{"slug":"alterlab-alphafold-db","title":"alterlab-alphafold-db","summary":"Access the AlphaFold DB of 240M+ AI-PREDICTED protein structures (v6, plus precomputed homodimer/heterodimer complexes) — retrieve models by UniProt accession, download PDB/mmCIF files, and analyze prediction confidence metrics (pLDDT, PAE). Use when a UniProt ID needs a computat","platform":"Claude","tags":[],"authorName":"LLM Mart","authorSlug":"llm-mart","score":0,"source":"github","price":null,"verified":false,"createdAt":"2026-09-23T18:57:07.403467Z","repo":{"url":"https://github.com/AlterLab-IEU/AlterLab-Academic-Skills","stars":68,"forks":13,"license":"MIT","updatedAt":"2026-09-23T13:42:59Z"},"bodyHtml":"<hr>\n<h2>name: alterlab-alphafold-db\ndescription: Access the AlphaFold DB of 240M+ AI-PREDICTED protein structures (v6, plus precomputed homodimer/heterodimer complexes) — retrieve models by UniProt accession, download PDB/mmCIF files, and analyze prediction confidence metrics (pLDDT, PAE). Use when a UniProt ID needs a computationally predicted 3D structure or when no experimental structure exists, for homology modeling, protein engineering, or structure-based drug discovery; for EXPERIMENTALLY determined structures (X-ray, cryo-EM, NMR) prefer alterlab-pdb, to fold a NEW sequence or complex yourself prefer alterlab-alphafold, and for protein sequences, annotations, or accession ID mapping prefer alterlab-uniprot instead. Part of the AlterLab Academic Skills suite.\nlicense: MIT\nallowed-tools: Read WebFetch Bash(curl:<em>) Bash(python:</em>)\ncompatibility: Keyless AlphaFold DB (EBI) REST API (v6 models); EMBL-EBI FTP (v6) or Google Cloud/BigQuery (v4) for bulk proteome downloads\nmetadata:\nskill-author: AlterLab\nversion: \"1.2.0\"\nlast_updated: \"2026-09-23\"</h2>\n<h1>AlphaFold Database</h1>\n<h2>Overview</h2>\n<p>AlphaFold DB is a public repository of AI-predicted 3D protein structures maintained by Google DeepMind and EMBL-EBI. Release v6 (October 2025, synced to UniProt 2025_03) holds ~241 million predictions, including ~40k isoforms and the input MSAs; since March 2026 it also serves precomputed homodimer and heterodimer complex predictions. Access structure predictions with confidence metrics, download coordinate files, retrieve bulk datasets, and integrate predictions into computational workflows.</p>\n<h2>When to Use This Skill</h2>\n<p>This skill should be used when working with AI-predicted protein structures in scenarios such as:</p>\n<ul>\n<li>Retrieving protein structure predictions by UniProt ID or protein name</li>\n<li>Downloading PDB/mmCIF coordinate files for structural analysis</li>\n<li>Analyzing prediction confidence metrics (pLDDT, PAE) to assess reliability</li>\n<li>Accessing bulk proteome datasets (EMBL-EBI FTP or Google Cloud Platform)</li>\n<li>Comparing predicted structures with experimental data</li>\n<li>Performing structure-based drug discovery or protein engineering</li>\n<li>Building structural models for proteins lacking experimental structures</li>\n<li>Integrating AlphaFold predictions into computational pipelines</li>\n</ul>\n<h3>Does NOT Trigger</h3>\n<table>\n<thead>\n<tr>\n<th>Scenario</th>\n<th>Use Instead</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>Experimental X-ray / cryo-EM / NMR structure by PDB ID</td>\n<td><code>alterlab-pdb</code></td>\n</tr>\n<tr>\n<td>Folding a new or mutated sequence, or a custom protein–protein complex, yourself (ColabFold / AF2-Multimer)</td>\n<td><code>alterlab-alphafold</code></td>\n</tr>\n<tr>\n<td>Protein–ligand or protein–nucleic-acid co-folding</td>\n<td><code>alterlab-boltz</code></td>\n</tr>\n<tr>\n<td>Protein sequence, functional annotation, or accession ID mapping only</td>\n<td><code>alterlab-uniprot</code></td>\n</tr>\n</tbody>\n</table>\n<h2>Core Capabilities</h2>\n<p>Worked, copy-paste Python recipes for every capability below live in\n<code>references/code_examples.md</code>. Load it when you need runnable code; the summaries\nhere give the routing and the key decisions.</p>\n<h3>1. Searching and Retrieving Predictions</h3>\n<p>Three entry points, in order of preference:</p>\n<ul>\n<li><strong>Biopython</strong> (recommended): <code>Bio.PDB.alphafold_db.get_predictions(accession)</code>,\n<code>download_cif_for(...)</code>, <code>get_structural_models_for(...)</code> — simplest path.</li>\n<li><strong>Direct REST</strong>: <code>GET https://alphafold.ebi.ac.uk/api/prediction/{uniprot_id}</code>.\nThe response is a list of every model for that accession — the canonical\nsequence plus isoforms (<code>P00520-2</code>, …) and, for some entries, third-party\nmodels — so select the record whose <code>uniprotAccession</code> equals your query\nrather than trusting <code>[0]</code>. The model ID is <code>modelEntityId</code> (e.g.\n<code>AF-P00520-F1</code>); <code>entryId</code> is the legacy name that passed its announced\n2026-06-25 sunset, so don't build new code on it. Complex models are\nexcluded unless you pass <code>?include_complexes=true</code> (or call\n<code>/api/complex/{id}</code>).</li>\n<li><strong>Find accessions first via UniProt</strong> when you only have a gene name or PDB ID —\nuse the UniProt ID-mapping job API (<code>get_uniprot_ids</code> helper in\n<code>code_examples.md</code> §1; valid db names at\n<a href=\"https://rest.uniprot.org/configure/idmapping/fields\">https://rest.uniprot.org/configure/idmapping/fields</a>).</li>\n</ul>\n<h3>2. Downloading Structure Files</h3>\n<p>The <code>/prediction</code> response carries version-stamped file URLs — <strong>use those, don't\nhand-build a <code>_v{N}</code> suffix.</strong> The DB version advances (currently v6) and old\n<code>_v4</code> file URLs now 404:</p>\n<ul>\n<li><code>cifUrl</code> / <code>pdbUrl</code> / <code>bcifUrl</code> — atomic coordinates (mmCIF / PDB / binary CIF).</li>\n<li><code>plddtDocUrl</code> — per-residue pLDDT scores (0-100).</li>\n<li><code>paeDocUrl</code> — PAE matrix.</li>\n</ul>\n<p>Download recipe (resolve URLs from the API, write bytes) in <code>code_examples.md</code> §2.</p>\n<h3>3. Working with Confidence Metrics</h3>\n<ul>\n<li><strong>pLDDT</strong>: from <code>plddtDocUrl</code>, read <code>confidence['confidenceScore']</code> (keys:\n<code>residueNumber</code>, <code>confidenceScore</code>, <code>confidenceCategory</code>); thresholds in\n\"Confidence Interpretation Guidelines\" below.</li>\n<li><strong>PAE</strong>: from <code>paeDocUrl</code>. The endpoint returns a single-element JSON array of\none object, so index <code>[0]</code> before the key\n(<code>pae[0]['predicted_aligned_error']</code>). Visualization recipe in\n<code>code_examples.md</code> §3.</li>\n</ul>\n<h3>4. Bulk Data Access (FTP v6 or Google Cloud v4)</h3>\n<ul>\n<li><strong>Model organisms, global-health proteomes, Swiss-Prot (v6):</strong> one tar per\nproteome at <code>https://ftp.ebi.ac.uk/pub/databases/alphafold/latest/</code>\n(e.g. <code>UP000005640_9606_HUMAN_v6.tar</code>; the index is <code>download_metadata.json</code>\nin the parent directory). Lower-confidence complex predictions are bulk-only,\nunder <code>.../alphafold/collaborations/nvda/</code>.