Chapter 3 Examples of QC Failure Diagnosis

The following sections describe failure patterns observed in the quality control analyses and provide guidance for tracing them to their potential root causes.

3.1 Scenario 1: RNA vs DNA (2.21) failure

A common failure signature is often detected in the RNA vs DNA (2.21) analysis, in which most cCREs cluster along the normalized RNA = DNA diagonal, indicating that few sequences have RNA abundance greater than expected based on their DNA abundance.

This pattern indicates a narrow dynamic range, but it does not in itself identify the underlying cause. Possible root problems include:

  1. Failure in RNA transcription or quantification
  2. Cell identity error
  3. The tested cCRE library contains few active elements

The next diagnostic step is to examine the separation between positive and negative controls using the Activity of controls (2.22) analysis.

  • If positive and negative controls show little to no separation, this points to a likely assay-wide failure in RNA transcription or quantification.
  • If positive controls are active but the test cCREs are not, the assay likely generated a valid activity readout, and the narrow dynamic range probably reflects the biological composition of the test library: the tested cCRE library contains few active elements. In this case, the pattern reflects the library’s biological composition rather than a technical failure.
  • If broadly active positive controls behave as expected but cell-type-specific positive controls do not, the result may indicate a cell identity error.

3.2 Scenario 2: BCs per cCRE (1.3) failure

A common failure signature in the association step is often observed in the BCs per cCRE (1.3) analysis, in which many cCREs are linked to too few BCs.

This pattern indicates reduced library complexity, but does not in itself reveal the underlying cause. Possible root problems include:

  1. Synthesis, cloning, and PCR biases
  2. Small pool of unique BCs
  3. Sequencing or alignment errors
  4. Overly stringent computational filters

The next diagnostic step is to examine sequencing depth using the BCs per cCRE by sequencing depth (1.6) analysis and the cCRE retention by sequencing depth (1.5) analysis.

  • If the number of retained cCREs or BCs per cCRE continues to increase with additional sequencing, the issue likely can be solved by additional sequencing.
  • If the curve has already plateaued, additional sequencing is unlikely to improve complexity, suggesting that the loss of complexity occurred before sequencing or during computational processing.

Additional diagnostics should then be used to distinguish among the possible causes:

  • Synthesis, cloning, and PCR biases: Examine cCRE sequence features, such as GC content, using the PCR bias - GC (1.7) analysis. Bias-driven underrepresentation of specific sequence classes can reduce cCRE-BC library complexity.
  • Small pool of unique BCs: Use the cCREs per BC (1.1) analysis to identify excessive BC promiscuity, which reduces effective library complexity by decreasing BC specificity.
  • Overly stringent computational filters: Compare the number of retained cCREs before and after filtering using Retained cCREs (1.4).
  • Sequencing or alignment errors: Evaluate standard quality measures, including sequencing depth, mapping quality, and the number of independent observations supporting cCRE-BC associations using the Reads per association (1.2) analysis.

3.3 Scenario 3: Poor correlation between replicates (2.13)

Another common failure signature is often observed in the Correlation between replicates (2.13) analysis, in which RNA-to-DNA ratios show low concordance across replicates.

Although this pattern may indicate reduced reproducibility, it can also result from a narrow dynamic range: when true activity varies little among cCREs, measurement noise constitutes a larger fraction of the observed variation and consequently reduces the correlation between replicates.

To determine whether the poor correlation is attributable to a narrow dynamic range, the next diagnostic step is to examine:

  • RNA vs DNA (2.21)
  • Activity distribution (2.4)
  • Activity of controls (2.22)

A clear high-activity tail, together with good separation between positive and negative controls, indicates that the experiment likely has sufficient dynamic range.

In this case, poor replicate correlation is more likely to reflect a root problem connected to reproducibility rather than dynamic range, such as:

  1. Insufficient constructs or DNA reads
  2. Batch effects
  3. Jackpotting of transcripts

These possible causes can be evaluated as follows:

  • Insufficient constructs or DNA reads: Examine Retained cCREs and BCs (2.1), BC retention by DNA/RNA sequencing depth (2.8), and cCRE retention by DNA/RNA sequencing depth (2.9).
  • Jackpotting of transcripts: Examine whether RNA reads are dominated by a small number of highly active cCREs using the Cumulative RNA reads (2.11) analysis.
  • Batch effects: If insufficient constructs or DNA reads and jackpotting are ruled out, batch effects become a leading possibility. Evaluate whether replicate differences align with experimental or technical variables.