Generative Glide Target Property Keywords

The target block of the agent section defines what kind of molecules the agents generate. Each entry is a property name and a target value that tells the agent how to optimize for that property. This topic describes the target value formats and lists the available property keywords. For the rest of the configuration file, see Generative Glide Config File Keywords.

A typical target block looks like this:

agent:
  target:
    Glide_score: min,auto
    mol_wt: [150.0, 500.0]
    alog_p: [-1.0, 4.0]
    tpsa: [50.0, 120.0]
    qed: [0.3, 0.7]

Target Value Formats

There are two kinds of target value: a range, which applies a flat reward, and a directional cutoff, which applies a graded reward.

Range target: `[min, max]`

A range target rewards any molecule whose property falls between min and max. The reward is flat: a molecule anywhere inside the range receives the full reward, and a molecule outside receives none. There is no preference for values nearer the center.

A range is the right choice for physicochemical property filters, such as molecular weight, polar surface area, and hydrogen-bond counts, where you want to constrain a property to a drug-like window without biasing the agent toward any particular value within it.

Directional target: `min,cutoff` or `max,cutoff`

A directional target rewards molecules that pass a cutoff in proportion to how far they exceed it, rather than with a flat reward. With min,cutoff, a lower value earns more reward; with max,cutoff, a higher value earns more. Molecules that do not pass the cutoff receive no reward.

Use a directional target for the primary scoring objective, typically Glide_score, where you want the agent to keep improving a value rather than just clear a threshold.

Automatic cutoff: `min,auto`

The min,auto target behaves like min,cutoff, but the cutoff is set automatically from the prior sample distribution using the auto_target_percentage keyword. This is the recommended setting for Glide_score, because it adapts to the score distribution of your target without requiring you to choose a cutoff.

Glide_score

The Glide_score target is always required. It represents the supervised model's predicted docking score, and it is the primary objective the agents optimize. By default it is set to min,auto, which optimizes for the lowest (most favorable) scores.

Property Keywords

The following properties can be used as targets. Specify each one with a range, [min, max], to reward molecules whose property falls within that window. They are computed from the SMILES during agent training, with no external tools required.

Keyword	Property	Description
`mol_wt`	Molecular weight	The molecular weight of the compound.
`fraction_csp3`	Fraction of sp³ carbons	The ratio of sp³-hybridized carbons to total carbons.
`alog_p`	AlogP	The octanol-water partition coefficient, a measure of lipophilicity.
`tpsa`	Topological polar surface area	An indicator of permeability.
`qed`	Drug-likeness (QED)	The quantitative estimate of drug-likeness, a score from 0 to 1. Target values must be within the range 0 to 1.
`num_rotatable_bonds`	Rotatable bonds	The number of rotatable bonds, a measure of molecular flexibility.
`num_chiral_centers`	Chiral centers	The number of defined chiral centers.
`num_undefined_chiral_centers`	Undefined chiral centers	The number of chiral centers with unspecified stereochemistry.
`ring_count`	Ring count	The total number of rings.
`num_aromatic_rings`	Aromatic rings	The number of aromatic rings.
`num_saturated_rings`	Saturated rings	The number of fully saturated rings.
`num_spiro_atoms`	Spiro atoms	The number of atoms at the junction of two rings that share a single atom.
`num_bridgehead_atoms`	Bridgehead atoms	The number of atoms shared by two rings in bicyclic systems.
`num_h_acceptors`	Hydrogen-bond acceptors	The number of hydrogen-bond acceptors.
`num_h_donors`	Hydrogen-bond donors	The number of hydrogen-bond donors.
`largest_ring_atom_count`	Largest ring size	The number of atoms in the largest ring.
`num_s_atoms`	Sulfur atoms	The number of sulfur atoms.
`num_alkyne`	Alkyne groups	The number of alkyne groups (carbon-carbon triple bonds).
`log_s`	LogS	The predicted aqueous solubility.
`num_nonspiro_cr4`	Non-spiro 4-membered rings	The number of non-spiro four-membered rings.
`num_five_five_rings`	Fused 5-5 rings	The number of fused or connected five-membered ring systems.
`complexity`	Structural complexity	A measure of the structural complexity of the molecule.
`longest_consecutive_rb_count`	Longest rotatable bond chain	The length of the longest chain of consecutive rotatable bonds.

Custom ML Model Targets

You can also optimize for a custom machine-learning model by listing it in the other_ml_model keyword of the agent section and adding its name to the target block with a range. The name in the target block must match the name used in other_ml_model. See Generative Glide Config File Keywords.