Browser-based tools for analyzing protein-protein interactions from structure predictions. Prediction files are read in your browser and never uploaded, and no installation is needed; to name chains, LIVIA sends sequence checksums and unmatched sequences to UniProt and EBI BLAST (what leaves your browser).
Web: https://flyark.github.io/LIVIA/ · Preprint: Kim & Perrimon 2026, bioRxiv · Atlas: LIVIA Atlas
upd1–dome (Drosophila JAK-STAT pathway, orthologous to human IL-6/gp130) — ChimeraX visualization generated by LIVIA script (iLIS: 0.529, ipTM: 0.57)
Left: PAE map (A: upd1, B: dome) showing confident interaction region (blue). Right: ChimeraX structure with LIR (light) and cLIR (dark) coloring.
- Drag & drop — upload prediction files directly into your browser (.zip, .gz, .xz, .cif, .pdb, .json, .npz, .npy)
- Auto-detection — automatically identifies the prediction platform from filenames
- Interaction residue detection — LIR (PAE ≤ 12Å) and cLIR (PAE ≤ 12Å & Cβ ≤ 8Å)
- Confidence scoring — iLIS, iLIA, iLISA, ipSAE, ipTM per chain pair
- Interactive visualizations — PAE/LIS/cLIS heatmaps, sequence viewer, linear contact map, chord diagram
- 3D structure viewer — Mol* viewer with LIR/cLIR coloring
- ChimeraX and PyMOL scripts — color presets (gradient, solid, high contrast, pLDDT, bychain)
- TED domain annotations — domain boundaries from AlphaFold DB displayed alongside detected regions
- CSV download — full metrics with LIR/cLIR residue indices, LIpLDDT/cLIpLDDT per chain
- Interactome clustering (cLIP) — cluster a bait's partners by shared contact residues; per-partner interface maps, searchable sequence viewer, and AlphaFold DB structure resolution
- Network building (PPI Network) — Leiden communities from
lis.pyoutput, computed in-browser (Pyodide + python-igraph)
| Page | Description |
|---|---|
| Prediction Analysis | Upload and analyze multi-platform predictions |
| FlyPredictome | Drosophila PPI analysis from FlyPredictome |
| Ortholog Interactome | Non-fly predictions from FlyPredictome ortholog search |
| AFDB Dimer | Dimer analysis from AlphaFold DB |
| Monomer Subdomain | Intramolecular domain interaction analysis from AlphaFold DB |
| cLIP | Cluster a protein's interactome by shared contact residues (cLIR) from lis.py output |
| PPI Network | Build an interaction network with Leiden communities, computed in-browser (Pyodide + python-igraph) |
| Tutorials | Step-by-step visual walkthroughs with auto-advancing screenshots |
| About | Metric definitions, color schemes, and references |
Prediction Analysis auto-detects the platform from uploaded files:
AlphaFold3 · AlphaFold2 · ColabFold · Boltz-1/2 · Chai-1 · OpenFold3 · Protenix-v2 · ESMFold2
- iLIS —
sqrt(LIS × cLIS)— integrated Local Interaction Score (Kim et al. 2026) - LIS — average confidence of residue pairs with PAE ≤ 12Å (Kim et al. 2024)
- cLIS — contact-filtered LIS (PAE ≤ 12Å & Cβ ≤ 8Å)
- iLIA —
sqrt(LIA × cLIA)— integrated interaction area - iLISA —
iLIS × iLIA - ipSAE — interaction prediction Score from Aligned Errors (Dunbrack, 2025)
- actifpTM — actual interface pTM (Varga et al., 2025)
- LIR / cLIR — Local Interaction Residues / contact residues (LIR also in contact)
- LIpLDDT / cLIpLDDT — average pLDDT of LIR / cLIR residues per chain
For large-scale batch analysis without a browser, use the command-line tools from AFM-LIS. They support all the same platforms, auto-detect the prediction format, and output CSV / ChimeraX scripts.
Score predictions with lis.py (writes CSV):
python lis.py /path/to/predictions/ # auto-detect, process all models
python lis.py /path/to/predictions/ -w 4 # parallel with 4 CPUs
python lis.py alphafold3_output.zip # zip input
python lis.py /path/to/predictions/ -v # verbose error detailsFeatures: .gz/.xz decompression, incremental CSV output (safe to interrupt and resume), progress bar with ETA, sorted output by name and rank.
Generate batch ChimeraX visualizations with lis_to_cxc.py (reads the lis.py CSV, writes one .cxc per fold × rank):
python lis_to_cxc.py \
--csv results_lis_analysis.csv \
--pdb-root . \
--out cxc/Double-click any .cxc to open in ChimeraX with LIR (light shade) / cLIR (full shade) coloring and an iLIS/cLIS label panel.
See AFM-LIS for full documentation and output CSV column reference.
- Tested on Chrome and Safari (macOS/iOS).
- Each prediction platform may produce different confidence calibrations. The iLIS ≥ 0.223 threshold was established using ColabFold/AlphaFold-Multimer predictions. Other platforms may require adjusted thresholds.
- AFM-LIS — Python framework and CLI for iLIS/LIS calculation
- FlyPredictome — Large-scale Drosophila PPI predictions (>1.5 million)
- AlphaFold Protein Structure Database — Predicted protein structures
LIVIA Atlas is a searchable atlas of AlphaFold-Multimer protein–protein interaction predictions, scored with LIVIA and resolved to residues. It draws on many published interactome datasets across species: for a protein, the partners predicted to bind it, how confident each prediction is, where each partner binds, and its orthologs side by side. Details, including the datasets and their sources, are on the LIVIA Atlas site.
Site: https://flyark.github.io/livia-atlas/
- Kim, A.-R. & Perrimon, N. (2026). LIVIA: a browser-based tool for assessing and visualizing predicted protein interactions. bioRxiv. https://doi.org/10.64898/2026.05.01.721633
- Kim, A.-R. et al. (2026). FlyPredictome: a structural atlas of predicted protein-protein interactions in Drosophila. bioRxiv. https://doi.org/10.64898/2026.04.14.718529
- Kim, A.-R. et al. (2024). Enhanced Protein-Protein Interaction Discovery via AlphaFold-Multimer. bioRxiv. https://doi.org/10.1101/2024.02.19.580970
If you use LIVIA in your research, please cite:
@article{livia2026,
author = {Kim, Ah-Ram and Perrimon, Norbert},
title = {LIVIA: a browser-based tool for assessing and visualizing predicted protein interactions},
year = {2026},
journal = {bioRxiv},
doi = {10.64898/2026.05.01.721633},
url = {https://doi.org/10.64898/2026.05.01.721633}
}MIT