PFBNet

PFBNet infers gene regulatory networks (GRNs) from time-series gene expression data by combining non-linear boosting models with prior-information fusion, incorporating knockout data to improve inference accuracy.

Key Features:

• Non-Linear Boosting Model: Uses a boosting-based model to compute confidence scores for regulatory relationships while accounting for accumulation effects of gene expression at previous time points to capture non-linear temporal dependencies.
• Prior Information Fusion: Integrates prior knowledge such as knockout experiments into the inference process to elevate confidence in regulatory relationships associated with known regulators.
• Performance and Validation: Validated on DREAM challenge benchmark datasets and E.coli datasets and reported to outperform Jump3, GEINE3-lag, HiDi, iRafNet, and BiXGBoost, demonstrating robustness to noisy, high-dimensional data with many potential interactions.

Scientific Applications:

• Dynamic GRN reconstruction: Reconstructs time-resolved gene regulatory networks to elucidate temporal regulatory mechanisms.
• Systems biology with prior data: Integrates knockout and other prior biological knowledge to refine GRN hypotheses in systems biology studies.
• Cross-organism and experimental analysis: Applies to diverse organisms and experimental conditions to provide insights into fundamental biological processes and potential therapeutic targets.

Methodology:

PFBNet first calculates confidence scores for regulatory relationships using a boosting-based model that accounts for temporal gene expression accumulation, then applies an information fusion strategy to integrate prior information such as knockout data into the inference process.