Hierarchical mixed-effects model for batch effect correction in primary-tissue single-cell metabolomics by mass spectrometry.

Single-cell mass spectrometry (MS) metabolomics is a powerful tool for profiling metabolites in primary tissues. However, daily instrumental drift and sample preparation introduce severe batch effects, which are highly intertwined with interindividual biological variation. Additionally, single cells are fully consumed after MS detection, making conventional quality control (QC) unavailable for error correction. Herein, we established a hierarchical mixed-effects correction strategy tailored to the three-level nested structure of analytical days, individual samples and single cells. A total of 1355 single cells were used for model training and 871 independent cells for validation using zebrafish liver samples. Variance decomposition revealed day-level batch effects accounted for 32.0% of total variation and sample variation for 23.3%. After correction, the average batch silhouette coefficient dropped from 0.333 to 0.081, while 98.4% of metabolite pairwise correlations were retained. Compared with ComBat, Harmony and z-score scaling, our method achieves a better trade-off between removing technical artifacts and preserving metabolic profiles. As a QC-free workflow, this approach is reliable and widely applicable to single-cell MS datasets from diverse tissues and analytical platforms.