Methodological exploration of cross-platform single-cell comparison and joint exploratory analysis of IgG4-ROD

Abstract

Using the same reproducible pipeline, this study examined how merging and filtering strategies in cross-platform single-cell comparison alter functional enrichment conclusions, and accordingly proposed actionable practical recommendations. Cases were 10x data from four IgG4-ROD lacrimal glands, and controls were three SORT-seq technical plates of healthy lacrimal glands from the same donor. Because platform and disease were completely confounded and the controls were technical replicates, this article reports only methodological conclusions and makes no biological inferences. Methodologically, we defined four differential gene criteria from broad to strict. We used within-sample proportions as the analysis unit, exhaustively enumerated all 35 possible exact permutations, and applied Benjamini–Hochberg (BH) correction. Because only processed matrices were available for controls, we used sample-level permutation instead of DESeq2. We also performed doublet detection, ambient RNA quantification, and linear correction. Main findings: before merging and without taking a gene intersection, 53.7% (9,254/17,241) of genes were detected in fewer than 5% of cells in at least one group, and enrichment was dominated by cytoskeletal/neural/ion-related terms. After taking the intersection and filtering, enrichment shifted to RNA splicing and immune-related terms, suggesting that the filtering criterion was more critical than filtering stringency. Proportion tests showed no significant between-group differences in permutation tests, with the smallest BH-adjusted value of 0.1286. For single-gene sample-level permutation, the smallest attainable P was 0.0556, and the smallest BH q was 0.133 (above the lower limit by 0.0556). Beyond the single-cell data, serum miRNA analysis was also incorporated into the joint exploratory analysis (methodological replication only, not a biological finding): under the primary criterion (excluding zero-variance probes) all 15 miRNAs were non-significant (BH 0.0651–0.0711); the all-probe value BH P = 0.0476 reflects step-up accumulation of same-order raw P values and is not presented as a finding.