Physics Reference
LatticeZero's scoring function consists of 14 physics-based terms that capture the key interactions governing protein-ligand binding. This page describes each term in detail: what it measures, how it's computed, and when it matters.
Scoring Function Overview
The total score is a weighted sum of all 14 terms:
Score = w1*E_disp + w2*E_rep + w3*E_coul + w4*E_hbond + w5*E_desolv
+ w6*E_clash + w7*burial + w8*depth + w9*strain
+ w10*aromaticBurial + w11*hbondGeo + w12*E_aromatic
+ w13*contactArea + w14*E_metal
Where w1 through w14 are the profile weights. More negative scores indicate stronger predicted binding.
Term 1: E_disp (Dispersion)
Physical meaning: Attractive van der Waals interactions (London dispersion forces).
What it captures: The favorable non-bonded attraction between all ligand-protein atom pairs at moderate distances. Dispersion is the primary driver of hydrophobic binding — nonpolar ligand groups nestling into hydrophobic pockets.
Computation: Sum of pairwise Lennard-Jones attractive terms (r^-6 component) between ligand atoms and pre-computed grid values.
Sign convention: Negative = favorable (attraction).
When it matters most: Hydrophobic binding sites, large nonpolar ligands, lipophilic drug candidates. Dominant for targets like nuclear receptors with deep hydrophobic pockets.
Typical range: -30 to 0
Term 2: E_rep (Repulsion)
Physical meaning: Short-range steric repulsion (Pauli exclusion).
What it captures: The energetic penalty when ligand atoms overlap with protein atoms. High repulsion indicates steric clashes — the ligand is physically colliding with the receptor.
Computation: Sum of pairwise Lennard-Jones repulsive terms (r^-12 component) with a softened potential to avoid infinities at very short distances.
Sign convention: Positive = unfavorable (clash).
When it matters most: Pose quality assessment. High E_rep values indicate the pose is physically unrealistic. Important for all targets but especially tight binding sites.
Typical range: 0 to 20
Term 3: E_coul (Electrostatics)
Physical meaning: Coulombic electrostatic interactions.
What it captures: Charge-charge interactions between ligand and protein atoms. Favorable when opposite charges meet (e.g., carboxylate near lysine) and unfavorable when like charges are close.
Computation: Distance-dependent Coulomb potential with distance-dependent dielectric screening to approximate solvent effects.
Sign convention: Negative = favorable (complementary charges).
When it matters most: Charged binding sites, catalytic residues, salt bridges. Critical for metalloproteases (ACE), serine proteases (FXa), and targets with charged active sites.
Typical range: -25 to 10
Term 4: E_hbond (Hydrogen Bonds)
Physical meaning: Hydrogen bond strength.
What it captures: The energetic contribution of hydrogen bonds between ligand and protein. Evaluates both the distance and directionality of donor-acceptor pairs.
Computation: Directional H-bond potential that considers donor-hydrogen-acceptor angle and heavy atom distance. Includes both classical (N-H...O, O-H...N) and weak (C-H...O) hydrogen bonds.
Sign convention: Negative = favorable (strong H-bonds).
When it matters most: Targets with polar binding sites, kinase hinge regions, catalytic triads. Essential for targets like EGFR where hinge hydrogen bonds anchor the ligand.
Typical range: -15 to 0
Term 5: E_desolv (Desolvation)
Physical meaning: Desolvation penalty.
What it captures: The thermodynamic cost of stripping water molecules from both the ligand and protein surface upon binding. Polar groups that were well-solvated in water pay a higher desolvation penalty.
Computation: Based on atomic solvation parameters and the change in solvent-accessible surface area upon binding.
Sign convention: Positive = unfavorable (penalty for burying polar groups).
When it matters most: Ligands with many polar groups binding to hydrophobic pockets (high penalty). Conversely, hydrophobic ligands in hydrophobic pockets have low desolvation costs.
Typical range: 0 to 15
Term 6: E_clash (Close-Contact Penalty)
Physical meaning: Hard-sphere clash detection.
What it captures: A binary-like penalty for atom pairs that violate hard-sphere radii. This is more aggressive than E_rep — it flags poses with severe, physically impossible overlaps.
Computation: Count of atom pairs closer than a threshold fraction (typically 0.6x) of the sum of van der Waals radii, multiplied by a penalty factor.
Sign convention: Positive = unfavorable (severe clashes).
When it matters most: Pose filtering. A high E_clash value means the pose is likely wrong regardless of other scores. Used as a quality gate in docking.
Typical range: 0 to 10
Term 7: burial (Ligand Burial Fraction)
Physical meaning: Fraction of ligand surface area buried upon binding.
What it captures: How deeply the ligand is enclosed by the protein. Higher burial means more of the ligand is in contact with the receptor, generally indicating better shape complementarity.
Computation: Ratio of ligand solvent-accessible surface area lost upon binding to total ligand SASA.
Sign convention: More negative = more buried = favorable.
When it matters most: Shape-driven binding, deep pockets, targets where size/fit matters (PPI hotspots, nuclear receptors). The dominant term for HIVRT optimization (weight -83.2 in the Platinum profile).
Typical range: -100 to 0
Term 8: depth (Pocket Penetration)
Physical meaning: How deep the ligand penetrates into the binding pocket.
What it captures: The distance of the ligand centroid from the pocket opening. Ligands that reach the bottom of deep pockets generally bind more tightly.
Computation: Distance-based metric from the ligand center of mass to the pocket surface, normalized by pocket depth.
Sign convention: More negative = deeper penetration = favorable.
