Documentation

Login
Docs / Scoring Profiles

Scoring Profiles

Scoring profiles control how LatticeZero weights the 14 physics-based scoring terms. Different target classes respond better to different weightings — a kinase prioritizes hinge hydrogen bonds, while a metalloprotease emphasizes metal coordination. Profiles let you tune this balance.

How Profiles Work

Every score in LatticeZero is computed as a weighted sum:

Total Score = w1*E_disp + w2*E_rep + w3*E_coul + w4*E_hbond + ... + w14*E_aromatic

A scoring profile defines the 14 weights (w1 through w14). The default profile uses equal-ish weights derived from physics. Target-class profiles use optimized weights validated against known actives and decoys.

Profile Tiers

Profiles are ranked by validation performance on DEKOIS2 benchmarks:

Tier AUC Threshold Badge Meaning
Platinum >= 0.90 Platinum Exceptional discrimination; validated on multiple seeds
Gold >= 0.80 Gold Strong discrimination; reliable for virtual screening
Silver >= 0.60 Silver Moderate discrimination; useful with caveats
Bronze < 0.60 Bronze Baseline performance; consider optimization

Tip: Platinum and Gold profiles have been validated with holdout cross-validation and bootstrap confidence intervals. They're ready for production use.

Available Target Classes

Kinases

Kinase profiles emphasize hinge region hydrogen bonds and hydrophobic gatekeeper interactions.

Profile Target AUC Tier Key Terms
SRC Kinase SRC 0.711 Silver strain (0.174), E_disp (0.143)
EGFR Kinase EGFR 0.82 Gold E_hbond (0.19), burial (0.16)

Nuclear Receptors

Nuclear receptor profiles weight aromatic burial and deep pocket shape complementarity.

Profile Target AUC Tier Key Terms
PPARG PPARg 0.934 Platinum aromaticBurial (5.3), depth (3.0)
ESR1 ERa 0.87 Gold E_disp (0.21), burial (0.18)

Proteases

Protease profiles emphasize catalytic residue interactions and substrate-like binding.

Profile Target AUC Tier Key Terms
ACE ACE 0.95 Platinum E_coul (0.22), metal (0.18)
CATL Cathepsin L 0.845 Gold E_coul (0.223), E_disp (0.15)

Viral Targets

Profile Target AUC Tier Key Terms
HIVRT HIV-RT 0.944 Platinum burial (-83.2), E_rep (0.12)

Reductases

Profile Target AUC Tier Key Terms
HMGR HMG-CoA Reductase 0.967 Platinum E_coul (0.31), depth (0.22)

Using Profiles

Selecting a Profile

When running IsoDock or IsoScore:

  1. In the Scoring Profile dropdown, browse available profiles
  2. Filter by target class or tier
  3. Select the profile closest to your target
  4. Run scoring as usual

Profile Recommendations

  • Known target class: Use the highest-tier profile for that class
  • Unknown target class: Start with the default profile, then try class-specific ones
  • Multiple candidates: Run scoring with 2-3 profiles and compare rankings

Custom Profiles

Creating a Profile

  1. Go to Scoring Profiles in the sidebar
  2. Click + New Profile
  3. Set weights for each of the 14 scoring terms
  4. Name and save your profile

Starting from a Template

  1. Select an existing profile as a starting point
  2. Click Duplicate
  3. Adjust weights as needed
  4. Save with a new name

Profile Optimization

Use the Optimizer to automatically tune weights for your specific target:

  1. Provide known actives and decoys
  2. The optimizer uses differential evolution to find weights that maximize AUC
  3. The optimized profile is validated with holdout cross-validation
  4. Save the result as a new profile

The 14 Scoring Terms

Each profile assigns a weight to these terms:

# Term Physical Meaning
1 E_disp Dispersion / van der Waals attraction
2 E_rep Steric repulsion
3 E_coul Electrostatic interactions
4 E_hbond Hydrogen bond strength
5 E_desolv Desolvation penalty
6 E_clash Close-contact penalties
7 burial Fraction of ligand surface buried
8 depth Ligand penetration into pocket
9 strain Internal ligand strain energy
10 aromaticBurial Aromatic ring burial fraction
11 hbondGeo H-bond geometry quality
12 E_aromatic Aromatic stacking interactions
13 contactArea Protein-ligand contact surface area
14 E_metal Metal coordination score

For detailed descriptions of each term, see the Physics Reference.

Best Practices

  1. Start with validated profiles — Don't manually tune weights unless you have validation data
  2. Match target class — A kinase profile will outperform a generic profile on kinases
  3. Validate on knowns — If you have known actives, score them first to verify the profile works for your target
  4. Use the Optimizer — For targets without a pre-built profile, automatic optimization is more reliable than manual tuning
  5. Compare tiers — If both Gold and Platinum profiles exist for your class, try both and compare
On this page
How Profiles Work Profile Tiers Available Target Classes Kinases Nuclear Receptors Proteases Viral Targets Reductases Using Profiles Selecting a Profile Profile Recommendations Custom Profiles Creating a Profile Starting from a Template Profile Optimization The 14 Scoring Terms Best Practices