Learning stochastic processes with intrinsic noise from cross-sectional biological data.

Area of Science:

• Computational Biology
• Systems Biology
• Biophysics

Background:

• Inferring dynamical models from biological data is challenging due to stochastic processes.
• Omics data often consists of independent cross-sectional samples at limited time points.
• Existing methods often oversimplify or ignore intrinsic system noise, impacting accuracy.

Purpose of the Study:

• To develop a novel inference method that accurately models stochastic biological processes.
• To infer the underlying diffusion process from time-series omics data.
• To disentangle system forces from intrinsic stochasticity.

Main Methods:

• Developed Probability Flow Inference (PFI) to model phase-space probability flow.
• PFI retains time-dependent marginal distributions of the stochastic process.
• Utilized ordinary differential equation (ODE) inference principles for algorithmic ease.

Main Results:

• Analytically proved unique solutions for Ornstein-Uhlenbeck processes under PFI.
• Demonstrated accurate parameter and force estimation in high-dimensional stochastic reaction networks.
• Successfully inferred cell differentiation dynamics with molecular noise, surpassing current methods.

Conclusions:

• PFI offers a robust framework for inferring dynamical models from noisy biological data.
• The method accurately captures system dynamics without compromising for optimization ease.
• PFI advances computational biology by enabling more precise modeling of complex biological systems.