# LIVIA — Local Interaction Visualization and Analysis > LIVIA (https://flyark.github.io/LIVIA/) is a client-side web suite for scoring, clustering, and visualizing predicted biomolecular interactions from AlphaFold-Multimer, AlphaFold 3, ColabFold, Boltz-1/2, Chai-1, OpenFold3, Protenix-v2, and ESMFold2, paired with the proteome-scale LIVIA Atlas (https://flyark.github.io/livia-atlas/). ## Web Applications - **Prediction Analysis (`universal.html`)**: `https://flyark.github.io/LIVIA/universal.html` - Client-side evaluator for local or remote structure predictions (`.pdb`/`.cif` + PAE `.json`/`.npz`/`.pkl`, or `.zip`/`.tar.gz` archives). - Computes `LIS`, `cLIS`, `iLIS`, `LIA`, `cLIA`, `iLIA`, `iLISA`, `ipSAE`, `actifpTM`, `pDockQ`, `LIpDockQ`, `pDockQ2`, `LIpDockQ2`, `ipTM`, `pTM`, per-chain `LIR` and `cLIR` residue indices, interactive PAE/LIS/cLIS heatmaps (hover readout + click-to-highlight in 3D), linear/circular interaction maps, 3D Mol* views (`Isolate selected pair`, PNG snapshot), and refold exports (`FASTA`, `AlphaFold 3 JSON`, `Boltz-2 YAML`). - Supports URL loading via `?url=` or postMessage handshakes (`?post=1`, `livia-load` and zero-copy `livia-load-files`), plus keyboard shortcuts (`↑`/`↓` switch pair, `1`–`5` switch model, `i` isolate pair). - **Lightweight bundles**: Shrinks large predictions (tens to hundreds of MB of numeric PAE) to ~1–5 MB shareable ZIPs (`lis.csv` + structures + static PAE images) via the in-browser **↓ Lightweight bundle** button or `python/make_lightweight_bundle.py`; loads identically via drag-and-drop or `?url=` (4 built-in 6–14-subunit examples on `universal.html`). - **cLIP (`clip.html`)**: `https://flyark.github.io/LIVIA/clip.html` - Clustered Local Interaction Profiler for `lis.py` CSV outputs (`?data=&gene=`). - Clusters passing models by cosine distance over query `cLIR` contact fingerprints to map binding sites along a query protein, overlaid with Pfam/UniProt domains, AlphaFold DB `pLDDT`, and AlphaMissense scores, with `by cluster`, `all models` (combined total curve), and `classic` frequency modes. - **PPI Network (`network.html`)**: `https://flyark.github.io/LIVIA/network.html` - Interactive network builder and community detector (Leiden, Louvain, MCL) for `lis.py` screen tables. - **FlyPredictome Viewer (`flypredictome.html`)**: `https://flyark.github.io/LIVIA/flypredictome.html` - Inspects *Drosophila* PPI runs from FlyPredictome (`?url=https://www.flyrnai.org/tools/fly_predictome/web/famdb_details////`). - **Ortholog Interactome Viewer (`ortholog_predictome.html`)**: `https://flyark.github.io/LIVIA/ortholog_predictome.html` - Inspects kinase-TF and kinase-kinase runs hosted on FlyPredictome across species. - **AlphaFold DB Dimer & Monomer Viewers (`dimer.html`, `monomer.html`)**: - `https://flyark.github.io/LIVIA/dimer.html?id=` - `https://flyark.github.io/LIVIA/monomer.html?id=` (subdomains via PAE Leiden clustering). - **LIVIA Atlas**: `https://flyark.github.io/livia-atlas/` (see `https://flyark.github.io/livia-atlas/llms.txt` for URL deep-links, `window.LIVIA_ATLAS`, and `scripts/atlas_query.py`). ## One-Click Markdown Export & Read-Only JavaScript API (`window.LIVIA`) Every interactive LIVIA page (`universal.html`, `clip.html`, `flypredictome.html`, `ortholog_predictome.html`, `dimer.html`, `monomer.html`, `network.html`) provides a **Copy Markdown** button (`#livia-copy-md`) next to its CSV/report export controls and exposes `window.LIVIA` for browser-capable AI agents: - `window.LIVIA.status()` — `{ page, ready, title, build }`. - `await window.LIVIA.waitReady(timeoutMs = 15000)` — Waits until predictions/clustering finish, then returns `window.LIVIA.inspect()`. - `window.LIVIA.inspect()` — Returns a JSON-serializable snapshot of chains, per-rank and average scores (`iLIS`, `iLISA`, `ipSAE`, `actifpTM`, `ipTM`, `pTM`, `LIS`, `cLIS`, `pDockQ2`), exact `cLIR` (`PAE ≤ 12 Å, Cβ ≤ 8 Å`) and `LIR` (`PAE ≤ 12 Å`) residue ranges, `cLIR` amino-acid sequence stretches, cLIP clusters/footprints, or network communities. - `window.LIVIA.toMarkdown()` — Formats `inspect()` as a concise Markdown report with tables and exact residue stretches for pasting into an LLM chat. ## Metric & Cutoff Reference - **LIR (Local interaction residues)**: Residue pairs across chains with `PAE ≤ 12 Å`. - **cLIR (Contact residues)**: Residue pairs across chains with `PAE ≤ 12 Å` and `Cβ ≤ 8 Å` (`Cα` for Gly). - **LIS**: Mean of `(12 - PAE) / 12` over cross-chain pairs with `PAE ≤ 12 Å`. - **cLIS**: Mean of `(12 - PAE) / 12` over cross-chain pairs with `PAE ≤ 12 Å` and `Cβ ≤ 8 Å`. - **iLIS**: `sqrt(LIS * cLIS)`. - **Benchmarked `iLIS` FPR cutoffs** (calibrated on the top-ranked of five models per pair): - `10% FPR`: `0.223` - `5% FPR`: `0.339` - `1% FPR`: `0.551` ## Command-Line Scoring (`AFM-LIS` / `lis.py`) Source repository for batch scoring AlphaFold-Multimer, AlphaFold 3, ColabFold, and Boltz outputs: `https://github.com/flyark/AFM-LIS`. ## Citation Kim A-R & Perrimon N (2026). LIVIA, bioRxiv. https://doi.org/10.64898/2026.05.01.721633