Research Radar — 2026-05-12
Methods & AI
Computational
Deep Computational Anatomy via Latent-Aligned Multiview Normalizing Flows
bioRxiv Published 2026-05-05 preprint DOI:
normalizing flows deep learning computational anatomy multiview learning representation learning medical imaging
Summary: Proposes latent-aligned multiview normalizing (LAMNr) flows that learn shared latent subspaces across heterogeneous multimodal datasets while topologically unfolding sampled data manifolds. Formal latent-alignment constraints model shared structural features separate from view-specific variations. Applied to biological imaging (imaging-derived phenotypes and multimodal MRI), with a 2D/3D open-source PyTorch implementation integrated into the ANTsX ecosystem (ANTsTorch).
Why it matters: Extends normalizing flows to multiview medical imaging with formal latent-alignment guarantees. The ANTsX integration makes this practically deployable for population-level neuroimaging studies.
Why for Yiru: Relevant to representation learning for multimodal biomedical data. The multiview latent alignment framework could be adapted for spatial multi-omics integration or single-cell multimodal analysis.
Biomedical discoveries
Biomedicine
All-trans retinoic acid destabilizes ADAR1 protein through retinoylation-mediated USP7 dissociation and improves immunotherapy in pancreatic cancer
Nature Communications Published 2026-05-11 research article DOI: 10.1038/s41467-026-72271-5
ADAR1 immunotherapy pancreatic cancer retinoic acid PD-L1 tumor microenvironment USP7
Summary: ADAR1 contributes to immunotherapy resistance by suppressing interferon signaling. This study discovers that all-trans retinoic acid (ATRA) promotes ADAR1 protein degradation via retinoylation, which dissociates the deubiquitinase USP7 from ADAR1. ATRA also induces PD-L1 expression, and combining ATRA with PD-1 blockade reprograms the tumor microenvironment to unleash antitumor immunity, impeding tumor growth in pancreatic cancer models.
Why it matters: First demonstration that ADAR1 can be pharmacologically targeted for degradation. ATRA is already clinically approved, making this combination strategy readily translatable. Addresses a major unmet need in pancreatic cancer immunotherapy.
Why for Yiru: Directly relevant to tumor immunology and immunotherapy resistance mechanisms. The ADAR1-USP7-retinoylation axis is a new mechanistic angle linking post-translational modification to immune evasion. Connects to interests in TME reprogramming and combination immunotherapy strategies.
Temperate phage microdiversity reflects infant gut microbiome maturation independent of chronic undernutrition
bioRxiv Published 2026-05-10 preprint DOI:
gut microbiome temperate phages virome infant development metagenomics microdiversity
Summary: Uses longitudinal fecal metagenomes from Zimbabwean infants with normal and stunted growth to characterize gut bacterial and temperate phage community development from birth to 18 months. Finds that temperate phages target hallmark early-life bacteria (Bifidobacteriaceae) and exhibit age-dependent maturation parallel to bacterial succession. Viral microdiversity models predict chronological age better than abundance-based models, especially around weaning. Stunting did not significantly delay phage maturation.
Why it matters: Demonstrates that within-phage genomic variation (microdiversity) is a previously underappreciated dimension of gut microbiome development. Challenges the assumption that chronic undernutrition universally impairs microbiome maturation.
Why for Yiru: Marginal relevance. Demonstrates how metagenomic-derived features (viral microdiversity) can capture biological age with high resolution. The analytical framework for tracking strain-level variation could inform computational methods for microbiome biomarker discovery.