A Systematic Evaluation Study on Parameter Inference of Gene Regulatory Networks via Physics-Informed Neural Networks
Inferring parameters of gene regulatory networks from time-series data is challenging due to non-convexity and poor identifiability. We propose Gene-PINN, a physics-informed neural network that embeds the governing ODEs into the loss function to jointly learn dynamics and estimate parameters. Systematic evaluation reveals that inhibitory edges and feedback loops trade off data fitting against parameter accuracy, network density has an optimal range, and topologies cluster into distinct identifiability groups. This work systematically characterizes how network structure governs parameter inference, guiding physics-informed machine learning in systems and synthetic biology.
Authors
- Dongyan Zhang (ORCID: https://orcid.org/0000-0002-5929-1282)
- Liu Hong
- Xiaoqiang Sun
Institutions
- Sun Yat-sen University (CN)
Publication Details
- Journal
- CSIAM transactions on life sciences.
- Published
- 2026-11-18
- DOI
- https://doi.org/10.4208/csiam-ls.so-2026-0721
- Primary Topic
- Gene Regulatory Network Analysis
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- National Natural Science Foundation of China