Documentation

Docs / Platform Workflow

Platform Workflow

A visual overview of the LatticeZero IsoPose scoring and validation pipeline — from target preparation through lead optimization and export.

IsoPose Platform Workflow End-to-end molecular scoring and validation pipeline

STEP 1 Upload PDB STEP 2 Select Target STEP 3 Configure Docking STEP 4 Run IsoPose (WebGPU-accelerated) FAST PATH IsoScore ~4,000 lig/sec OUTPUT Dock AUC + Poses results feed back STEP 5 Optimize Scoring Profile Autotune weights with unsupervised or labeled mode STEP 6 Validate Cross-validation + Y-scramble + bootstrap OPTIONAL LP+ Lead Optimization STEP 7 Export Results CSV, JSON, PDF LEGEND Primary pipeline Fast path Optional / feedback Optional step

Pipeline Stages

1. Upload PDB

Upload a receptor structure in PDB format, or choose from the pre-validated target library (46 GOLD_LOCKED targets available at launch).

2. Select Target

Choose the binding site and configure the receptor grid. For library targets, the binding site is pre-defined with validated pocket coordinates.

3. Configure Docking

Set docking parameters: exhaustiveness, scoring profile, and conformer options. FFT docking mode is available for PoseCompass-certified targets.

4. Run IsoPose (WebGPU)

IsoPose performs GPU-accelerated molecular docking directly in the browser using WebGPU compute shaders. Typical throughput: 50-200 ligands/second depending on hardware.

Fast Path — IsoScore: For pre-docked libraries, IsoScore provides ultra-fast rescoring at ~4,000 ligands/second without re-docking. This is useful for rapid virtual screening of large compound sets.

5. Optimize Scoring Profile

Use the Autotune optimizer to calibrate scoring weights for your target. Two modes are available:

  • Unsupervised mode — classifies target family and generates weights automatically
  • Labeled mode — uses known active/inactive labels to train a target-specific profile via logistic regression

Both docking results and IsoScore results feed back into the optimizer to improve discrimination.

6. Validate

Statistical validation ensures the scoring profile is robust:

  • Cross-validation (CV) — k-fold AUC to measure generalization
  • Y-scramble — permutation test to rule out overfitting
  • Bootstrap — confidence intervals on AUC and enrichment metrics

7. Export Results

Download results in multiple formats:

  • CSV — ranked ligand list with scores, poses, and ADME properties
  • JSON — structured output for programmatic integration
  • PDF — formatted report with plots and statistics

LP+ Lead Optimization (Optional)

For validated hits, the LP+ module provides FEP-calibrated binding affinity predictions to guide lead optimization. This step is optional and can be entered after any scoring or validation stage.

Alternate Entry Points

Not every project follows the full pipeline. Common abbreviated workflows:

Workflow Entry Point Exits At
Quick screen Upload PDB Dock AUC + Poses
Rescore existing poses IsoScore Export Results
Profile calibration Optimize Scoring Profile Validate
Full discovery Upload PDB Export Results