Geinium
Home › Ligand parameters

Ligand parameters

How the ligand receives GAFF2 atom types and AM1-BCC charges, how to read the warnings, and how to use parameters made elsewhere.

What is computed

The protein force field covers amino acids; the ligand needs parameters of its own. Geinium computes them with the GAFF2 general force field and AM1-BCC charges, in its own code and without starting any other program. The chain of steps reproduces that of an established reference implementation stage by stage:

  1. Perception of bonds, bond orders, aromaticity and formal charges from the coordinates and the file.
  2. Protonation at the pH (see Protonation at pH).
  3. GAFF2 atom typing.
  4. A semi-empirical AM1 calculation (self-consistent field and geometry optimisation) with Geinium's own AM1 code, for closed-shell molecules of H, C, N, O, F, P, S, Cl, Br and I.
  5. Bond charge corrections (BCC) applied to the AM1 charges.
  6. Missing bonded parameters taken by analogy, with a penalty for each.
  7. The ligand topology and its position restraints.
The two-dimensional structure of tricin 7-O-glucoside as perceived from its docked pose. Look at the wedge and hash bonds of the glucose ring: the stereochemistry was read from the three-dimensional coordinates, which is a quick check that perception found the molecule you docked.
The two-dimensional structure of tricin 7-O-glucoside as perceived from its docked pose. Look at the wedge and hash bonds of the glucose ring: the stereochemistry was read from the three-dimensional coordinates, which is a quick check that perception found the molecule you docked.

Name and net charge

SettingDefaultMeaning and when to change it
"Ligand residue" nameLIGThe residue name of the ligand in the topology and the results, 1 to 4 characters. Change it only to keep the name of your own files; the analyses find the ligand either way.
"Net charge"auto"auto" takes the sum of the formal charges after the pH rules. An integer forces that charge. Set it only when you know the charge and the rules disagree; a forced charge that does not match the hydrogens present gives wrong charges.

On Run MD and the Batch page the net charge is always "auto"; the System Builder shows both fields.

Penalties and warnings

When a bond, angle or torsion of the ligand has no exact GAFF2 parameter, one is taken from a similar chemical environment and given a penalty that grows with the difference. Above 50 a note is written; above 150 a warning, for example:

parameter by analogy with a high penalty 232.0 (> 150): dihedral ... not validated for this chemistry

High penalties are common in drug-like molecules: 16 of 44 molecules of the development set had a maximum above 150. A term with no substitute at all receives an estimated parameter and a warning; this is the only deliberate difference from the reference chain. The report states the result in the "Ligand parameters" row, for example "GAFF2 with AM1-BCC charges (net charge +1 at pH 7.4, input +0; 2 parameters by analogy with a penalty > 150)".

What to do with a warning: look at the torsion it names. If it lies in a part of the ligand that matters for binding, check its behaviour in the torsion profile of the results (see Ligand analyses), and for a publication consider fitting that torsion to quantum-mechanical data and importing the parameters as described below.

Parameters from elsewhere

Parameters made with another method can be used on the System Builder page: set "Ligand parameters" to "Existing topology (.itp)", choose the .itp file as "Ligand" and its coordinates, in the frame of the receptor, as "Ligand coordinates" (.gro or .pdb). In a project file the key inputs.ligand_topology does the same; it skips the ligand stage and the contact check, so make sure the coordinates are the docked pose. GAFF2 parameters are combined with the AMBER protein force fields only. The charge methods "gas" and "input" (in place of "bcc") are available from the project file only.

How long it takes

A ligand of 26 atoms takes about 3 to 5 s. In an older measurement of 44 drug-like molecules on eight threads of a shared machine the total was 451 s and the largest molecule took 69 s; later optimisation brought the test set to 0.6 to 30 s per molecule. The whole preparation of a 4DAJ complex took 16 s. Large, flexible molecules can take a minute or more.

Accuracy

Against the reference implementation: AM1 single points of 88 test cases differed by at most 2.8e-8 kcal/mol in the heat of formation, with Mulliken charges within 5.0e-4 e; atom types showed no mismatch over 98 molecules; and end to end from SDF, 44 drug-like molecules gave identical topology lines with charges within 1.0e-3 e. Details are in Accuracy and validation. These figures say that the method is implemented faithfully; they say nothing about how well GAFF2 itself describes a given molecule.

Limits