Challenges in Lipidomics Biomarker Identification: Avoiding the Pitfalls and Improving Reproducibility

Area of Science:

• Lipidomics
• Mass Spectrometry
• Biomarker Discovery

Background:

• Accurate feature identification in liquid chromatography-mass spectrometry (LC-MS) lipidomics is crucial for biomarker discovery.
• Existing software platforms often yield inconsistent results, impacting the reproducibility of lipid identification.

Purpose of the Study:

• To quantify the reproducibility gap between open-access lipidomics software platforms.
• To investigate the impact of spectral data and settings on identification agreement.
• To propose methods for improving the accuracy and reproducibility of lipid identification.

Main Methods:

• Comparative analysis of two lipidomics platforms (MS DIAL and Lipostar) using identical LC-MS spectra.
• Evaluation of identification agreement using default settings and MS2 fragmentation data.
• Application of a machine learning approach (support vector machine regression with leave-one-out cross-validation) for data-driven outlier detection.

Main Results:

• Low identification agreement (14.0%) between platforms using default LC-MS settings.
• Improved, but still limited, agreement (36.1%) when utilizing MS2 spectra.
• Demonstration of a machine learning method to enhance spectral output assessment.

Conclusions:

• Significant reproducibility gaps exist between lipidomics software platforms, posing challenges for biomarker discovery.
• Manual curation of spectra and software outputs, alongside data-driven validation, is essential to minimize identification errors.
• Novel machine learning approaches can supplement validation, reducing false positives and improving cross-platform consistency.