Combining Motif Accessibility With RNA Expression

From a Sequence Motif to a Regulatory Hypothesis

An accessible chromatin peak can contain a DNA sequence resembling the binding motif of a transcription factor (TF). If that peak is linked to a gene, and the TF’s gene is expressed in the relevant cells, the observations can motivate a regulatory hypothesis: the TF may bind that region and influence the linked gene. The evidence is useful because it connects a candidate regulator, a possible cis-regulatory region, and a potential target.

The connection is not proof of binding or regulation. A motif match indicates sequence compatibility, not occupancy. Accessibility indicates that the region is available to regulatory proteins, not that a particular protein occupies it. RNA expression indicates that the TF transcript is present, not necessarily that active TF protein is present in the nucleus. Combining the signals makes a hypothesis more coherent, but does not remove these uncertainties.

The region must also be connected to a candidate target gene. Here, a linked peak means a peak assigned to that gene using an explicitly stated peak-to-gene procedure, such as the approaches covered in the earlier material on linking accessible peaks to gene expression. Such assignments can be uncertain; proximity alone does not establish a regulatory connection. Denote the selected peaks linked to target gene \(g\) by \(P_g\). A motif-accessibility summary for \(T\) and \(g\) should be calculated from the motif-bearing peaks in that set, not from an unspecified collection of all peaks.

Summarizing Motif Accessibility

For a transparent cell-level summary, first define \(B_{ip}=1\) if peak \(p\) is called accessible in cell \(i\), and \(B_{ip}=0\) otherwise. The call may be based on a prespecified rule for the observed accessibility counts. For the set of peaks linked to \(g\) that contain a match to \(T\)'s motif, define the motif-accessibility score as the fraction of those peaks called accessible:

This score is defined only when at least one peak in \(P_g\) has the motif (\(\sum_{p\in P_g}M_p>0\)). It ranges from 0 to 1 and describes the fraction of eligible motif-bearing peaks accessible in cell \(i\). It does not measure TF occupancy, binding strength, or regulatory effect. If peaks are weighted by a prespecified quantity, the weights and their interpretation must be reported; changing the peak set or weighting rule can change the score.

Before interpreting this score, check that the denominator is meaningful and that accessibility calls are comparable across cells. A score based on one motif-bearing peak is much less stable than one based on many peaks. The score can also change if peak calling, motif thresholds, or the peak-to-gene assignment changes. Report these choices and, where practical, assess whether the conclusion depends on them.

Motif Enrichment Is a Different Question

Motif enrichment asks whether a selected peak set contains more motif matches than expected relative to a defined background. It is not the same as asking whether those motif-bearing peaks are accessible in particular cells, and neither question establishes binding. An enrichment calculation is interpretable only if the tested peak set and background are appropriate for the selection process. For example, a background consisting of all genomic regions may be unsuitable if the selected peaks were already restricted to accessible chromatin.

A simple reference for enrichment is the expected number of motif-bearing peaks under random allocation. If a tested set contains \(K\) peaks drawn from a universe of \(N\) peaks, of which \(M\) contain the motif, then the expected motif count under random selection is \(KM/N\). An exact enrichment test can use the hypergeometric distribution, or a contingency-table method can compare motif presence in the tested set and its background. The random-selection assumption is often only an approximation: peak length, GC content, accessibility and related sequence properties can affect motif-match rates.

Combining Accessibility and TF RNA

A useful analysis keeps the evidence components visible rather than collapsing them into an unexplained score. For a defined cell population or condition, summarize motif accessibility across cells and summarize RNA expression of the matching TF gene in the same population. Consider whether both signals are present in the relevant biological context and whether their patterns are compatible with the proposed hypothesis. If comparing conditions, report the accessibility and RNA contrasts separately, with units and the direction of each contrast.

The two measurements need not have identical scales. A motif-accessibility fraction is between 0 and 1, while normalized RNA expression is on the RNA analysis’s scale. Adding them directly would not have a natural interpretation. Nor should their p-values simply be multiplied: the measurements may be dependent, and the tests may have different null hypotheses and assumptions. A joint evidence summary is most informative when it states what each component contributes and what it cannot establish.

The strongest interpretation depends on design and measurement. If RNA and accessibility come from paired cells, measurements can be aligned within the same cell or population, but paired observation does not establish direction of regulation. If modalities come from different cells or samples, the joint summary must respect that structure rather than implying cell-level co-occurrence. In either case, comparisons should use the same cell type or a design that accounts for changes in cell composition; otherwise, a difference in cell-type proportions can create apparent modality changes.

Assumptions, Diagnostics, and Alternatives

A defensible joint hypothesis depends on several choices. Motif scanning needs a specified reference sequence, motif database or model, and match threshold. The TF-to-motif assignment should acknowledge that similar motifs can be recognized by related TFs. The accessibility analysis needs a consistent peak definition and a stated rule for calling or summarizing accessibility. Peak-to-gene links need a stated method and should be treated as candidate links rather than confirmed targets. Finally, RNA measurements should be comparable across the cells or replicates being summarized.

Diagnostics should target these choices. Check how many eligible peaks contribute to each score, whether motif matches or peak links are concentrated in a few regions, and whether conclusions persist under reasonable alternative motif thresholds or background sets. Examine replicate-level summaries rather than relying only on pooled cells. Check that the compared populations have comparable cell identities and that technical differences do not align with condition. When many TFs, targets, or motif thresholds are examined, account for the resulting multiplicity in any inferential claims. A visually compelling pattern selected after screening many candidates is not equivalent to a prespecified confirmatory test.

The main alternatives answer different questions. Motif enrichment alone screens for sequence patterns in a set but does not use TF RNA or cell-specific accessibility. TF RNA alone identifies expressed candidate regulators but does not indicate which genomic regions are available. Joint evidence narrows candidates by combining these complementary observations. Direct occupancy assays, such as TF-specific chromatin immunoprecipitation or CUT&Tag, provide evidence closer to binding in the assayed context. Perturbation followed by a suitable readout can test functional consequences more directly, although its interpretation also depends on experimental design. These methods are complementary: a motif-plus-RNA hypothesis can help prioritize candidates for follow-up, but should not be reported as a binding result.

Common Mistakes and Limits

• Calling a motif match a binding event. A matching sequence may be inaccessible, unoccupied, or recognized by another TF with a similar motif. Describe it as a motif match, not confirmed occupancy.
• Interpreting TF RNA as TF activity. Transcript abundance does not measure translation, protein stability, localization, post-translational activation, or DNA binding.
• Using an arbitrary enrichment background. The background should reflect the peak universe from which the tested set could reasonably have been selected. A mismatched universe can make enrichment appear stronger or weaker.
• Ignoring uncertainty in peak-to-gene links. A linked peak is a candidate regulatory region, not automatically a causal enhancer for that gene.
• Counting cells as independent replicates. Cells from the same biological sample share sample-level conditions. For condition-level inference, biological replication and the analysis unit must be respected.
• Adding unlike evidence into an opaque score. A numerical sum of RNA and accessibility on different scales has no clear interpretation unless a justified construction and validation are supplied.