#5 · ML / AI methods
Structure-aware statistical methods for single-cell and spatial transcriptomics inference
The dominant thread is statistical inference tailored to the structure of single-cell and spatial measurements rather than generic models. QuasiMed performs quasi-regression mediation analysis for zero-inflated single-cell data without strict distributional assumptions. sctrial provides participant-level differential analysis of longitudinal scRNA-seq with calibrated error rates in simulations, and block-SAM respects nested sample structure in mass spectrometry imaging and scRNA-seq. In spatial transcriptomics and tissue imaging, GP-GHS infers spatially varying cell-cell interaction networks using HSGP coefficients and group horseshoe priors, while a counterfactual directionality score and a geometry-aware ligand-receptor framework separate directional influence and interface association from spatial localization. A benchmark of nine cell-cell interaction tools on simulated and real spatial datasets adds a comparative reference point. The novelty lies in explicit modeling of spatial variation, directionality, and hierarchical sample design.
Preprints
Combining Bayesian and Frequentist Inference for Laboratory-Specific Performance Guarantees in Copy Number Variation Detection
Proposes a hybrid Bayesian-frequentist framework giving tolerance intervals with valid coverage for per-gene CNV detection on targeted amplicon oncology…
Reconstructing intra-tumor fitness landscapes from scSeq CNA genotypes via simulation-based Bayesian inference and Deep Learning
Simulation-based neural posterior estimation with normalizing flows infers intra-tumor CNA selection coefficients from single-cell clonal genotypes,…
Testing and Estimating Causal Treatment Effect Heterogeneity in Observational Studies via Revised Deep Semiparametric Regression: A Lung Transplant Case Study
deepHTL: a gatekept semiparametric deep learning workflow testing causal effect heterogeneity of bilateral lung transplant on FEV1 in registry data, with…