Single-Cell Multi-Omics for Mapping Regulatory Elements

Bulk chromatin and transcriptome assays report the average of every cell in the sample. In a tissue with six cell types, an enhancer active in 8% of cells may produce a signal indistinguishable from noise, or worse, a signal attributed to the dominant population. Single-cell methods solve this specific problem: they assign regulatory signal to cell identities.

What they do not do is establish that an element regulates a gene. Co-accessibility, correlation, and gene activity scores are all association measures, and sparsity makes them easy to over-read. Understanding where the method's authority ends is the difference between a useful discovery paper and a dataset that cannot support its conclusions.

This article maps the practical decision space. For the functional validation that must follow, see our pillar article and the enhancer–gene pair validation roadmap.

The multiome is preferable when the question is element–gene association, because matching both modalities in the same cell removes a major confound: cell-type composition differences that mimic regulatory correlation. It is not automatically better for element discovery, where deeper scATAC-seq in more cells may be worth more than paired transcriptomes.

Single-cell accessibility data are too sparse for per-cell peak calling. The field convention is aggregation into pseudobulk profiles — by cluster, by sample, or by both — followed by standard peak calling and differential analysis.

Three consequences that shape interpretation:

• • Cluster definitions determine peak sets. Merge or split a cluster and the element map changes. Report the clustering resolution and marker genes used.
• • Pseudo-replication is easy to commit. Treating cells as replicates inflates significance; the replicate unit is the donor or sample, not the cell.
• • Rare populations lose power. A 2% population may be identifiable but too sparse for reliable differential accessibility without targeted enrichment.

Three association strategies are used, and each should be reported for what it is:

Co-accessibility. Elements whose accessibility correlates across cells are inferred to share regulation. This captures covariation, which can reflect shared cell state rather than a direct relationship.

Gene activity scores. Accessibility near a gene is summed as a proxy for expression. The scores are useful for annotation and weak as evidence of regulation, because the weighting is arbitrary and proximity is a poor predictor of enhancer–gene pairing.

Peak–gene correlation in multiome data. The most informative of the three, particularly when restricted to cells of one type and corrected for expression level. Even here, correlation across cells may reflect a third variable — the cell state — rather than a regulatory interaction.

A useful discipline: generate candidate pairs computationally, then state explicitly which ones the study will test experimentally, and test at least a few. Papers that stop at correlation invite the criticism that the analysis produced hypotheses and called them findings.

• • Cell number per sample. Sparse modalities need thousands of cells per cluster for pseudobulk signal; plan for at least two donors per condition.
• • Nucleus quality. Frozen tissue and nuclei isolation protocols dominate data quality far more than sequencing depth. Pilot before scaling.
• • Multiplexing. Hashing or genotype-based demultiplexing reduces batch effects and is close to essential for cross-condition comparisons.
• • Depth allocation. More cells at lower depth usually beats fewer cells at high depth for element discovery; the reverse is true for splicing or allele-specific questions.

• • A homogeneous cell line. Bulk ATAC and RNA are cheaper, deeper, and easier to interpret.
• • A question about a specific element's function. Single-cell mapping nominates; it does not validate. CRISPR perturbation remains the decisive experiment.
• • Low-input clinical samples. If the sample yields a few hundred cells, bulk with spike-in normalization may be more informative than a sparse single-cell dataset.
• • Rare-event detection without enrichment. If the population of interest is under 1%, consider sorting or targeted enrichment first.

EDITGENE builds the bridge from single-cell maps to experiments: cell lines or iPSC-derived models matched to the identified cell type, CRISPRi/a-ready backgrounds for testing candidate elements, and reporter lines for quantitative follow-up.

Share the tissue context and the candidate elements; we will help scope a validation path. Contact EDITGENE to request a project consultation or quote.

Single-cell association measures are hypothesis-generating. Peak sets, clusters, and gene activity scores depend on analysis choices that should be reported alongside conclusions.

Cell-type identity in a dataset does not guarantee an equivalent in vitro model; where validation uses a derived cell type, marker and functional equivalence should be demonstrated.

• • Buenrostro JD, et al. Single-cell chromatin accessibility reveals principles of regulatory variation. Nature. 2015;523:486–490.
• • Ma S, et al. Chromatin potential identified by shared single-cell profiling of RNA and chromatin. Cell. 2020;183:1103–1116.
• • Stuart T, et al. Single-cell chromatin state analysis with Signac. Nature Methods. 2021;18:1333–1341.
• • Zhang K, et al. A single-cell atlas of chromatin accessibility in the human genome. Cell. 2021;184:5985–6001.
• • Murphy AE, Skene PJ. SPATAC-seq and the analysis of chromatin accessibility. Nature Methods. 2023;20:1–6.