Structure Identification and Dynamical Reconstruction of Gene Regulatory Networks Based on Neural Ordinary Differential Equations
Ingegneria Sismica · 10.06.2026 · 42 (4) · 1-19
DOI: 10.65102/is20261131 ↗Kurz gefasst
Fachartikel in der Zeitschrift Ingegneria Sismica, online erschienen am 10.06.2026. Verfasst von Kangsheng Song.
Abstract (Originalsprache)
To solve the split between structural recovery and continuous dynamics’ characterisation in gene regulatory network identification, we put forward a Neural ODE-GRN joint modelling method which is based on neural ordinary differential equations. This method uses time-series bulk RNA-seq and pseudo-time-series single-cell RNA-seq expression matrices as input materials, introduces continuous-time evolution equations into a hidden state space, and combines sparse structural masks, symbolic constraints, biological priors and temporal consistency loss to realize co-optimisation of regulatory adjacency matrix learning and expression trajectory reconstruction. Therefore, validation works were conducted with simulated data from GeneNetWeaver and SERGIO, as well as real-world data from hESCs and mESCs. Hence, results show that on a 100-gene simulated network, the model’s AUPR achieved 0.631, which is an improvement of 0.042 to 0.114 compared with various baselines; under 20% noise conditions, the AUROC kept above 0.80; on real-world data, the trajectory correlation coefficient reached 0.914-0.928, at the same time maintaining good topological consistency for hub nodes and representative regulatory modules. Therefore, this method provides a new computational tool for explaining dynamic regulatory mechanisms in developmental differentiation and disease progression.
Zitieren
APA
Song, K. (2026). Structure Identification and Dynamical Reconstruction of Gene Regulatory Networks Based on Neural Ordinary Differential Equations. Ingegneria Sismica, 42(4), 1-19. https://doi.org/10.65102/is20261131
Vancouver
Song K. Structure Identification and Dynamical Reconstruction of Gene Regulatory Networks Based on Neural Ordinary Differential Equations. Ingegneria Sismica. 2026;42(4):1-19. doi:10.65102/is20261131
BibTeX
@article{Song2026,
author = {Song, Kangsheng},
title = {Structure Identification and Dynamical Reconstruction of Gene Regulatory Networks Based on Neural Ordinary Differential Equations},
journal = {Ingegneria Sismica},
year = {2026},
volume = {42},
number = {4},
pages = {1--19},
doi = {10.65102/is20261131}
}
EUVIMED