Are the tools fit for purpose? Network inference algorithms evaluated on a simulated lipidomics network.

Area of Science:

• Metabolomics
• Systems Biology
• Bioinformatics

Background:

• Constructing metabolic networks from metabolomic data is crucial for understanding cellular functions.
• Existing methods face challenges due to small sample sizes, confounding factors, indirect interactions, and biological randomness.

Purpose of the Study:

• To benchmark correlation- and regression-based network inference algorithms for metabolomic data.
• To evaluate algorithm performance across varying sample sizes using a generative model.
• To assess the ability of these methods to recover known metabolic networks.

Main Methods:

• Benchmarking of existing correlation- and regression-based network inference algorithms.
• Performance evaluation using a generative model with different sample sizes.
• Analysis using standard interaction-level tests and network centrality scores.

Main Results:

• Significant challenges exist in metabolic network inference and result interpretation, even with large sample sizes and modeled data.
• Correlation-based network inference showed limited ability to discriminate between two metabolic states in a computational model.
• The methods were not effective in identifying direct metabolic pathways but could indicate broader metabolic state changes.

Conclusions:

• Metabolomic network inference presents considerable challenges, impacting the reliability of identified pathways.
• Correlation-based approaches may offer insights into overall metabolic state shifts rather than specific pathway alterations.
• Further methodological development is needed for accurate reconstruction of metabolic networks from metabolomic data.