Reducing double-dipping in single-cell topic models with three-way count splitting

Zheng Qu

Home/Archive/arXiv

arXiv · 2610.24130

Reducing double-dipping in single-cell topic models with three-way count splitting

Zheng Qu

View original source ↗Read source document ↗Export metadata ↓

Abstract

Topic models learn gene programs from single-cell RNA sequencing (scRNA-seq) data, and the same counts are then used to test which genes and donors are associated with those programs. Count splitting is the standard remedy for this reuse, but it has not been tested when an encoder reads each cell's counts. On a 104,400-cell lupus cohort with 300 permuted null genes per run, naive tests rejected 36% of null gene-topic pairs at a nominal 5% level, and the usual two-way split rejected 38%, because the encoder sees the tested gene's own held-out count. Encoding topic usage on a separate third of the counts, with beta-binomial thinning, lowered this to 7.7% (6.7% for genes outside the model); for two other models, LDVAE and scETM, the rate was 7.0-7.3% (naive 51-59%). A label-free interferon program shifts in held-out counts in the lupus cohort and in a pre-specified COVID-19 cohort, and donor-level topic usage predicts disease (AUROC 0.96) beyond cell-type proportions (0.86 and 0.58 in the two cohorts). Calibration and prediction did not depend on the model. We do not claim better clustering, prediction or topic stability. Code, figures and result tables are public (MIT license; DOI 10.5281/zenodo.23307193).

Explore related subjects

q-bio.GN cs.LG

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Zheng Qu. 2026-10-16. Reducing double-dipping in single-cell topic models with three-way count splitting. https://arxiv.org/abs/2610.24130

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

PACE: Geometry-Aware Bridge Transport for Single-Cell Trajectory Inference

Single-cell trajectory inference from destructive time-course snapshots is fundamentally ill-posed: neither cross-time cell correspondences nor continuous trajectories are observed, so the snapshot distributions alone do not uniquely determine the underlying dynamics. Existing optimal transport and flow-based methods typically couple cells by Euclidean proximity at observed clock times, which can misalign trajectories when development is asynchronous and cells sampled at the same experimental time occupy different latent pseudotime stages. We propose PACE, a trajectory inference framework that recovers geometry-consistent continuous transport dynamics from destructive time-course snapshots through three coupled components. First, PACE constructs a state- and time-dependent anisotropic Riemannian metric that assigns low transport cost along locally supported tangent directions while penalizing normal velocity components. Second, it alternates between refining cross-time couplings under the induced path-action cost and fitting endpoint-preserving neural bridges between adjacent snapshots. Third, it distills the learned bridge dynamics into a global continuous-time velocity field over cellular states. Across seven controlled and biological datasets covering nine held-out reconstruction experiments, PACE achieves the strongest overall reconstruction performance, reducing MMD, Wasserstein-1 distance, and Wasserstein-2 distance by 23.7% on average relative to the strongest competing baseline. PACE also improves RNA-velocity alignment by 15.4% on an embryoid body differentiation benchmark, without requiring explicit cell pairing, lineage tracing, or RNA-velocity supervision during training. Code is available at https://github.com/AI4Science-WestlakeU/PACE.

q-bio.GN↗

arXiv2026

Application of Single-cell Deep Learning in Elucidating the Mapping Relationship Between Visceral and Body Surface Inflammatory Patterns

The subcutaneous fascia (scFascia) is a widely distributed connective tissue, yet its contribution to systemic responses during visceral inflammation remains insufficiently defined. In the rat model of Salmonella-induced dysentery, scRNA-seq of scFascia combined with deep neural network (DNN)-based nonlinear feature discovery revealed transcriptional alterations indicative of Warburg-like reprogramming in macrophages (MPs), accompanied by reduced activity of the citrate cycle. CD34+/PDGFRα+ telocytes (CPTCs) regulated MP differentiation and proliferation through Wnt/Fgf signaling, suggesting a pathological communication pattern between scFascia and visceral organs, termed the fascia-visceral inflammatory crosstalk pattern (FVICP). PySCENIC analysis showed increased activity of transcription factors Fosl1, Nfkb2, and Atf4, modulated by CPTC signaling to MPs, leading to downregulation of aerobic respiration and upregulation of the cell cycle, DNA replication, and transcription. Co-culture assays demonstrated that CPTCs suppressed the pro-inflammatory activity of MPs by transcriptionally downregulating TLR4 and CPT1A while upregulating ATF4, and enhanced anti-inflammatory potential via post-transcriptional MTOR upregulation. These findings highlight the role of scFascia in immunomodulation and metabolic reprogramming during visceral inflammation, emphasizing its contribution to systemic homeostasis.

q-bio.GN↗

arXiv2026

Evaluating Conformal Reliability of Pathway-Level Transcriptomic Signatures Under Cross-Cohort Shift in Sepsis Mortality Prediction

Blood transcriptomic profiling enables prognostic modeling by capturing the host immune response at the molecular level. Yet, the within-cohort evaluation strategies employed by many transcriptomic models inadequately reflect deployment across independent hospitals. Outside deployment scenarios introduce a cohort shift that can substantially degrade predictive performance and reliability of uncertainty estimates. We present a framework for evaluating transcriptomic sepsis mortality prediction under realistic cross-cohort deployment, systematically comparing gene-level, pathway-level and hybrid molecular representations. Four publicly available whole-blood transcriptomic cohorts consisting of 936 patients and 248 mortality events were harmonized into a shared 7,660-gene feature space and evaluated under leave-one-cohort-out validation using logistic regression, random forests, XGBoost and LightGBM. Beyond AUROC and AUPRC, model behavior was evaluated via conformal prediction, calibration analysis, selective prediction and the proposed Pathway Stability Index. Gene-level and hybrid representations were found to generally achieve the strongest discriminative performance, whereas pathway-level representations exhibited greater robustness across model families, more reliable uncertainty behavior under cross-cohort shift and stable molecular signatures enriched for immune and host-defense processes identified through Gene Ontology and KEGG enrichment analyses. These findings demonstrate that molecular representation influences not only predictive discrimination but also calibration, uncertainty reliability, biological coherence and transferability under external validation.

q-bio.GN↗