Correlated measurement error hampers association network inference.

Area of Science:

• Analytical Chemistry
• Systems Biology
• Bioinformatics

Background:

• Metabolomics studies generate large datasets requiring robust analysis methods.
• Association networks are increasingly used to interpret complex metabolomics data.
• Chromatography-based metabolomics data possess complex error structures, including correlated measurement errors.

Purpose of the Study:

• To investigate the impact of correlated measurement errors on partial correlation-based association networks in metabolomics.
• To raise awareness about the influence of correlated measurement errors on network interpretation.
• To differentiate the effects of correlated errors on direct versus indirect associations.

Main Methods:

• Analysis of chromatography-based, time-resolved lipidomics data from a human intervention study.
• Construction and evaluation of partial correlation-based association networks.
• Visual distinction of correlated measurement error from biological variation in time-series data.

Main Results:

• Correlated measurement errors can significantly impact the reliability of association networks.
• The effect of correlated errors differs between direct and indirect associations.
• Indirect associations are particularly vulnerable to becoming unreliable due to correlated measurement errors.

Conclusions:

• Correlated measurement errors pose a significant challenge to the biological interpretation of metabolomics association networks.
• Underestimating these errors can lead to false biological discoveries.
• Experimental designs that quantify error structures are essential for identifying spurious network connections.