Improving metabolomics data comparability without long term quality controls using a post-acquisition correction strategy

Background

Recent advances in analytical techniques for metabolomics allowed generating data of increasing quality in terms of sensitivity and robustness, thus opening the door to its large-scale application. However, the integration of separately collected metabolomic data is currently limited by the lack of methods able to correct for the analytical bias without long-term quality controls. This significant bottleneck prevents inter-comparisons across studies and limits metabolomics impact in precision biology. Overcoming these major challenges is therefore of great importance in many application fields to improve interoperability across studies and offer more reliable and reproducible conclusions.

Results

In this work, we propose a post-acquisition strategy (PARSEC) to improve metabolomics data comparability that consists in a three-step workflow starting from the combined extraction of raw data from the different studies or cohorts analyzed, through standardization, to the filtering of features based on analytical quality criteria. This workflow was applied to two case studies to evaluate the performance of the developed correction approach and to compare it with the classically used locally estimated scatterplot smoothing (LOESS) method. The PARSEC strategy allowed reducing the inter-group variability, and producing a more homogeneous sample distribution. In addition, results showed an improvement in the comparability of the data in both case studies, allowing biological information initially masked by unwanted sources of variability to be revealed more clearly than with the LOESS method.

Significance

The proposed post-acquisition correction strategy, which combines batch-wise standardization and mixed modeling, enhances data comparability and scalability for metabolomics studies. By addressing both batch and group effects, this approach minimizes the influence of analytical conditions while preserving biological variability. Therefore, it offers a valuable tool for harmonizing datasets across different studies or cohorts without common long-term quality control samples.