Reconstructing noisy gene regulation dynamics using extrinsic-noise-driven neural stochastic differential equations

Area of Science:

• Computational Biology
• Systems Biology
• Biophysics

Background:

• Cell signaling and gene expression require precise regulation for cellular function and adaptation.
• Cellular processes exhibit noise from intrinsic stochasticity and extrinsic heterogeneity.
• Understanding noise is key to deciphering complex biological systems.

Purpose of the Study:

• To develop a novel computational framework for modeling cellular heterogeneity and reaction dynamics.
• To accurately reconstruct stochastic differential equations (SDEs) from noisy, heterogeneous cell population data.
• To provide a robust method for analyzing complex biophysical processes where mechanistic models are challenging.

Main Methods:

• Introduction of the extrinsic-noise-driven neural stochastic differential equation (END-nSDE) framework.
• Utilizing Wasserstein distance for accurate SDE reconstruction from stochastic trajectories.
• Validation with simulated data and experimental data from circadian rhythms, RPA-DNA binding, and NFκB signaling.

Main Results:

• The END-nSDE framework successfully models how cellular heterogeneity modulates reaction dynamics.
• The method accurately reconstructs SDEs from heterogeneous cell population data, capturing extrinsic noise effects.
• END-nSDE outperforms existing time-series analysis methods like RNNs and LSTMs in modeling noisy biological dynamics.

Conclusions:

• The proposed END-nSDE method offers a powerful surrogate modeling approach for complex biological systems.
• This framework effectively infers cellular heterogeneities and reproduces observed noisy dynamics.
• The study advances the analysis of stochastic processes in cell biology by accounting for extrinsic noise.