GeinDock Suite
Home › Constraint docking

Constraint docking

Stating a geometric condition that a pose must satisfy, how each kind of constraint reaches the search, and how to read a run whose condition was not met.

Sometimes you know something about the answer before you start. A series may be known to hinge on a hydrogen bond to one catalytic residue. A scaffold may be known to sit exactly where a crystallised analogue sits. A pharmacophore model may say that a basic nitrogen has to reach a particular acidic pocket. Constraint docking lets you state that knowledge and have it enforced, instead of hoping an unconstrained search stumbles onto it.

Select Docking › Docking Mode › Constraint Docking. Step 4 becomes Constraints.

The constraint editor with nothing defined
The editor on opening. The definition panel on the right stays disabled until there is a constraint to define.

How a constraint reaches the result

This is the part most worth understanding, because it decides what a constraint can and cannot promise. The search has no general restraint term, so a constraint acts through one of three mechanisms depending on what kind it is, and the dialog names the mechanism in force whenever you select a constraint.

Mechanism What it does Which kinds use it
Biased samplingThe constraint is turned into an attractive feature that steers the search itself towards satisfying it. This is the only mechanism that changes what gets sampled. It steers but it does not guarantee, so the poses are still measured afterwards.Sphere, distance, hydrogen bond
GenerativeThe pose is built already obeying the restraint and is then only locally optimised. Nothing that violates the restraint is ever generated.Core, or scaffold
SelectiveThe search runs normally and the reported poses are then filtered. This changes which pose you are shown, not what was explored, so the result stays directly comparable with an unconstrained run.Whole ligand RMSD, and the final check applied to all kinds
Why a scaffold constraint has to be generative. If a constrained conformer were handed to an ordinary global search, the first thing that search would do is throw the scaffold placement away and start looking elsewhere. A core constraint that is not built into the pose from the start is not a core constraint at all, so this kind is generated rather than filtered.

Adding a constraint

Add offers the five kinds, each with a one line description of what it asks for.

The list of constraint kinds
The five kinds of constraint, described by what they require rather than by how they are implemented.

Suggest from contacts is the quickest way to start. It reads the contacts the crystal ligand makes with the receptor, lists them with their distances, and offers to turn each one into either a hydrogen bond constraint or a distance constraint. That gives you a defensible starting point rather than a guess.

Suggesting constraints from the crystal contacts
Each contact the crystal ligand makes is offered as a ready made constraint, with the distance it was measured at.

The five kinds of constraint

Sphere

Requires the ligand to be positioned at a point in space, within a radius you set. The Inside control decides what has to be there.

Setting Meaning When to use it
Ligand centreThe ligand's own centre of geometry must lie within the radius.The default, and the honest reading of put the compound here.
Any atomSome qualifying ligand heavy atom must lie within the radius.Rarely what you want on its own, for the reason below.
A sphere on any atom constrains very little. A drug sized molecule spans ten to twelve Ångström, so a small sphere placed anywhere inside its envelope is satisfied by almost every orientation and almost every translation. A constraint that everything satisfies has restrained nothing, and a run that reports every pose as satisfying it has told you nothing. This is why the centre is the default for a newly created sphere.

Use grid box centre fills the coordinates from the box you already defined, and Pick atom from 3D view lets you click the location in the viewer. Ligand elements narrows which ligand atoms qualify, for example to nitrogen and oxygen only.

Distance

Requires a ligand heavy atom to lie between a minimum and a maximum distance of one named receptor atom. Both bounds matter. Without a lower bound, physically impossible contacts satisfy the test, so the default window runs from 2.4 to 4.0 Å, which is the range a real non bonded contact occupies.

H-bond

Requires a ligand nitrogen, oxygen or sulfur to come within the distance limit of a nitrogen, oxygen or sulfur of one of a set of named receptor residues. This is the kind to use when a series is known to depend on a specific polar contact, and naming several residues lets any of them satisfy it. The test is on heavy atom separation and does not include a donor angle, which the dialog states plainly so that you know what is being measured.

Two hydrogen bond constraints defined
Two constraints, each naming a residue and the distance measured in the crystal structure. The blue note reports which mechanism is in force for the selected constraint.

Core

Requires named scaffold atoms to sit within an RMSD tolerance of a template, usually a crystallised analogue. This is scaffold constrained docking, and it is the right tool when you are exploring substituents around a core whose binding mode is already known. Because it is generative, poses are built obeying it from the start.

RMSD

Requires the whole ligand to sit within a tolerance of a reference pose. Use it to test whether a compound can adopt a specific known binding mode at all, rather than to discover a new one.

Requiring some rather than all

By default every constraint must be satisfied. The Require control at the top of the dialog changes that to at least N of M.

Why this matters more than it looks. Constraints are a conjunction by default, so one unreachable anchor annihilates the whole run. A pharmacophore with three hydrogen bond anchors, one of which no pose can reach, reports that nothing satisfied it and discards the fact that poses satisfying the other two existed and were perfectly usable. Requiring two of three turns that bare refusal into a usable answer without weakening any individual constraint. Use it when your constraints are alternatives rather than a checklist.

Reading the result

Two columns appear in the pose table that no other mode fills in.

The pose table with the constraint columns
Each pose reports how many constraints were in force and how many it satisfies, and whether it therefore counts as a satisfying pose.
Column What it tells you
ConstraintsHow many constraints were in force for this run, and how many this pose meets, written as a fraction.
SatisfiedWhether this pose meets the requirement, taking the Require setting into account.

The job's verdict summarises the same thing for the run as a whole. A run in which no pose met the requirement is reported as a criterion that was not met, and it is not treated as a failure, because the poses it produced are real and are worth examining.

The diagram of a constrained pose
The two dimensional diagram of a satisfying pose. The constrained contacts are drawn along with everything else the pose makes, which is how you check that the constraint bought you a sensible binding mode rather than merely a satisfied test.
A run where every pose satisfies the constraint is a warning sign. It usually means the constraint was too loose to restrain anything, most often a sphere on any atom. Tighten the radius, switch to the ligand centre, or use a distance or hydrogen bond constraint to a named atom instead. A constraint is only informative if some poses could fail it. In a healthy constrained run, as in the figure above, some poses satisfy the requirement and some do not.

A suggested way to work

  1. Dock without constraints first. You need to know what the unconstrained search does before you can tell what constraining it changed.
  2. Start from the contacts you can see. Suggest from contacts builds constraints from the crystal ligand's own interactions, which is a defensible starting point rather than a guess.
  3. Constrain one thing at a time. Add a single constraint, run, and look at how many poses satisfied it. That tells you whether the constraint is doing any work.
  4. Loosen the requirement rather than the constraints. If several constraints together give nothing, set Require to a smaller number before you start widening tolerances. Weakening each individual condition makes them all less meaningful; requiring fewer of them keeps each one as strict as it was.