When it matters most: Deep pockets (reductases, nuclear receptors). Less important for shallow surface pockets (PPI interfaces).
Typical range: -10 to 0
Term 9: strain (Internal Ligand Strain)
Physical meaning: Conformational strain energy of the bound ligand.
What it captures: The energetic cost for the ligand to adopt its bound conformation versus its relaxed (lowest-energy) conformation. Ligands that must twist, stretch, or compress to fit the pocket pay a strain penalty.
Computation: Torsional energy evaluation based on rotatable bond dihedral angles relative to preferred conformations.
Sign convention: Positive = unfavorable (high strain).
When it matters most: Flexible ligands, kinase inhibitors with multiple rotatable bonds. The key discriminating feature for the SRC kinase profile (weight 0.174). High strain should be a red flag regardless of other scores.
Typical range: 0 to 15
Term 10: aromaticBurial (Aromatic Ring Burial)
Physical meaning: Fraction of aromatic ring surface area buried in the pocket.
What it captures: How well aromatic rings in the ligand are enclosed by protein residues. Aromatic rings involved in pi-stacking, cation-pi, or hydrophobic interactions should be well buried.
Computation: SASA calculation restricted to aromatic ring atoms, comparing free vs. bound states.
Sign convention: More negative = more buried = favorable.
When it matters most: Targets with aromatic binding sites — nuclear receptors (PPARG profile weight 5.3), MDM2 (hydrophobic cleft), and targets with Phe/Trp/Tyr lining the pocket.
Typical range: -10 to 0
Term 11: hbondGeo (H-Bond Geometry Quality)
Physical meaning: Geometric quality of hydrogen bonds.
What it captures: Not just whether H-bonds exist (that's E_hbond), but how ideal their geometry is. Perfect H-bonds have linear donor-H-acceptor angles (180 degrees) and optimal heavy atom distances (2.8-3.0 A for N/O).
Computation: Geometric scoring based on deviation from ideal angles and distances, weighted by the number of H-bonds.
Sign convention: More negative = better geometry = favorable.
When it matters most: Targets where H-bond quality distinguishes actives from decoys — kinase hinge binding, serine protease oxyanion holes, HIV protease catalytic aspartates.
Typical range: -5 to 0
Term 12: E_aromatic (Aromatic Stacking)
Physical meaning: Pi-pi and cation-pi stacking interactions.
What it captures: The favorable energy from aromatic ring stacking (parallel-displaced or T-shaped) between ligand and protein aromatic residues (Phe, Tyr, Trp, His).
Computation: Distance and angle-dependent potential between aromatic ring centers, considering both face-to-face and edge-to-face orientations.
Sign convention: Negative = favorable (good stacking).
When it matters most: Aromatic-rich binding sites. Important for kinase ATP-binding sites (adenine stacking), nuclear receptors, and targets with aromatic cages.
Typical range: -8 to 0
Term 13: contactArea (Contact Surface Area)
Physical meaning: Total protein-ligand contact surface area.
What it captures: The absolute area of close contact between ligand and protein surfaces. Related to burial but measured as area rather than fraction. Larger contact area generally indicates more extensive interactions.
Computation: Connolly surface intersection calculation between ligand and protein at a probe distance.
Sign convention: More negative = more contact = favorable.
When it matters most: PPI inhibitors (large flat interfaces), allosteric sites, targets where binding area correlates with potency.
Typical range: -20 to 0
Term 14: E_metal (Metal Coordination)
Physical meaning: Metal ion coordination energy.
What it captures: The interaction between ligand atoms and catalytic or structural metal ions (typically Zn2+, Mg2+, Ca2+, Fe2+/3+) in the binding site. Strong metal coordination is critical for metalloenzyme inhibitors.
Computation: Distance and angle-dependent potential between known metal-coordinating ligand atoms (carboxylates, hydroxamates, thiols, phosphonates) and metal ions in the receptor.
Sign convention: Negative = favorable (strong coordination).
When it matters most: Metalloproteases (ACE, MMPs), carbonic anhydrase, histone deacetylases, and other metalloenzymes. The ACE Platinum profile assigns weight 0.18 to this term.
Typical range: -15 to 0
Summary Table
| # | Term | Type | Sign | Range | Key For |
|---|---|---|---|---|---|
| 1 | E_disp | Attraction | - | -30 to 0 | Hydrophobic contacts |
| 2 | E_rep | Repulsion | + | 0 to 20 | Steric clashes |
| 3 | E_coul | Electrostatic | -/+ | -25 to 10 | Charged sites |
| 4 | E_hbond | H-bond | - | -15 to 0 | Polar sites |
| 5 | E_desolv | Penalty | + | 0 to 15 | Polar in hydrophobic |
| 6 | E_clash | Penalty | + | 0 to 10 | Pose quality |
| 7 | burial | Geometry | - | -100 to 0 | Deep pockets |
| 8 | depth | Geometry | - | -10 to 0 | Pocket penetration |
| 9 | strain | Penalty | + | 0 to 15 | Flexible ligands |
| 10 | aromaticBurial | Geometry | - | -10 to 0 | Aromatic pockets |
| 11 | hbondGeo | Quality | - | -5 to 0 | H-bond geometry |
| 12 | E_aromatic | Stacking | - | -8 to 0 | Pi interactions |
| 13 | contactArea | Geometry | - | -20 to 0 | PPI, allosteric |
| 14 | E_metal | Coordination | - | -15 to 0 | Metalloenzymes |
Further Reading
- Scoring Profiles — How to weight these terms for your target
- Optimizer — Automatically find the best weights
- Glossary — Definitions of key terms