BCTI: a Bayesian network-based method for revealing critical transitions in complex biological systems

The identification of critical states during disease progression is essential yet challenging for preventing disease deterioration and developing precision therapies. Traditional methods often rely on the dynamic feature of coordinated molecular variation to provide early-warning signals of impending critical transitions. However, these methods typically overlook the causal relationships among variables, potentially limiting their interpretability in uncovering underlying molecular regulatory mechanisms.

With the rapid advancement of sequencing technologies and the surge in high-throughput data, we propose Bayesian Critical Transitions Inference (BCTI), inspired by the time-varying nature of gene regulatory networks. BCTI integrates mutual information and structural equation models to qualitatively capture dynamic changes in network topology and quantitatively evaluate system states through a network scoring mechanism, thereby enabling the efficient and robust dual detection of early-warning signals associated with critical transitions in disease progression.

The proposed BCTI was validated by a series of applications on simulated and real datasets of complex biological systems. BCTI achieved superior or comparable accuracy to benchmark methods in inferring gene regulatory networks (GRNs) and detecting critical states. All the results demonstrate the high effectiveness of the proposed method in analyzing time-course/stage-course high-dimensional expression data, providing new insights into precision medicine for clinical applications and the underlying regulatory mechanisms of biological systems.

The proposed method enables effective detection of critical transitions and reveals dynamic regulatory mechanisms in complex biological systems, demonstrating strong potential for applications in systems biology, precision medicine, and the exploration of key molecular regulation driving disease progression and development.