> ## Documentation Index
> Fetch the complete documentation index at: https://docs.revilico.bio/llms.txt
> Use this file to discover all available pages before exploring further.

# Cereblon Binder Affinity: Boltz-2 and Docking

> Boltz-2 affinity predictions and docking scores compared with measured affinities for 288 cereblon binders, the target of molecular glue degraders.

Cereblon (CRBN) is the E3 ligase receptor that thalidomide-like molecular glues bind to, and ranking its binders is a first step in designing new degraders. This page checks two of our predictors on 288 cereblon ligands with measured affinities: Boltz-2's affinity prediction and the docking engine's best-pose score.> [!NOTE]
>
> All predictions are ours. Measured values are pChEMBL values (the negative log of the potency) for each ligand, collected from the literature as Ki, Kd or IC50 measurements. The two predictors were run on exactly the same 288 ligands.## What the comparison showsBoth predictors find only a weak relationship with measured affinity on this set, and neither should be used on its own to rank cereblon binders. Over all 288 ligands the docking score correlates at about 0.32 and Boltz-2 at about 0.12. The picture changes with the assay: on the 73 Ki measurements the two are level on Pearson r at about 0.5, on the 87 IC50 measurements docking is ahead, and on the 128 Kd measurements Boltz-2 is weakly positive while docking is not. Cereblon binders tend to share a small core, and the measurements come from different assays, which both make ranking harder.## By assay typeSpearman ρ and Pearson r between each prediction and the measured pChEMBL value.

Metric


      

\<table><thead><tr><th>Name</th>
              \<th>Ligands</th></tr></thead>
          \<tbody /></table>


      

\**Bold:** a value for  that beats at least one other method in its row. *Italic, underlined:* the best value in the row.
        \**-**: no comparable result. Bold is a Boltz-2 value that beats the docking score. A positive value means tighter predicted binding goes with tighter measured binding.

## Prediction against measurementPick a group and a predictor. Boltz-2 predicts on the same scale as the measurements, so its panel has a 1:1 line; the docking score is in kcal/mol, so it does not.Boltz-2's predictions span a narrow range, 3.8 to 6.6, while the measurements run from 3.3 to 9.0. It under-predicts potency by 0.76 log units on average, and so it cannot separate weak binders from strong ones even where its ranking is right.## How the comparison was made> **Ligands and measurements**
>
> The set is 288 cereblon ligands, each with one pChEMBL value from the literature. The values come from three kinds of assay (Ki, Kd and IC50), which are not strictly comparable, so each is also shown on its own.> **Predictions**
>
> Boltz-2 predicts each ligand's complex with cereblon and reports an affinity as a predicted pIC50. The docking engine reports the affinity of its best pose in kcal/mol, which is negated here so that a higher value means tighter binding. The mean over poses, fixed as the alternative, gives about the same correlation (0.35).## Reading these results> **Mixed assays**
>
> Ki, Kd and IC50 values measure different things and depend on conditions, so a single correlation over all of them mixes real signal with assay differences. Treat the per-assay rows as the more careful comparison, and keep in mind that each has under 130 ligands.> **Similar ligands**
>
> Cereblon binders tend to share a small core, so the differences between them can be small relative to measurement noise and the predictors' own error. We have not measured the chemical diversity of this particular set.> **Possible overlap with training data**
>
> Boltz-2 and learned docking scores are trained on public protein–ligand data. We have not checked whether any of these cereblon ligands were part of it, so treat the results as agreement with public data, not a strict test on unseen ligands.> **No competing method is shown**
>
> This page reports our two predictors on their own. It makes no comparison with other methods.## Sources- **Measured affinities:** pChEMBL values for 288 cereblon ligands from the ChEMBL database.
- **Predictions:** Boltz-2 affinity predictions and our docking engine's best-pose scores for the same ligands.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.