Documentation

Docs / Guides / Understanding Scores

Understanding LatticeZero Scores

A common first question after docking: "Is this score good?" This guide explains what LatticeZero scores mean, how to interpret them, and what they can't tell you.

The Quick Answer

  • Lower scores = stronger predicted binding (scores are negative; -6.0 is better than -4.0)
  • Scores are relative rankings, not absolute binding energies
  • Compare scores within the same target only - ACE scores can't be compared to CDK2 scores
  • A score of -5.0 doesn't mean 5 kcal/mol - it's a composite physics index

Score Components

Your dock score is a weighted combination of physics terms. The exact weight of each term depends on the scoring profile used for that target.

Term What It Measures Good Values
E_vdw Net van der Waals (shape fit + clashes) More negative = better fit
E_disp Van der Waals attraction (shape complementarity) More negative = better
E_rep Steric clashes (atoms too close) Closer to zero = fewer clashes
E_hb / E_hbq Hydrogen bond quality and count More negative = more/better H-bonds
E_coul Electrostatic complementarity More negative = better charge match
insideFrac Fraction of ligand inside the pocket Closer to 1.0 = better burial
aromaticBurial Aromatic ring burial in hydrophobic regions Higher = more aromatic contacts
triad Catalytic triad engagement (proteases) Higher = better catalytic interaction
ZBG Zinc binding group score (metalloproteases only) Higher = better zinc coordination
strain Conformational strain energy of docked pose Lower = less-strained pose

Each target has an optimized scoring profile that weights these terms differently based on what matters most for that binding pocket. See Scoring Profiles for details.

:::note A score near zero or positive often indicates the compound doesn't fit in the pocket at all - it's clashing severely or sitting outside the binding site. :::

PQS (Pose Quality Score) Badges

The colored badge on each result is the Pose Quality Score - a percentile-based quality assessment calibrated against the DEKOIS2 benchmark set for that target.

Badge Meaning What to Do
🟢 EXCELLENT Top 5% of scores for this target Strong hit - investigate further, analyze interactions
🟡 GOOD Top 25% Promising - consider analog synthesis
🟠 MODERATE Top 50% Borderline - inspect physics breakdown for weak terms
🔴 POOR Bottom 50% Unlikely to bind well
TRASH Severe clashes or outside pocket Discard - compound doesn't fit the pocket geometry

PQS thresholds are calibrated per-target and per-tier (Standard vs Elite scoring). A "Good" score in Standard mode maps to a different raw score than "Good" in Elite mode.

Scoring Profiles

Each target uses an optimized scoring profile (e.g., ace_metal_v5 for ACE, cdk2_hinge_v3 for CDK2). The profile determines:

  • Which physics terms are included
  • How much weight each term gets
  • Whether special terms (ZBG for metalloproteases, triad for serine proteases) are active

The profile name and tier are shown in the results panel under "Scored with: [profile]". Higher-tier profiles (Platinum > Gold > Silver > Internal) give more reliable rankings.

Benchmark Validation

The Benchmarks page shows the AUC (Area Under the ROC Curve) for each target - a measure of how well the scoring separates known drugs from non-drugs:

Tier AUC Meaning
Platinum ≥ 0.90 Excellent discrimination - highly reliable
Gold ≥ 0.70 Good discrimination - reliable for lead discovery
Silver ≥ 0.55 Some signal - use with caution
Internal ≤ 0.55 Not yet validated - experimental only

A Platinum-tier target with AUC 0.95 means the scoring correctly ranks 95% of known drugs above non-drugs.

What Scores Can't Tell You

Scores are powerful but have real limitations:

  • No bioavailability - a compound might bind perfectly but never reach the target in vivo
  • No selectivity - a good score on one target doesn't mean it won't bind others
  • No ADMET - absorption, distribution, metabolism, excretion, and toxicity are separate questions
  • No absolute affinity - scores rank compounds relative to each other, they don't predict Kd values
  • Rigid receptor - docking assumes the receptor doesn't move; induced-fit effects aren't captured

For lead candidates, always follow up with experimental confirmation. Use the scores to prioritize, not to certify.

Reading the Physics Breakdown

The physics breakdown panel (available in IsoPose and IsoScore results) shows each term's contribution:

  • Bars represent magnitude (longer = bigger contribution)
  • Green bars (negative terms) improve the score
  • Red bars (positive terms) penalize the score
  • Zero bars mean the term is absent or inactive for this target

If E_rep is the largest single bar, your compound has steric clashes - consider modifying the clash region. If E_hbq is small, you may be missing a hydrogen bond that a key residue expects.


What QuickDock Is and Is Not

QuickDock is a rapid virtual screening tool for evaluating single compounds against validated protein targets. Here's how to use it responsibly.

What QuickDock Does

  • Physics-based docking with HBQ-optimized scoring (van der Waals, H-bonds, electrostatics, desolvation)
  • Pose quality control with target-class-aware checks (kinase, metalloenzyme, PPI, default)
  • Automatic escalation to higher-accuracy GA sampling when initial poses fail QC
  • Conformer ensemble docking for flexible compounds that resist single-conformer docking
  • Calibrated per-target thresholds for STRONG / MODERATE / WEAK verdict labels
  • Reference receptor alignment for targets requiring structural superposition

What QuickDock Cannot Do

  • Predict binding affinity in absolute terms. Scores are relative rankings, not Kd values.
  • Guarantee selectivity. A compound scoring well on one target may also bind others.
  • Replace experimental validation. Docking is a computational filter, not a substitute for assay data.
  • Model receptor flexibility. The receptor is treated as rigid; induced-fit effects are not captured.
  • Assess ADMET properties. Absorption, distribution, metabolism, excretion, and toxicity require separate tools.
  • Score covalent binders. The current scoring does not model covalent bond formation.
  • Handle very large ligands. Compounds with >50 heavy atoms or >12 rotatable bonds may not dock reliably.

Interpretation Checklist

Before acting on a QuickDock result:

  1. Check the Pose QC badge — FAIL means the pose has steric clashes or is outside the pocket.
  2. Check the verdict — STRONG means the score is in the top percentile for that target's benchmark set.
  3. Look at the target class — kinase hinge H-bonds, metalloenzyme coordination, and PPI burial depth each have different QC thresholds.
  4. Compare against known binders — dock a known drug for the same target as a positive control.
  5. Consider multiple targets — if a compound scores STRONG on both your target and an unrelated one, it may be a promiscuous binder.
  6. Follow up experimentally — virtual hits should always be confirmed with biochemical assays.