Documentation

Docs / Demo & Benchmarks

Demo & Benchmarks

LatticeZero includes a built-in demo system powered by the DEKOIS2 benchmark suite. Run real molecular scoring experiments directly in your browser to evaluate platform performance and learn the tools.

The DEKOIS2 Benchmark

DEKOIS2 (Demanding Evaluation Kits for Objective In Silico Screening) is a widely-used benchmark suite for evaluating molecular docking and scoring methods. Each target includes:

  • Known active ligands — experimentally confirmed binders
  • Matched decoys — non-binders designed to have similar physicochemical properties to actives
  • Pre-docked poses — 3D coordinates for all ligands in the binding site

The challenge: can the scoring function distinguish actives from decoys?

Running the Demo

  1. Navigate to Try Demo in the sidebar (or go to /demo/run)
  2. Browse available targets — each shows the target name, protein class, and number of ligands
  3. Click a target to select it
  4. Choose Score Mode (IsoScore) or Dock Mode (IsoDock)
  5. Click Run — scoring begins immediately in your browser

Score Mode (IsoScore)

Rescores all pre-docked poses at ~4,000 ligands/sec. Best for:

  • Quick evaluation of scoring accuracy
  • Comparing different scoring profiles
  • Large target datasets

Dock Mode (IsoDock)

Performs full docking at ~3 sec/ligand. Best for:

  • Evaluating pose prediction quality
  • Testing on individual targets
  • Comparing docked vs. pre-docked poses

Available Demo Targets

The demo includes validated DEKOIS2 targets spanning multiple protein classes:

Target Protein Class Ligands Mode
ACE Angiotensin-converting enzyme Metalloprotease 40A + 1200D Score & Dock
EGFR Epidermal growth factor receptor Kinase 542A + 27420D Score & Dock
CDK2 Cyclin-dependent kinase 2 Kinase 219A + 6570D Score & Dock
HSP90 Heat shock protein 90 Chaperone 88A + 2640D Score & Dock
MDM2 Mouse double minute 2 PPI 78A + 2340D Score & Dock
P38a p38 MAP kinase alpha Kinase 137A + 4110D Score & Dock
UROK Urokinase Serine protease 55A + 1650D Score
FXa Factor Xa Serine protease 146A + 4380D Score
ESR1 Estrogen receptor alpha Nuclear receptor 135A + 4050D Score
AMPC AmpC beta-lactamase Hydrolase 48A + 1440D Score
PDE5 Phosphodiesterase 5 Phosphodiesterase 74A + 2220D Score
GBA Glucocerebrosidase Hydrolase 52A + 1560D Score

A = Actives, D = Decoys

Understanding Results

After scoring, the demo produces:

ROC Curve

The Receiver Operating Characteristic curve plots the true positive rate (sensitivity) against the false positive rate (1 - specificity) at every score threshold.

  • AUC = 1.0 — perfect discrimination (all actives scored above all decoys)
  • AUC = 0.5 — random (no better than coin flip)
  • AUC > 0.8 — strong performance suitable for virtual screening

Enrichment Metrics

Metric Description
AUC Area under the ROC curve (0-1)
EF1% Enrichment factor at 1% of ranked list
EF5% Enrichment factor at 5% of ranked list
BEDROC Boltzmann-Enhanced Discrimination of ROC

Score Distribution

A histogram showing the score distributions for actives vs. decoys. Good separation = good discrimination.

Validated Results

LatticeZero has been validated across the DEKOIS2 benchmark suite. Results with optimized profiles:

Target Default AUC Optimized AUC Tier
ACE 0.82 0.95 Platinum
HIVRT 0.51 0.94 Platinum
PPARG 0.71 0.93 Platinum
HMGR 0.72 0.97 Platinum
CATL 0.70 0.85 Gold
SRC 0.55 0.71 Silver

Note: Optimized results use the Optimizer with target-class-specific priors. Default AUC uses the standard equal-weight profile.

Using Demo Results

The demo is useful for:

  1. Evaluating the platform — See real performance before committing to a project
  2. Learning the tools — Practice with validated data where you know the answer
  3. Profile comparison — Test different scoring profiles against known benchmarks
  4. Conference presentations — Run live demos with reproducible results

Next Steps

  • Golden Packs — Download pre-validated target bundles for your own analysis
  • Scoring Profiles — Understand how profile selection affects results
  • Optimizer — Tune scoring weights for your specific targets