tigre

tigre infers transcription factor (TF) activity and regulatory model parameters from gene expression time series of single input motif networks using Gaussian process differential equation models.

Key Features:

• Gaussian Process Priors: Uses Gaussian process priors to represent latent chemical species in biochemical interaction networks, enabling estimation of structure and parameters of genetic, metabolic, and protein interaction networks.
• Bayesian Marginalization: Applies Bayesian marginalization to infer TF activity when active protein concentrations are unobserved.
• Likelihood Function Derivation: Integrates out uncertainty in inferred TF activity to derive a likelihood function for estimating regulatory model parameters.
• Efficient Inference Schemes: Implements exact inference for linear regulation and approximate inference for non-linear regulation that are more computationally efficient than sampling-based approaches.
• Avoidance of Coarse-Grained Discretization: Avoids coarse-grained discretization of continuous time functions to reduce the number of estimated parameters and improve computational efficiency.

Scientific Applications:

• Inference of Unobserved TF Concentrations: Inferring unobserved transcription factor protein concentrations from target gene expression time series.
• Target Ranking: Ranking candidate TF targets based on inferred TF activity.
• Regulatory Mechanism Analysis: Analyzing activation and repression mechanisms within single input motif transcriptional regulation.

Methodology:

Uses Gaussian process differential equation models for gene expression time series in single input motif networks; employs Bayesian marginalization of latent chemical species, derives likelihoods by integrating out TF activity uncertainty, implements exact inference for linear regulation and approximate schemes for non-linear regulation, and avoids coarse-grained time discretization.