Comparison of Batch Effect Removal Methods for High Dimensional Mass Cytometry Data

Mass cytometry is one of the most popular single-cell technology that can measure over 50 markers simultaneously for millions of cells. Due to the high-dimensional nature of the dataset, manual analysis is difficult. Furthermore, during biological sample preparation, some problems may arise that need to be dealt with. One of the problems is a batch effect that can be introduced to the data because of experimental design or different experimental settings and techniques used. There are several bioinformatical solutions to correct the batch effect. Some of them require technical replicates (CytofBatchAdjust, CytoNorm and CytofRUV), others can work for a limited number of cells only (iMUBAC). An interesting aspect is how the batch correction method affects the results in terms of the number and quality of identified cell groups and to what extent the batch effect was removed. In the study, the two batch effect methods were compared, that do not require technical replicates, cyCombine and iMUBAC, applied to a real dataset with over 2 million bronchoalveolar lavage cells. Results were presented with the original mISO plots. Cells were clustered based on the original and corrected marker profiles with the PARC algorithm. After the correction, the number of clusters decreased from 24 to 22 (iMUBAC) and 18 (cyCombine). The homogeneity of clusters expressed as an effect size measure increased after the cyCombine correction (p-value = 4.38*10–7) in contrast to iMUBAC (p-value = 0.4628). The results indicate the superiority of cyCombine over iMUBAC for the real dataset if the within-cluster marker profile similarity is considered.