Benchmark · NOBBLE

Benchmark panel

Seven pairs with published crystallographic ternaries, run first in every campaign. This is the page a sceptical reader opens first, so it reports what each pair tests, what the pipeline predicted and where it disagreed with the deposited structure. Every accession, chain and ligand code below was resolved programmatically at build time: none is hardcoded.

GSPT1 and BRD4 lead, as community standards. BRD4 appears twice, on VHL and on CRBN, which tests an E3 swap on the same substrate.

Construct comparison: does the whole domain recover the pose?

Six of the seven pairs submitted their substrate as a 15-mer cut out of a folded domain, and all six missed the crystal pose by 12 to 29 angstrom. The one whose epitope is linear, IKZF1, recovered at 5.4. This re-runs the panel with each domain-mediated substrate submitted as its whole domain and changes nothing else.

IKZF1 is the control. It keeps its 15-mer in both campaigns, so if its recovery moves, something other than the construct changed and nothing below can be attributed to the construct.

Verdict not concluded: the IKZF1 control has no number in both campaigns, so a difference in the other pairs cannot yet be attributed to the construct control: not yet run

Pair	Substrate	15-mer RMSD	Domain RMSD	Change	Verdict
BRD4 / CRBN	folded_domain	20.399	—	—	incomplete: one campaign has no recovery number yet
BRD4 / VHL	folded_domain	19.235	—	—	incomplete: one campaign has no recovery number yet
CSNK1A1 / CRBN	folded_domain	12.24	—	—	incomplete: one campaign has no recovery number yet
GSPT1 / CRBN	folded_domain	29.221	—	—	incomplete: one campaign has no recovery number yet
IKZF1 / CRBN control	peptide_15mer	5.357	—	—	incomplete: one campaign has no recovery number yet
RBM39 / DCAF15	folded_domain	6.36	—	—	incomplete: one campaign has no recovery number yet

All RMSD values are peptide backbone, superposed on the receptor, against the same deposited structures in both campaigns.

1. GSPT1 / CRBN

5HXB

Canonical glycine-degron glue benchmark: the highest-value single test in the panel.

Bridge %ile 53.9 3/10 seeds bridged

Δ-ipTM -0.756 median 0.024

PLIP match 1.00 glue-degron contacts 1

Convergence 0.000 mean 19.2 A

Docked 14/15 1 undocked

Lysines 4 not a validated signal

Peptide benchmark_83c557c3dc
seed(s) [4] collapsed well below the median of their condition and set the mandated minimum, so delta_iptm here reports worst-case seed behaviour rather than interface confidence. delta_iptm_median is reported alongside

Open in Pose Open in Interactions

2. BRD4 / VHL

5T35

Canonical bivalent ternary benchmark: tests that a linked degrader is handled as such rather than as a glue.

Bridge %ile 85.4 4/10 seeds bridged

Δ-ipTM -0.101 median -0.004

PLIP match 0.50 glue-degron contacts 0

Convergence 0.000 mean 18.3 A

Docked 15/15 0 undocked

Lysines 0 not a validated signal

Peptide benchmark_6e0551f38f

Open in Pose Open in Interactions

3. BRD4 / CRBN

6BNB

Second BRD4 arm on a different E3: tests an E3 swap on the same substrate.

Bridge %ile 76.2 4/5 seeds bridged

Δ-ipTM 0.002 median 0.080

PLIP match 0.60 glue-degron contacts 2

Convergence 0.000 mean 18.7 A

Docked 10/10 0 undocked

Lysines 4 not a validated signal

Peptide benchmark_c9b55ab831
conditioning disagreement not computable: one arm absent

Open in Pose Open in Interactions

4. IKZF1 / CRBN

6H0F

The founding IMiD neosubstrate observation.

Bridge %ile 46.4 10/10 seeds bridged

Δ-ipTM 0.316 median 0.136

PLIP match 0.80 glue-degron contacts 0

Convergence 0.728 mean 2.7 A

Docked 15/15 0 undocked

Lysines 1 not a validated signal

Peptide benchmark_888fd2c63e

Open in Pose Open in Interactions

5. CSNK1A1 / CRBN

5FQD

Same glue, different neosubstrate: tests specificity, not just recall.

Bridge %ile 44.7 10/10 seeds bridged

Δ-ipTM 0.081 median 0.046

PLIP match 0.80 glue-degron contacts 2

Convergence 0.356 mean 6.4 A

Docked 15/15 0 undocked

Lysines 1 not a validated signal

Peptide benchmark_b14fa633af

Open in Pose Open in Interactions

6. RBM39 / DCAF15

6Q0W

Aryl sulfonamide arm: exercises the DCAF15 token exception.

Bridge %ile — 0/0 seeds bridged

Δ-ipTM — median —

PLIP match — glue-degron contacts —

Convergence — —

Docked 5/5 0 undocked

Lysines — not a validated signal

Peptide benchmark_8be125d899
delta_iptm not computed: 5 binary and 0 ternary seeds present; conditioning disagreement not computable: one arm absent; no seed produced a glue bridge between receptor and peptide

Open in Pose Open in Interactions

7. CDK12 / DDB1

6TD3

Glue acting without a substrate receptor at all.

Queued

This pair has no completed jobs yet. The header chip shows campaign progress.