#5 · Single-cell

Methods separate coordinated gene programs from confounding variation in transcriptomes

The dominant thread is separating coordinated biological structure from confounding variation in transcriptomic data. DeSPA uses linear mixed models to detect cross-state spatial pattern remodeling across multi-slice atlases, and SORT applies matrix factorization with orthogonality and graph total variation to recover multilevel spatial gene programs across patients. These two items most directly pair multi-slice spatial organization with cross-state or cross-patient comparison. ALBIS contributes a simulator that generates multi-resolution 2D sections from 3D tissues with ground truth, which supports evaluation of such methods. Other items address distinct problems: DKO predicts cell-specific virtual perturbation effects and reports rising heterogeneity with aging, IsoGraph separates isoform usage from gene abundance in the aging human brain, PRISM estimates sample-specific correlation networks, and BEAST and FateScape use lineage barcodes for phylogenetic and division-history inference.