CaiNet

CaiNet simulates and infers gene regulatory networks (GRNs) with combined stochastic and deterministic molecular detail to model gene expression dynamics, noise-induced behaviors (bi-stability and oscillations), and to estimate regulatory connections and steady-state parameters.

Key Features:

• Stochastic simulation: Decouples network elements over short time intervals to apply local stochastic or deterministic solutions and then synchronizes molecule numbers across elements.
• Modular approach: Treats each network element in isolation during brief periods and integrates results to approximate solutions of deterministic differential equations while incorporating stochastic events.
• Incorporation of stochastic detail: Models genes switching stochastically between active and inactive states to reproduce noise-induced phenomena such as bi-stability and oscillations.
• Deterministic delays: Includes deterministic delays in gene expression to represent temporally extended processes.
• Inference capabilities: Infers regulatory connections and steady-state parameters for GRNs of up to ten genes by mapping each gene to a perceptron within an artificial neural network and applying gradient descent methods adapted from recurrent neural network training.
• Explicit stochastic events: Represents stochastic molecular interactions such as transcription factor-DNA interactions and gene product production.

Scientific Applications:

• Cellular and organismic biology: Analysis of GRN-driven processes underpinning homeostasis, differentiation, and development.
• Temporal reaction network analysis: Prediction and analysis of the temporal progression of reaction networks under stochastic and deterministic influences.
• Noise impact studies: Investigation of how stochastic events affect gene regulation, including noise-induced bi-stability and oscillations.
• Parameter and topology inference: Estimation of regulatory connections and steady-state parameters in small-scale GRNs (up to ten genes).

Methodology:

Decoupling of network elements over small time intervals with local stochastic or deterministic solutions followed by synchronization of molecule numbers; modular isolation and integration approximating deterministic differential equations; explicit stochastic gene switching between active/inactive states; inclusion of deterministic gene expression delays; mapping genes to perceptrons in an artificial neural network and inferring parameters via gradient descent adapted from recurrent neural network training.