Map the Molecular System

A degrader can bind both of its partners and still fail to arrange them productively. For a discovery team, that is a costly ambiguity: should the next experiment test the compound, the protein constructs, or the assumed mechanism? A map of the complete molecular system helps separate those questions before the assay queue fills up.

The 5T35 crystal structure captures one such system. BRD4 binds the PROTAC MZ1, MZ1 binds VHL, and the two proteins create contacts that do not exist in either binary complex.

We compared a minimal assembly with a fuller one to ask a specific question: would including the surrounding partners change recovery of the interface? That comparison tests a modeling choice; it does not establish whether a new degrader will work.

This is a known-case reconstruction. The 2017 release of 5T35 falls before Boltz-2's structural training cutoff, and the reference is not a documented holdout. It cannot establish generalization to unseen systems. The value here is a traceable comparison of two assembly definitions.

Both specifications recovered the known protein interface in all five samples. DockQ measures agreement with an experimental protein interface on a scale from 0 to 1, with higher values indicating closer agreement. It is a reference-based accuracy metric, not a model confidence score.

The minimal specification contained BRD4 BD2, pVHL, and MZ1. The full specification added ElonginB and ElonginC. We used the same computational protocol for both, with five samples per system.

The five minimal-system structures scored 0.903 to 0.970 DockQ for the induced BRD4-pVHL interface. The five full-system structures scored 0.922 to 0.940. Each recovered 13 to 15 of the 15 native protein contacts.

MZ1 also returned to the experimental pose. Its symmetry-aware heavy-atom RMSD was 0.6 to 1.4 Å after aligning the relevant protein pocket. For the retrospective visual, we show full-system sample 1, the highest-DockQ full-system pose. It aligns to the measured assembly at 0.892 Å across 451 matched protein C-alpha atoms.

These are ten correlated samples from one known case and a shared seed, not ten independent validation cases. Both biological specifications recovered the interface; the full system was not automatically more accurate than the reduced one.

Here, adding the two partners did not change the central structural conclusion. A team can use that comparison to decide whether the extra assembly context is needed for its question, rather than assuming that a larger input is always a better one.

For a novel system without a measured reference, DockQ and reference ligand RMSD are unavailable. Confidence scores cannot substitute for them. The design brief must say which predicted contacts or assembly assumptions an experiment will test.

The map also helps choose where to intervene. Blocking a circulating protein, changing its production, and recruiting another partner lead to different programs. The next article in the reading guide examines that choice of molecular route.

The original 5T35 study did the work a structural model cannot. The authors measured binary and ternary binding, quantified cooperativity, mutated interface residues, and connected the structural interface to degradation behavior. Read the primary experimental study.

For a new system, we would use the predicted assembly to choose interface mutations and orthogonal assays. SPR can quantify binary and ternary binding, kinetics, and cooperativity; BLI can provide an orthogonal binding readout. Mutagenesis can test whether the proposed contacts matter. Cellular degradation can test whether the assembled geometry produces the intended phenotype.