RESIDE: Reconstructing Network Interactions and Stochastic Dynamics from Single-Cell Snapshot Expression Data

Abstract

Deciphering stochastic gene regulatory dynamics and their governing network architectures remains a central challenge in systems biology. While single-cell sequencing advancements have significantly enhanced our understanding of large-scale gene regulatory networks, these snapshot measurements inherently lack temporal resolution, thereby limiting our ability to capture underlying dynamical processes. Current dynamical reconstruction approaches face complementary limitations: data-driven methods usually lack interpretation of molecular mechanisms, whereas model-driven strategies usually lack integration of quantitative experimental data. Here, we introduce RESIDE, a unified single-cell framework that integrates model- and data-driven methods for stochastic dynamics reconstruction, coupled with concurrent network structure inference and quantification of steady-state distributions, vector fields, and underlying landscapes. Across the tested in silico systems, RESIDE showed favorable performance relative to the compared methods. Applied to single-cell data from mouse preimplantation development, it recovered differentiation features consistent with the observed cell-state organization and predicted directionally asymmetric transdifferentiation paths. Overall, RESIDE provides a mechanistically interpretable framework for jointly inferring network interactions and effective stochastic dynamics from single-cell snapshot data, with potential value for hypothesis generation and experimental investigation.