Research Radar — 2026-08-29
Methods & AI
Computational
A pretrained unified model enables cellular functional profile prediction and multi-objective virtual drug screening
bioRxiv (Bioinformatics) Published 2026-08-28 Preprint DOI:
biomedical AI spatial omics drug discovery perturbation biology
Summary: InsilicoCell is a pretrained multimodal, multitask model trained on more than 88 million measurements across seven tasks, linking molecular profiles to cellular phenotypes and perturbation responses. It extends to patient, spatial, and single-cell settings and identifies candidate compounds with experimental validation.
Why it matters: The unified representation connects cellular state modeling to multi-objective virtual screening and tests transfer across entities, contexts, and conditions rather than optimizing a single task.
Why for Yiru: It is unusually close to the intersection of cellular state modeling, spatial omics, biomedical AI, and drug discovery, with a useful bridge from prediction to experimentally tested intervention hypotheses.
PROFET predicts continuous gene expression dynamics from scRNA-seq data to elucidate heterogeneity of cancer treatment responses
Cell Systems Published 2026-08-28 Research article DOI:
single-cell omics oncology disease dynamics computational immunology
Summary: PROFET reconstructs continuous single-cell trajectories from sparse scRNA-seq time series by combining particle-based gradient flows with neural force matching. The framework captures nonlinear cell-state dynamics and reveals heterogeneous treatment responses and candidate surface markers associated with breast cancer resistance.
Why it matters: A trajectory-first dynamical model addresses a central limitation of sparse single-cell time series and links inferred state changes to heterogeneous treatment response and resistance-associated markers.
Why for Yiru: The approach is directly transferable to immune and tumor-state modeling, where continuous disease dynamics may be more informative than disconnected snapshots.
CytoGate-Bench: an LLM benchmark for cross-panel cell gating in cytometry
bioRxiv (Bioinformatics) Published 2026-08-26 Preprint DOI:
computational immunology biomedical AI cytometry evaluation generalization
Summary: CytoGate-Bench evaluates LLMs on hierarchical, cross-panel cell gating, emphasizing panel-agnostic transfer and distribution shift instead of only in-panel accuracy. The benchmark uses expert-curated cytometry data and explicit ablations to test where model performance comes from.
Why it matters: It turns a clinically relevant immune-phenotyping workflow into a generalization benchmark that makes cross-panel reliability and expert-defined evaluation criteria explicit.
Why for Yiru: The benchmark offers a concrete template for evaluating biomedical agents and computational-immunology systems under realistic panel changes rather than relying on narrow held-out accuracy.
Label Noise Limits TCR-pMHC Specificity Prediction: Improved Performance Through AlphaFold3-Based Structural Modeling and Data Denoising
bioRxiv (Bioinformatics) Published 2026-08-28 Preprint DOI:
computational immunology protein modeling immunotherapy data quality
Summary: This study uses an AlphaFold3-based TCR-pMHC structural-modeling pipeline and cluster-based denoising to improve specificity prediction. It reports state-of-the-art performance and more than 70% relative improvement in binder-ranking accuracy after removing mislabeled points from a large specificity dataset.
Why it matters: The work identifies label noise as a major ceiling on TCR specificity prediction and shows that data quality and structural modeling can matter as much as architecture.
Why for Yiru: Its diagnosis of noisy immune-receptor labels is methodologically transferable to computational-immunology and multimodal biomedical datasets used for target selection.
Biomedical discoveries
Biomedicine
Spatially resolved transcriptional programs link fallopian tube precursor lesions to immune activation and stromal reorganization
bioRxiv (Cancer Biology) Published 2026-08-28 Preprint DOI:
spatial omics oncology tumor microenvironment early detection
Summary: Using tissue-wide, single-cell-resolution Visium HD spatial transcriptomics across normal tissue, precursor lesions, and invasive cancer, the study maps immune and stromal programs, collagen architecture, and therapeutic antigen candidates during fallopian-tube tumor development.
