#2 · Single-cell
Frozen single-cell foundation models extended to multimodal completion and reconstruction
The dominant thread is reuse of pretrained single-cell representations beyond their original scope. scAnyOmics distills a frozen single-cell foundation model into a multimodal student for any-to-any completion of RNA, protein and ATAC. MUTUAL aligns frozen foundation-model embeddings into a shared RNA-ATAC latent space trained on 3.0M multiome cells, enabling cross-modality analysis. Top2All benchmarks a Geneformer-based model that reconstructs transcriptome-wide expression from sparse high-expression genes in mouse adrenal RNA-seq across sex- and sex-steroid-defined conditions. Evaluation work runs alongside: BioTrajX scores pseudotime plausibility via Directionality, Order Consistency and Endpoint Validity, and a published review gives a unified mathematical categorization of 16 cell potency estimators. A second strand addresses dynamics and lineage through ManifoldVelo (joint pseudotime and RNA velocity), bowside (trajectories coupled to GRNs), LOTUS (lineage-aware unbalanced optimal transport) and lineage inpainting. VIGAT, CloneVI and IsoPrism cover gene activity, mitochondrial clones and isoform usage. The set is heterogeneous, so this is a dominant pattern, not a shared method.
Published papers
Categorization and analysis of 16 computational methods for estimating cell potency from single-cell RNA-seq data.
Review and unified mathematical categorization of 16 computational methods estimating cell potency from scRNA-seq, grouped as average-based, entropy-based, and…
IsoPrism predicts isoform usage from short-read transcriptomes through structure-guided competitive allocation.
IsoPrism is a structure-guided allocation framework that predicts isoform proportions from short-read single-cell and spatial transcriptomes using paired…
Preprints
Learning developmental dynamics from lineage context
Introduces lineage inpainting, a generative model using tree-structured transformers and stochastic interpolants to learn cell state dynamics across lineage…
scAnyOmics: Foundation-Model Distillation for Any-to-Any Completion of Multi-omics Virtual Cells
scAnyOmics distills a frozen single-cell foundation model into a multimodal student for any-to-any completion of RNA, protein, and ATAC modalities across…
Coupling cellular trajectories and RNA kinetics on continuous expression manifolds with ManifoldVelo
ManifoldVelo jointly infers pseudotime and RNA velocity from single-cell transcriptomes via a topology-conditioned manifold model, outperforming compared…
VIGAT: VLM-guided gene activity quantification from single-cell ATAC-seq
VIGAT uses a vision-language model to visually interpret ATAC coverage plots and select gene-specific regions for scATAC gene activity quantification,…
Mechanistic gene-regulation-based trajectory inference with bowside
Bowside, a shallow neural stochastic differentiation model, jointly infers single-cell trajectories and gene regulatory networks, tested on hematopoiesis and…
A single-cell multimodal foundation model enables cross modality analysis between transcriptome and chromatin
MUTUAL aligns frozen single-cell foundation model embeddings into a shared RNA-ATAC latent space trained on 3.0M multiome cells, enabling cross-modality…
LOTUS: lineage-aware optimal transport reconstructs continuous clonal fate dynamics
LOTUS, a lineage-aware unbalanced optimal transport framework, reconstructs continuous clonal fate dynamics from time-resolved scRNA-seq snapshots with barcode…
Top2All: Benchmarking sparse-to-transcriptome reconstruction from high-expression profiles across sex- and sex-steroid-defined conditions
Benchmarks Top2All, a Geneformer-based model reconstructing transcriptome-wide expression from sparse high-expression genes in mouse adrenal RNA-seq; gene-wise…
BioTrajX: evaluating biological plausibility in single-cell trajectory inference
BioTrajX, an R package with Directionality, Order Consistency, and Endpoint Validity metrics, evaluates biological plausibility of single-cell pseudotime…
CloneVI: A deep generative model for deciphering the relationships between cell state and mitochondrial genotype
CloneVI is a deep generative model that infers mitochondrial DNA mutation clones from single-cell RNA-seq and ATAC-seq cell states, validated in simulations…
Topic history
| Date | Rank |
|---|---|
| #2 | |
| #3 | |
| #1 | |
| #7 | |
| #23 | |
| #3 | |
| #1 | |
| #6 | |
| #2 | |
| #15 | |
| #23 | |
| #13 | |
| #5 | |
| #35 | |
| #35 | |
| #32 | |
| #32 | |
| #30 | |
| #31 | |
| #34 | |
| #33 | |
| #37 |