#6 · Single-cell

Mechanistic and autoencoder models decode single-cell transcriptional dynamics

The dominant thread is inference of transcriptional dynamics and cell or gene states from single-cell expression data using explicitly structured models. DeepTX couples mechanistic transcription models to scRNA-seq through deep learning, inferring genome-wide burst kinetics and linking DNA-damage-induced bursting to the response. SPAE applies a piecewise linear autoencoder to cell cycle dynamics and cell states, and is reported to outperform existing methods. NIRD combines matrix factorisation with a tree ensemble to infer non-linear gene networks from sparse single-cell profiles. What is new is the move from descriptive state mapping toward parameterised, perturbation-linked kinetics at single-cell resolution. The histopathology pathway-prediction preprint operates on whole-slide images rather than single-cell data and does not share this thread; it is included for its comparison of direct image-based prediction against expression-mediated models.