Dissecting the Influence of Batch Effects on Immunomics Pattern Discovery in High‐Dimensional Mass Cytometry

High-dimensional cytometry, such as mass cytometry (CyTOF), measures protein expression in single cells. When paired with AI-enhanced data analytics, it facilitates the discovery of immune biomarkers that can assist in diagnosing and treating immune-related diseases. However, fluctuating instrument readouts, known as batch effects, can obscure biological patterns. To enhance our previously reported EPIC immune atlas platform, we developed the ImmuneMapBuilder app to integrate, annotate, and cluster CyTOF data. To evaluate biological stratification and batch effects in clustering results, we created a two-step visualization method, called Group Similarity Analysis (GSA). First, multidimensional immune profiles are projected into two-dimensional embeddings to reveal similarities between samples. Second, silhouette analysis of embedding coordinates quantifies cohesion within technical and biological groups. We illustrate the scope and effectiveness of GSA using six batch normalization methods across three datasets, demonstrating that batch-effect correction enhances the separation of age groups and tissue types. The goals of batch normalization are increased biological and decreased technical GSA scores, both of which correlate with decreased Earth Mover's Distance (EMD), an alignment indicator of signal distributions. To dissect the contributions of individual marker misalignments to batch effects, we introduce a mix-and-match strategy that combines normalized and raw channels. GSA also helps to compare meta-clustering outputs. In a case study, we confirmed the significance of CD62L expression in T cells for age-based stratification. Overall, GSA is a versatile method to evaluate clustering results from cytometry data.