Documentation

Docs / Profiles Library

Profiles

The Profiles page lets you browse, compare, and manage scoring weight profiles. Each profile defines how LatticeZero's 14 physics terms are weighted when scoring ligand-protein interactions.

Overview

Property Value
Access Profiles (sidebar)
Library profiles 70+ validated profiles
User profiles Unlimited (saved to your account)
Tiers Platinum (≥0.90), Gold (≥0.80), Silver (≥0.60)

Understanding Scoring Profiles

LatticeZero uses three profile categories with different derivation pathways:

1. Validated Demo Profiles (18 targets)

Per-target profiles tuned on DEKOIS2 active/decoy benchmarks. Weights are iteratively optimized to maximize AUC and validated with 5-fold CV plus Y-scramble controls.

2. Dock-Mode Profiles (8 targets)

Profiles optimized for IsoPose-generated poses using rank normalization, representing true end-to-end docking performance.

3. Class-Level Predictions (any target)

Fallback class recipes (kinase, protease, metalloenzyme, etc.) used when no validated target profile is available. These are untuned for the specific target and are the real cold-pass mode. On a 5-target DUD-E holdout, this untuned mode reached Silver-tier AUC on PNPH (0.642).

Profile Library

The LatticeZero Library contains a mix of validated profiles and class-level fallback recipes, organized by:

  • Target class - Kinase, Protease, GPCR, Nuclear Receptor, Metalloenzyme, PPI, Chaperone, Enzyme
  • Tier - Platinum, Gold, Silver, or Internal (based on validated AUC on DEKOIS2)
  • Method - E2E Validated (full docking pipeline), Score Validated (rescoring only), Class Recipe (untuned fallback)

Tier System

Tier AUC Threshold Meaning
Platinum ≥ 0.90 Excellent discrimination
Gold ≥ 0.80 Good discrimination
Silver ≥ 0.60 Moderate discrimination
Internal < 0.60 Experimental / research use

Browsing Profiles

Use the filters at the top of the Profiles page:

  • Search - Find by target name
  • All Classes - Filter by target family
  • All Tiers - Filter by validation quality
  • All Methods - Filter by validation method

Each profile card shows:

  • Target name and class
  • Validated AUC score
  • Tier badge (color-coded)
  • Description of the scoring approach

Profile Details

Click View on any profile to see:

  • Full weight breakdown across all 14 physics terms
  • Validation methodology and dataset
  • Recommended use cases
  • Scoring mode (geometric, hybrid, linear, etc.)

My Profiles

Creating Profiles

You can create custom profiles through:

  1. Autotune - Unsupervised profile generation from receptor + ligands
  2. Optimizer - Supervised optimization using active/decoy data
  3. Duplicate - Copy a library profile and modify weights manually

Using Profiles

When submitting a docking job in IsoPose:

  1. Select a Target Class from the dropdown
  2. Choose a specific Profile from the filtered list
  3. The profile's scoring weights are applied during docking

Profiles can also be exported as JSON for use with the CLI tools.