A comparative study of statistical methods used to identify dependencies between gene expression signals

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• Understanding gene dependencies is crucial for modeling gene regulatory networks.
• Pearson's correlation is widely used but limited to linear associations in gene expression data.
• Non-linear and non-functional relationships necessitate alternative dependency measures.

Purpose of the Study:

• To summarize and evaluate methods for identifying dependencies between random variables, particularly gene expression data.
• To assess the strengths and limitations of various statistical and information-theoretic measures.
• To provide guidelines for selecting appropriate methods based on data characteristics.

Main Methods:

• Comparative analysis of dependency measures including Pearson's, Spearman's, Kendall's correlations, distance correlation, Hoeffding's D, Heller-Heller-Gorfine measure, mutual information, and maximal information coefficient.
• Systematic Monte Carlo simulations to evaluate method performance under varying sample sizes and relationship types (linear, non-linear, non-functional).
• Application and comparison of methods on actual gene expression datasets.

Main Results:

• Pearson's correlation is effective only for linearly associated data.
• Rank correlation and information-theoretic measures are more suitable for non-linear or non-functional dependencies.
• Simulation and real-data analyses demonstrate the performance variations of different methods.

Conclusions:

• Clear guidelines are needed for applying advanced dependency measures in gene expression analysis.
• The study provides a framework for selecting appropriate methods based on data characteristics and relationship types.
• This research aids in more accurate gene regulatory network modeling by recommending suitable dependency identification techniques.