LIVIA — Prediction Analysis

Analyze predictions from AlphaFold3, ColabFold, Boltz, Chai-1, and OpenFold

Upload Prediction Files  (upload your prediction files — all analysis runs locally in your browser)
📦
Drop prediction files or folders here (.zip, .gz, .xz, .cif, .pdb, .json, .npz, .npy) or click to browse
p53–MDM2 examples:
AlphaFold3 predicted via AlphaFold Server. ColabFold, Boltz-2, Chai-1, and OpenFold3 predicted via Tamarind Bio.
For batch analysis of many predictions, use lis.py from AFM-LIS.

Interaction Analysis Summary

↓ Download CSV
Click a row to switch models. This updates all visualizations and the script below.
Rank/ModelPairiLISiLIAiLISAipSAEipTMLIScLISLIR (i/j)cLIR (i/j)
iLIS: ≥0.551 (1% FDR) ≥0.339 (5% FDR) ≥0.223 (10% FDR) <0.223 (below threshold)
iLIS (average): ≥0.303 (1% FDR) ≥0.120 (5% FDR) ≥0.073 (10% FDR) <0.073 (below threshold)
Thresholds based on large-scale Y2H reference sets in yeast, fly, and human predicted using ColabFold. See Kim et al. 2025 for iLIS benchmark details. Different platforms may require different thresholds.

Predicted Aligned Error (PAE) Maps

0Predicted Aligned Error (Å)30

Local Interaction Score (LIS) Maps (PAE ≤ 12 Å)

0Local Interaction Score (LIS)1

contact Local Interaction Score (cLIS) Maps (PAE ≤ 12 Å & Cβ ≤ 8 Å)

0contact Local Interaction Score (cLIS)1

Visualization Script

Fill gaps ≤ residues | Min segment ≥ residues
Gap filling bridges short breaks for continuous cartoon. Min segment removes isolated LIR fragments shorter than the threshold.
PAE cutoff (Å):
Cβ cutoff (Å):
Teal / Coral
Blue / Orange
Purple / Gold
Slate / Rose
Indigo / Tangerine
Green / Red
Cyan / Navy / Peach / Crimson
Sky / Indigo / Salmon / Red
Lime / Forest / Gold / Maroon
Mint / DarkTeal / Apricot / Burnt
Orchid / Purple / Yellow / Fire
Blue / Navy / Sage / Ember
Teal / Coral
Blue / Orange
Purple / Gold
Steel / Wheat
Cornflower / Tomato
pLDDT
bychain
bypolymer

Chain A (Protein 1)

LIR
#80CBC4
cLIR
#00897B

Chain B (Protein 2)

LIR
#FFAB91
cLIR
#E64A19

Place the script and structure files in the same folder, then open the script in ChimeraX or PyMOL. "All Chains" shows every chain with LIR regions, colored by chain.
How to Use  (Full guide)

Quick Start

  1. Run a prediction on any supported platform
  2. Download/export the results
  3. Drop the files above — the platform is auto-detected
  4. Click Process to calculate AFM-LIS metrics
  5. Click a chain pair in the table to generate a ChimeraX script
  6. Download the script (.cxc or .pml) and structure file, place in the same folder

Platform-Specific Tips

  • AlphaFold3: Drop the zip from AlphaFold Server directly
  • ColabFold: Upload the output folder or select individual _unrelaxed_rank_*.pdb and _scores_rank_*.json files
  • Boltz-1/2: Upload output folder containing .cif, confidence_*.json, and pae_*.npz files
  • Chai-1: Upload output folder with pred.model_idx_*.cif and scores files
  • OpenFold: Upload .cif with *_full_data*.json files
  • Generic: Any .cif/.pdb with a PAE JSON file (predicted_aligned_error field)

AFM-LIS Metrics (Kim et al. 2024, 2025)

iLISintegrated LIS — √(LIS × cLIS)
iLIAintegrated LIA — √(LIA × cLIA), geometric mean of interface area counts
iLISAintegrated LISA — iLIS × iLIA, overall binding strength
LISLocal Interaction Score — normalized PAE confidence (0–1)
cLIScontact-filtered LIS — restricted to direct contacts
LIRLocal Interaction Residues (PAE ≤ 12 Å)
cLIRcontact-filtered LIR (PAE ≤ 12 Å & Cβ ≤ 8 Å)

Confidence Metrics (from prediction platform)

ipTMinterface predicted TM-score — global interface confidence from the prediction model
ipSAEinteraction prediction Score from Aligned Errors (Dunbrack, 2025)

PAE (Predicted Aligned Error) measures how confidently the model predicts the relative position of two residues — lower values indicate higher confidence.

References

  • Kim et al. 2024 — Enhanced PPI Discovery via AlphaFold-Multimer
  • Kim et al. 2025 — A Structure-Guided Kinase–Transcription Factor Interactome Atlas