bAIcis: A Novel Bayesian Network Structural Learning Algorithm and Its Comprehensive Performance Evaluation Against...

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning

Background:

• Structural learning of Bayesian networks (BNs) is crucial for analyzing observational data across various scientific fields.
• Existing open-source BN learning tools struggle with large feature spaces and accurate structure recovery.

Purpose of the Study:

• To evaluate the performance of bAIcis, a novel BN learning software, against established open-source BN learners.
• To assess bAIcis's capability in handling diverse data types and large feature spaces.

Main Methods:

• Performance evaluation using synthetic datasets with discrete, continuous, and mixed data types.
• Comparison of bAIcis with publicly available Bayesian network learning algorithms.
• Investigation across small and large feature space scenarios.

Main Results:

• bAIcis demonstrated superior performance in structure recovery precision compared to open-source alternatives.
• Achieved high true positive rates (0.9) and precision (0.8) in structure recovery.
• bAIcis effectively handles continuous, discrete, and mixed data types and scales to large feature sets.

Conclusions:

• bAIcis offers a significant advancement in Bayesian network learning, particularly for "Big Data" applications in healthcare and omics research.
• The software's parallelization and scalability make it suitable for datasets previously considered infeasible for accurate analysis.