Benchmarking time-series data discretization on inference methods

Area of Science:

• Systems biology
• Computational biology
• Bioinformatics

Background:

• Quantitative measurement of DNA, RNA, and proteins drives interest in reverse-engineering biological networks.
• Many reverse-engineering methods require discrete data, but experimental data are often continuous.
• The impact of data discretization on inference method performance requires further systematic study.

Purpose of the Study:

• To systematically evaluate the impact of different data discretization methods on reverse-engineering performance.
• To develop and validate a method for selecting optimal discretization techniques for time-series biological data.
• To address the need for comparative methods to guide the choice of discretization strategies.

Main Methods:

• Applied various discretization methods to time-series datasets from four published intracellular networks.
• Evaluated the performance of reverse-engineering methods using the discretized data.
• Developed DiscreeTest, a two-step metric to rank discretization methods based on preserving dynamic patterns.
• Validated DiscreeTest using the same datasets and networks.

Main Results:

• Data discretization significantly impacts reverse-engineering performance across all tested datasets.
• No single discretization method universally outperformed others across different time-series datasets.
• DiscreeTest successfully identified appropriate discretization methods from candidate options.
• The proposed metric assumes optimal discretization preserves original dynamic patterns.

Conclusions:

• Appropriate data discretization is crucial for improving reverse-engineering accuracy in systems biology.
• DiscreeTest provides a systematic approach to benchmark and select discretization methods for time-series data.
• This work presents the first method for evaluating and choosing discretization techniques for time-series biological data.