Research Radar — 2026-04-28

Generated 2026-04-28 21:45 +0800 DeepSeek-V4-Pro Academic articles only

Methods & AI

Computational

3 selected
Computational #1 READ FULL

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

Authors: Jia Zhao et al.

spatial transcriptomics deep learning NMF cross-dataset integration

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.

Why for Yiru: Directly aligned with spatial transcriptomics methods, multi-sample integration, and interpretable representation learning for tissue organization.

Computational #2 READ FULL

Dango: Predicting higher-order genetic interactions

Cell Systems Published 2026-04-24 research article DOI:

Authors: Ruochi Zhang et al.

genetic interactions hypergraph neural networks self-attention genotype-to-phenotype

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.

Why for Yiru: Representation learning on biological graphs, perturbation modeling, and computational functional genomics — strong overlap with method interests.

Computational #3 READ FULL

Learning multi-cellular representations of single-cell transcriptomics data enables characterization of patient-level disease states

Cell Systems Published 2026-03-31 research article DOI:

Authors: Tianyu Liu et al.

single-cell foundation models patient-level phenotypes representation learning

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.

Why for Yiru: Bridges single-cell modeling, biomedical AI, and translational disease stratification — a key direction for computational immunology applications.

Biomedical discoveries

Biomedicine

5 selected
Biomedicine #1 READ FULL

Spatial 5mC-seq profiling of embryos and decidua after implantation in mammals

Nature Methods Published 2026-04-27 research article DOI: 10.1038/s41592-026-03079-w

Authors: Xun Shan et al.

spatial epigenomics DNA methylation single-cell resolution development

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.

Why for Yiru: High-impact spatial method that broadens the toolkit beyond RNA/protein readouts — relevant for multi-omics spatial analysis.

Biomedicine #2 READ FULL

Reference-free discovery with barcoded single-cell sequencing

Nature Biotechnology Published 2026-04-22 research article DOI: 10.1038/s41587-026-03084-6

Authors: Jacob M. Musser et al.

single-cell spatial transcriptomics reference-free bioinformatics

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.

Why for Yiru: Methodologically relevant: robust discovery from barcoded single-cell and spatial data without relying on predefined references.

Biomedicine #3 READ FULL

Sialylated CD43 forms a glyco-immune barrier that restrains antileukemic immunity

Science Published 2026-04-09 research article DOI: 10.1126/science.ady5196

Authors: Authors not specified in RSS

tumor immunology macrophage phagocytosis CRISPR screen glycosylation

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.

Why for Yiru: Strong match for tumor microenvironment, macrophage biology, and translational cancer immunotherapy — with a CRISPR screen methodology angle.

Biomedicine #4 READ FULL

A convergent uPAR-positive tumor ecosystem creates broad vulnerability to CAR T cell therapy

Cell Published 2026-03-30 research article DOI:

Authors: Zeda Zhang et al.

CAR T solid tumors tumor ecosystem stroma senescence

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.

Why for Yiru: Directly maps to CAR-T, tumor microenvironment, and translational cancer biology interests.

Biomedicine #5 READ FULL

Global genetic interaction network of a human cell maps conserved principles and informs functional interpretation of gene co-essentiality profiles

Cell Published 2026-04-27 research article DOI:

Authors: Maximilian Billmann et al.

genetic interactions CRISPR screen functional genomics cancer dependencies

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.

Why for Yiru: Systems-level reference for perturbation biology and computational interpretation of cancer dependencies — complements the computational methods in this digest.

Cross-disciplinary watchlist

Other Fields

3 selected
Field #1 SKIM

Today's robots walk, swim, fly, and manipulate objects on the go — and they're just getting started

Science Published 2026-04-16 review DOI: 10.1126/science.aeg9576

Authors: Science Robotics Editors

robotics AI control systems locomotion manipulation

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.

Why for Yiru: A clean AI + robotics breakthrough with no biomedical angle. Tracks how AI is transforming physical systems outside the life sciences.

Field #2 SKIM

AIs can 'memorize' data they shouldn't. Can they be forced to forget?

Science Published 2026-04-06 research news DOI:

Authors: Science News Staff

machine unlearning AI safety data privacy LLMs

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.

Why for Yiru: A fundamental AI safety research topic — relevant to anyone working with or thinking about foundation models, without being tied to biomedicine.

Field #3 SKIM

Higher education must bridge the AI gap

Science Published 2026-04-02 editorial DOI: 10.1126/science.aeh5777

Authors: Science Editorial

AI in education higher education AI literacy policy

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.

Why for Yiru: A cross-cutting AI policy and education perspective — relevant to academic career planning and the evolving role of AI in scientific training.