Single-cell approaches can reveal which cell types express a gene and how cellular states differ within a tissue. They are especially attractive in medical genetics because a disease-associated variant may affect only a subset of cells or a specific developmental state. The resulting resolution is valuable, but it does not remove the need for careful study design.
Use atlases as references, not universal baselines
The 2022 Tabula Sapiens study created a multi-organ human transcriptomic atlas and demonstrated substantial diversity across cell types and tissues. Such resources can help generate hypotheses about where a disease mechanism may act. They do not establish that every cell state in a patient sample has a direct equivalent in the reference, or that reference expression predicts a clinical phenotype. Read the Tabula Sapiens study.
Tissue collection, dissociation, preservation and sampling can influence which cells are recovered. A missing population may reflect sampling rather than biological absence. Annotation should therefore incorporate established markers, multiple lines of evidence and an explicit category for uncertain assignments.
Separate batch removal from biological discovery
The atlas-level integration benchmark by Luecken and colleagues evaluated methods according to both batch correction and preservation of biological variation. This distinction matters: a visually well-mixed embedding can conceal meaningful differences, while residual separation may reflect technical effects. Integration quality cannot be judged from an attractive plot alone. Read the integration benchmark.
Where possible, avoid designing a study in which every case is processed in one batch and every control in another. No computational method can reliably recover all information when disease status and technical conditions are inseparable. Use sensitivity analyses to assess whether conclusions depend on one preprocessing or integration choice.
Keep the unit of inference visible
- Report the number of participants as well as the number of cells.
- Account for cells nested within participants when comparing disease groups.
- Distinguish changes in cell abundance from changes within a cell type.
- Validate key findings in independent samples or with another measurement.
A large cell count can improve characterization of each sample while leaving uncertainty about population-level effects. For medical-genetics interpretation, a strong study links a reproducible cellular observation to a plausible molecular mechanism and makes the limits of that link explicit. Discovery of a disease-associated cell state is the beginning of that argument, not its completion.