#2 · Single-cell
Encoding spatial and relational structure in high-dimensional omics analysis
Most items encode spatial neighborhood, graph, or contact structure explicitly: GNN augmented graphs (Concordia), graph-regularized joint NMF (SpaNiche), reaction-diffusion dynamics (SpaFlow), and Monte Carlo G-function proximity analysis. Several add formal null or consistency criteria, including the embedding-permutation null in tensor-CP similarity networks, structural-zero-aware dissimilarity for single-cell Hi-C (szKendall), and augmented graphs for consistent domain detection across thousands of tissues. ISPAT-3D extends spatial structure to 3D tumor imaging via anisotropic Gaussian-process adjustment. The multi-omic benchmark and CITE-seq tensor integration are less explicitly spatial, relying on tensor and multi-modal structure instead. What is new relative to earlier relational approaches is the pairing of spatial or topological structure with formal null or consistency criteria, applied at 3D and population scale.
Published papers
A systematic delineation of 3'UTR regulatory elements and their contextual associations.
Massively parallel synthetic 3'UTR reporter screen (~2M inserts, A549/HCT116) maps short regulatory elements; a context-aware deep model prioritizes ClinVar…
TFXcan reveals transcription factor programs driving GWAS loci.
TFXcan predicts personalized TF binding from genotypes with Enpact (Enformer plus IMPACT) and links TF-tissue pairs to prostate cancer risk loci, plus…
An extensible deep learning framework for multi-type chromatin interaction prediction through hierarchical adaptation strategies.
UniLoopNet: a multi-task deep learning framework predicting enhancer-promoter, enhancer-enhancer, CTCF and stripe chromatin interactions from DNA sequence,…
Preprints
SPARK: Bayesian structure learning of enhancer-based gene regulatory networks from single-cell multiomics data and prior knowledge
SPARK is a Bayesian DAG-based framework that reconstructs enhancer gene regulatory networks from single-cell multiome data, using prior knowledge; it…
Representation learning of somatic missense mutation landscapes across cancer types
MMRL, a self-supervised pan-cancer framework, learns representations of somatic missense mutations from ~50,000 tumors, outperforming TMB for prognosis and…
Scooby-2: fine-tuning AlphaGenome for single-cell genomic track prediction
Scooby-2 fine-tunes AlphaGenome with a cell-specific decoder to predict single-cell RNA-seq and chromatin accessibility tracks from DNA sequence, outperforming…
Topic history
| Date | Rank |
|---|---|
| #2 | |
| #3 | |
| #2 | |
| #1 | |
| #1 | |
| #2 | |
| #3 | |
| #3 | |
| #1 | |
| #2 | |
| #4 | |
| #3 | |
| #2 | |
| #5 | |
| #5 | |
| #5 | |
| #3 | |
| #18 | |
| #18 | |
| #14 | |
| #12 |