Experimental design of time series data for learning from dynamic Bayesian networks.

Area of Science:

• Computational biology
• Systems biology
• Bioinformatics

Background:

• Bayesian networks (BNs) and dynamic Bayesian networks (DBNs) are increasingly utilized for inferring biological networks from high-throughput data.
• High experimental costs necessitate efficient data generation strategies for network modeling.

Purpose of the Study:

• To determine optimal properties of time series data for dynamic Bayesian network (DBN) modeling.
• To guide experimental design for generating efficient time series data for DBNs.

Main Methods:

• Theoretical analysis of dynamic Bayesian networks (DBNs) without hidden variables.
• Evaluation of DBN learning capabilities with proteomic time series data.

Main Results:

• Under specific assumptions, a fixed experimental budget yields more efficient DBN modeling when applied to numerous short time series datasets.
• Conversely, investing in a few long time series datasets is less efficient for DBN parameter learning.

Conclusions:

• Optimizing experimental design for time series data generation is crucial for effective DBN modeling.
• Prioritizing the collection of multiple short time series is recommended for cost-effective proteomic network inference using DBNs.