Why it matters: The study uses spatial context to connect early cancer precursor states with immune activation and stromal reorganization, offering a route toward interception biology rather than focusing only on established tumors.
Why for Yiru: It is a strong example of spatial omics clarifying early tumor-microenvironment transitions and generating hypotheses for immune or antigen-directed intervention.
Spatial multi-omics analysis reveals vimentin-high macrophages-endothelial cells niche shapes CAFs heterogeneity in colorectal cancer metastasis
bioRxiv (Cancer Biology) Published 2026-08-27 Preprint DOI:
spatial omics computational immunology oncology metastasis
Summary: High-plex spatial multi-omic mapping and neighborhood analysis of colorectal cancer primary tumors and paired liver metastases identifies a vimentin-high macrophage-endothelial niche associated with distinct cancer-associated fibroblast phenotypes and site-specific signaling.
Why it matters: The primary-versus-metastatic comparison uses spatial multi-omics to propose a mechanistic niche that shapes CAF heterogeneity, moving beyond a descriptive map of cell types.
Why for Yiru: It is a concrete fit for spatial immunology and metastasis modeling, especially approaches that connect tissue neighborhoods to tumor-stroma state transitions.
Genome-scale perturb-seq in primary human CD4+ T cells maps context-specific regulators of T cell programs and human immune traits
Cell Published 2026-08-28 Research article DOI:
computational immunology perturb-seq immune genetics T cells
Summary: A dynamic regulatory atlas was generated by perturbing every expressed gene across 22 million primary human CD4+ T cells under resting and re-stimulated conditions. The map reveals context-specific immune regulators and links pathways to naturally occurring T-cell states and autoimmune disease risk.
Why it matters: Genome-scale perturbation across resting and stimulated contexts directly tests how regulatory effects depend on cellular state and connects functional screens to human immune traits.
Why for Yiru: It offers a valuable bridge between perturbational functional genomics, immune-state modeling, and disease genetics for computational immunology.
Cross-disciplinary watchlist
Other Fields
A multimodal, all-optical platform for linking cell identity to metabolic function in intact tissues
Nature Methods Published 2026-08-27 Research article DOI:
spatial omics metabolic imaging oncology functional histopathology
Summary: REDCAT integrates Raman imaging, autofluorescence, and high-plex immunofluorescence to map metabolic activity alongside cell identity in intact tissue at single-cell and subcellular resolution. Applications include lipid-redox remodeling in lymphoma and intratumoral heterogeneity.
Why it matters: The platform moves spatial profiling beyond static expression by measuring cellular identity and metabolic function together in preserved tissue architecture.
Why for Yiru: It is a compelling measurement strategy for functional spatial phenotypes and histopathology relevant to tumor biology and metabolic-state modeling.
NELLY enables patient-centric drug prioritization through interpretable drug-conditioned gene weighting
bioRxiv (Cancer Biology) Published 2026-08-26 Preprint DOI:
drug discovery precision oncology interpretable biomedical AI patient-derived models
Summary: NELLY combines patient-centric drug-response prioritization with interpretable, drug-conditioned gene weighting and evaluates predictions in patient-derived organoids and out-of-distribution settings across cancer types.
Why it matters: The framework targets the translational gap between cell-line response prediction and individualized therapy selection while exposing patient-specific gene attributions.
Why for Yiru: It directly connects interpretable biomedical AI with precision oncology and offers a useful design for testing drug prioritization in patient-derived systems.
A binding-to-release strategy for targeted anticancer drug delivery
Nature Published 2026-08-26 Research article DOI:
oncology drug delivery therapeutic design
Summary: A binding-to-release drug-conjugate strategy enables targeted payload release without requiring cellular internalization, improving tumor specificity and efficacy while broadening the target space beyond conventional drug conjugates.
Why it matters: The strategy challenges the assumption that targeted payloads must be internalized and proposes a mechanistic route to expand the targetable antigen space.
Why for Yiru: It is a high-impact oncology drug-delivery concept with direct relevance to therapeutic design and a clear mechanistic innovation.