Research Radar — 2026-10-01

Summary: CytoVI uses a probabilistic latent model to integrate flow cytometry, mass cytometry and CITE-seq, impute missing markers and test protein differences. Benchmarks and a 350-protein B-cell atlas support applications to lymphoma-associated T-cell states.

Why it matters: Brings uncertainty-aware inference to heterogeneous antibody panels and permits joint use of large existing cytometry collections.

Why for Yiru: A practical comparator for immune-cell representation learning and multimodal state analysis.

Summary: Population-scale RNA and chromatin-accessibility profiling in a Finnish cohort links regulatory variants to cell-specific expression. The journal abstract reports 593,765 peak–gene links and MPRA support for 10,428 fine-mapped molecular QTLs.

Why it matters: Provides a population-scale test bed for tracing genotype-to-chromatin-to-expression relationships and studying regulatory buffering.

Why for Yiru: Highly relevant reference evidence for immune-state models and variant-to-function hypotheses.

Summary: ProxiNet learns pairwise cellular-proximity signatures from spatial references and transfers them to dissociated scRNA-seq. The abstract reports cross-region and cross-technology tests, spatially validated neighborhoods and Alzheimer-associated remodeling.

Why it matters: Offers a reusable representation of tissue organization without requiring a spatial assay for every query dataset.

Why for Yiru: Direct fit for spatial-to-single-cell transfer and interpretable neighborhood modeling.

Summary: BEACON models background barcode counts to improve sample demultiplexing and extends the approach to protein–transcriptome assays. The abstract reports multiple human benchmarks and prospective validation of a recovered CD161-positive memory CD4 T-cell population.

Why it matters: Demultiplexing errors can lose cells and distort downstream biological conclusions; background-aware inference tackles an upstream bottleneck.

Why for Yiru: Immediately relevant to reliable multiplexed single-cell and immunophenotyping workflows.

Summary: ClonoDynamics separates replicates used to condition abundance from those used to estimate change, exposing artificial restoring dynamics. Technical replicates, synthetic controls and ten adults sampled weekly for six weeks test longitudinal TCR behavior.

Why it matters: Shows how an estimator can manufacture apparent biological dynamics and provides a concrete control strategy.

Why for Yiru: Useful for designing longitudinal immune-state benchmarks and avoiding noise-driven trajectory claims.

Summary: Nailpolish corrects and deduplicates long-read UMIs through reference-free consensus generation while separating UMI collisions across diverse protocols.

Why it matters: Could reduce measurement error before downstream single-cell or spatial analyses.

Why for Yiru: A practical preprocessing candidate for long-read molecular profiling.

Summary: A Transformer learns from paired cell-position point clouds to organize tissue geometry over developmental time. The journal abstract reports landmark annotation and subtle-phenotype detection; the earlier full preprint evaluates C. elegans embryos and RNAi screens.

Why it matters: Cell coordinates become reusable features for morphology analysis alongside molecular embeddings.

Why for Yiru: A practical alternative representation for spatial single-cell data and perturbation phenotyping.

Summary: MIMOSA constrains diffusion-based antibody design using the geometry and chemistry of known cognate interfaces. Its abstract reports 20–26% experimental hit rates on three difficult targets with two different generators.

Why it matters: Tests whether structural prior knowledge can rescue targets where unconstrained generation yields few or no binders.

Why for Yiru: A concrete example of model-agnostic constraints paired with wet-lab evaluation for AI-guided discovery.

Summary: AGENT couples Bayesian optimization with experiments to optimize solid-state mRNA–LNP formulations, with thermostability and animal immune-response validation.

Why it matters: Demonstrates a closed experimental optimization loop for a practical formulation problem.

Why for Yiru: Transferable design for data-efficient AI4Science and drug-formulation workflows.

Summary: HARP combines about nine million PBMCs with 192 consensus cell subsets and introduces scTiger hierarchical label transfer.

Why it matters: Provides a reusable immune-annotation reference spanning studies and disease.

Why for Yiru: Useful for harmonizing computational-immunology datasets and testing annotation transfer.