Do DNA-Sequencing Counts Predict Potency in DEL Screening?

What we compared

For the 21 p38α hits carried forward into biochemical IC₅₀ testing, we also had the raw DNA-sequencing observation counts from the three replicate cBTE experiments run at the highest bait-DNA concentration. Plotting each compound’s average observation count against its measured potency (as pIC₅₀) makes the relationship — or lack of one — directly visible.

VPC00251 was among the most heavily observed compounds in the entire data set — roughly double the sequencing counts of VPC00628 — yet its biochemical potency (0.39 µM) was more than fifteen-fold weaker than VPC00628’s 25 nM. At the other end, VPC00257, with some of the lowest observation counts among the 21 compounds tested, was nonetheless the second most potent. Across the full set of 21 compounds, no correlation — not even a rough trend — was observed between average observation count and measured IC₅₀.

Why this is expected, not a flaw in the method

Enrichment in a trapping-based DEL screen like BTE or cBTE depends on two separable physical properties of a binding interaction: the dissociation constant (Kd), which governs how likely a library member is to be bound to the target at all at a given moment, and the off-rate, which governs whether that binding survives long enough — through dilution and compartmentalization — to actually be trapped and ligated. A compound can have an excellent Kd but a fast off-rate, meaning it binds well but also releases quickly, reducing how often it survives to be counted. A direct, one-to-one relationship between enrichment and either Kd or IC₅₀ is therefore not expected on mechanistic grounds — the sequencing count is a readout of trapping efficiency, which conflates both parameters, not of affinity alone.

There is also a statistical dimension. When screening a library of hundreds of millions of members, real trends that might be visible in a small, curated set of compounds can be washed out simply because of the scale and sampling variability involved. Weak signal-to-noise ratios in the count data for individual compounds, multiplied across a huge and chemically diverse hit list, make a clean dose-response-like relationship between counts and potency unlikely to emerge even where some underlying tendency exists.

What this means in practice

The practical takeaway is not that sequencing counts are meaningless — they remain essential for distinguishing genuine hits from background noise via the mathematical threshold, and for assessing reproducibility across replicate experiments. What they do not reliably do is rank hits by potency. A compound with a modest observation count sitting just above the significance threshold can turn out to be the most potent molecule in the set, while a compound with one of the highest counts can be a comparatively weak binder. The only way to know which is which is to test it: pull a chemically diverse spread of hits — not just the top of the count-ranked list — into an orthogonal biochemical or cellular assay, exactly as we did in selecting the 21 p38α compounds discussed here.