The Target Changes Shape

Imatinib fits an inactive ABL kinase state with an extended pocket. Dasatinib fits a more active-like state with a different pocket geometry. Both structures are experimentally measured. Choosing one as a design reference therefore chooses which binding hypothesis a model can explore.

This walkthrough uses published structures, not states generated or measured by Crane. ABL provides the main comparison. A shorter cryptic-pocket example then asks a different question: does observing a pocket with a ligand establish that it is accessible before binding?

Imatinib binds an inactive ABL kinase state, termed DFG-out, captured in PDB 1IEP. The DFG phenylalanine moves away from the ATP-site geometry and leaves room for imatinib's extended pose.

Dasatinib binds an active-like ABL1 state, termed DFG-in, captured in PDB 2GQG. The target is recognizably the same kinase, but the pocket now supports another inhibitor geometry and development strategy.

These measured structures do not reduce binding to a rigid lock and key. Studies of ABL1 support a combination of conformational selection and induced fit, with local chemistry influencing the DFG switch. The useful lesson is simpler: choosing a target structure also chooses which inhibitor hypothesis the design can see.

The M182T TEM-1 structure PDB 1JWP shows helices 11 and 12 packed against the protein core. From that view alone, the cryptic allosteric site is easy to miss.

In PDB 1PZO, a ligand separates those helices and exposes a hydrophobic pocket about 16 angstrom from the active site. The experimental study connected that structural change to inhibition.

The structure proves the pocket can exist. Because it is ligand bound, it does not establish how often unbound TEM-1 visits that state, whether the ligand selected or induced it, or how quickly the transition occurs.

Modern ensemble methods can propose alternative conformations much faster than an experimental campaign can measure them. That is useful when it exposes a missed pocket, challenges a preferred mechanism, or identifies the state measurement that deserves a more expensive test.

For the ABL comparison, the useful judgment is whether a proposed molecule remains plausible across the relevant measured states. If the conclusion depends entirely on one conformation, that dependence belongs in the design brief and the experimental question.

An ensemble suggests where to look. It does not establish a thermodynamic population, and it does not show a kinetic path between states. A 2026 cryptic-pocket benchmark found that current AI and physics-based approaches can recover some directional trends, while absolute opening probabilities remain unreliable, especially for rare states.

For ABL1, the next evidence depends on the inhibitor hypothesis. State-resolved binding measurements, resistance mutations, and cellular activity can test whether the proposed state preference matters beyond structural compatibility.

The cryptic-pocket example illustrates a broader evidence problem: observing a bound state does not establish its availability before binding. A program needs evidence matched to that question before treating a structural possibility as an accessible intervention.

With the relevant state defined, the question becomes more local: which surface would support the intended mechanism? The HER2 binding-site comparison examines that choice on a measured receptor complex.