elucidating cancer cell dependencies. | Why it matters: A human-scale genetic interaction map is a foundational resource for interpreting co-essentiality, pathway structure, and synthetic vulnerability. genetic interactions CRISPR screen functional genomics cancer dependencies https://www.science.org/doi/10.1126/science.aeg9576 2026-04-28-field-1-10.1126/science.aeg9576 Tue, 28 Apr 2026 13:45:00 +0000 Section: Other Fields | Source: Science | Published: 2026-04-16 | DOI: 10.1126/science.aeg9576 | Recommendation: SKIM | Summary: Recent advances in AI-driven control systems have enabled robots to transition between rolling, walking, flying, and swimming — and to use legs as manipulators while moving. This review covers the control challenges overcome to achieve multi-modal locomotion. | Why it matters: Multi-modal robots represent a convergence of AI, control theory, and mechanical engineering — a milestone toward general-purpose embodied intelligence. robotics AI control systems locomotion manipulation https://www.science.org/content/article/ais-can-memorize-data-they-shouldn-t-can-they-be-forced-forget 2026-04-28-field-2-https://www.science.org/content/article/ais-can-memorize-data-they-shouldn-t-can-they-be-forced-forget Tue, 28 Apr 2026 13:45:00 +0000 Section: Other Fields | Source: Science | Published: 2026-04-06 | DOI: | Recommendation: SKIM | Summary: A new tool helps researchers probe how large language models 'unlearn' sensitive training material, addressing a core tension between model capability and data privacy. | Why it matters: Machine unlearning is becoming a practical requirement as models are trained on increasingly broad data. This work addresses a key AI safety and privacy challenge. machine unlearning AI safety data privacy LLMs https://www.science.org/doi/10.1126/science.aeh5777 2026-04-28-field-3-10.1126/science.aeh5777 Tue, 28 Apr 2026 13:45:00 +0000 Section: Other Fields | Source: Science | Published: 2026-04-02 | DOI: 10.1126/science.aeh5777 | Recommendation: SKIM | Summary: As AI advances at a pace that outstrips regulatory and ethical frameworks, this editorial argues that higher education must develop students who can critically engage with AI systems rather than merely use them. | Why it matters: The AI literacy gap in higher education affects how the next generation of scientists and engineers will interact with AI — a structural challenge that shapes all fields. AI in education higher education AI literacy policy