Source research

Improved reconstruction of single-cell developmental potential with CytoTRACE 2

Minji Kang, José Juan Almagro Armenteros, Gunsagar Singh Gulati, Rachel Gleyzer, Susanna Avagyan, Erin L. Brown et al. · 2025 · 225 citationsRead the paper

Single-cell RNA sequencing (scRNA-seq) has transformed our understanding of cell fate in developmental systems. However, identifying the molecular hallmarks of potency - the capacity of a cell to differentiate into other cell types - has remained challenging. Here, we introduce CytoTRACE 2, an interpretable deep learning framework for characterizing potency and differentiation states on an absolute scale from scRNA-seq data. Across 31 human and mouse scRNA-seq datasets encompassing 28 tissue types, CytoTRACE 2 outperformed existing methods for recovering experimentally determined potency levels and differentiation states covering the entire range of cellular ontogeny. Moreover, it reconstructed the temporal hierarchy of mouse embryogenesis across 62 timepoints; identified pan-tissue expression programs that discriminate major potency levels; and facilitated discovery of cellular phenotypes in cancer linked to survival and immunotherapy resistance. Our results illuminate a fundamental feature of cell biology and provide a broadly applicable platform for delineating single-cell differentiation landscapes in health and disease.

What the full text shows

Working prototype

CytoTRACE 2 outperformed existing supervised machine learning and developmental potential inference methods in recovering experimentally determined potency levels and differentiation states. (Across 31 human and mouse scRNA-seq datasets encompassing 28 tissue types)
The model successfully reconstructed the temporal hierarchy of mouse embryogenesis. — 62 timepoints (Mouse embryogenesis data)
The framework identified pan-tissue expression programs that discriminate major potency levels and cellular phenotypes in cancer linked to survival and immunotherapy resistance. (Cross-tissue scRNA-seq data)

Setup: Training cohort of 312,523 cells (16 tissue types, 93 phenotypes) from 19 human and mouse datasets; test cohort of 93,535 cells (9 tissue types, 73 phenotypes) from 14 datasets.
How: Developed a deep learning framework using Gene Set Binary Networks (GSBN) that assign binary weights to genes to identify discriminative gene sets. It integrates rank-based and expression-based enrichment scores and uses Markov diffusion with a nearest neighbor approach to smooth final potency scores.
Limitations: Tabula Sapiens data exhibited reduced quality due to postmortem intervals (mean MTR 7.4% vs 3.9% expected). · Some benchmarking methods required downsampling of larger datasets (e.g., Tabula Muris to 30 cells per phenotype) for computational feasibility.

Summarized by Gemma 4 from the open-access full text · 9 Oct 2026. Read it