PT  - JOURNAL ARTICLE
AU  - Kip D. Zimmerman
AU  - Mark A. Espeland
AU  - Carl D. Langefeld
TI  - Pseudoreplication bias in single-cell studies; a practical solution
AID  - 10.1101/2020.01.15.906248
DP  - 2020 Jan 01
TA  - bioRxiv
PG  - 2020.01.15.906248
4099  - http://biorxiv.org/content/early/2020/01/15/2020.01.15.906248.short
4100  - http://biorxiv.org/content/early/2020/01/15/2020.01.15.906248.full
AB  - Cells from the same individual share a common genetic and environmental background and are not independent, therefore they are subsamples or pseudoreplicates. Empirically, we show this dependence across a range of cell types. Thus, single-cell data have a hierarchical structure that current single-cell methods do not address and subsequently the application of such tools leads to biased inference and reduced robustness and reproducibility. When properly simulating the hierarchical structure of single-cell data, commonly applied single-cell differential expression analysis tools exhibit highly inflated type I error rates, particularly when applied together with a batch effect correction for individual as a means of accounting for within sample correlation. As single-cell experiments increase in size and frequency, we propose applying generalized linear mixed models that include random effects for differences among persons to properly account for the correlation structure that exists among measures from cells within an individual.