Bayesian networks for network inference in biology.

Area of Science:

• Computational biology
• Bioinformatics
• Systems biology

Background:

• Bayesian networks (BNs) are graphical models used to represent probabilistic relationships between variables.
• In the 2000s, BNs showed promise for inferring biological networks from data across various disciplines.
• Their application spans molecular biology, ecology, and neuroscience.

Purpose of the Study:

• To review and evaluate the efficacy of Bayesian networks in reconstructing biological interactions from data over the past two decades.
• To assess the successes and limitations of BNs in biological network inference.

Main Methods:

• Systematic review of studies employing Bayesian networks for biological data analysis.
• Evaluation of BN performance based on reported successes in identifying novel interactions and network components.
• Analysis of factors influencing BN success, such as dataset size, data quality, and data discretization.

Main Results:

• Bayesian networks have demonstrated success in inferring biological networks, often uncovering previously unknown interactions or components.
• The accuracy of BNs is difficult to fully assess due to underreporting of false positive results.
• BN performance is optimized with smaller numbers of variables, high-quality datasets, discretized data, or perturbative data.

Conclusions:

• While BNs may not have fully met the initial high expectations from the 2000s, their limitations are largely attributed to experimental data constraints.
• Bayesian networks remain a powerful and versatile tool for biologists, capable of inferring complex networks across diverse biological contexts.