</li>\n<li><strong>Any taxon (v4):</strong> <code>gs://public-datasets-deepmind-alphafold-v4/proteomes/</code>\nwith <code>gsutil</code>, or query <code>bigquery-public-data.deepmind_alphafold.metadata</code> to\nfilter by organism/confidence. The species-download helper validates the\ntaxonomy ID and uses list-form <code>subprocess.run</code> (never <code>shell=True</code>).</li>\n</ul>\n<p>See <code>code_examples.md</code> §4 and <code>references/api_reference.md</code> (Bulk Downloads).</p>\n<h3>5. Parsing and Analyzing Structures</h3>\n<p>Parse mmCIF with <code>Bio.PDB.MMCIFParser</code>; pLDDT is stored in the B-factor column\n(<code>residue['CA'].get_bfactor()</code>). Contact-map and B-factor extraction recipes in\n<code>code_examples.md</code> §5.</p>\n<h3>6. Batch Processing Multiple Proteins</h3>\n<p>Loop accessions → predictions → confidence stats → summary DataFrame. Full\nexample in <code>code_examples.md</code> §6.</p>\n<h2>Installation and Setup</h2>\n<pre><code>uv pip install biopython requests          # core: structure access + API\nuv pip install numpy matplotlib pandas scipy  # analysis + PAE plots\nuv pip install google-cloud-bigquery gsutil   # optional: bulk GCP access\n</code></pre>\n<p><strong>3D-Beacons alternative:</strong> AlphaFold is also reachable via the 3D-Beacons\nfederated API (<code>https://www.ebi.ac.uk/pdbe/pdbe-kb/3dbeacons/api/uniprot/summary/{id}.json</code>),\nfiltering entries where <code>structures[i]['summary']['provider'] == 'AlphaFold DB'</code>. Recipe in\n<code>code_examples.md</code> (3D-Beacons section).</p>\n<h2>Common Use Cases</h2>\n<h3>Structural Proteomics</h3>\n<ul>\n<li>Download complete proteome predictions for analysis</li>\n<li>Identify high-confidence structural regions across proteins</li>\n<li>Compare predicted structures with experimental data</li>\n<li>Build structural models for protein families</li>\n</ul>\n<h3>Drug Discovery</h3>\n<ul>\n<li>Retrieve target protein structures for docking studies</li>\n<li>Analyze binding site conformations</li>\n<li>Identify druggable pockets in predicted structures</li>\n<li>Compare structures across homologs</li>\n</ul>\n<h3>Protein Engineering</h3>\n<ul>\n<li>Identify stable/unstable regions using pLDDT</li>\n<li>Design mutations in high-confidence regions</li>\n<li>Analyze domain architectures using PAE</li>\n<li>Model protein variants and mutations</li>\n</ul>\n<h3>Evolutionary Studies</h3>\n<ul>\n<li>Compare ortholog structures across species</li>\n<li>Analyze conservation of structural features</li>\n<li>Study domain evolution patterns</li>\n<li>Identify functionally important regions</li>\n</ul>\n<h2>Key Concepts</h2>\n<p><strong>UniProt Accession:</strong> Primary identifier for proteins (e.g., \"P00520\"). Required for querying AlphaFold DB.</p>\n<p><strong>AlphaFold model ID (<code>modelEntityId</code>):</strong> <code>AF-[UniProt accession]-F[fragment number]</code> for DeepMind monomer models (e.g., \"AF-P00520-F1\"; isoforms look like \"AF-P00520-2-F1\"). Complex and third-party models use opaque numeric IDs (e.g., \"AF-0000000365776990\"); both forms are accepted by <code>/api/prediction/{id}</code>.</p>\n<p><strong>pLDDT (predicted Local Distance Difference Test):</strong> Per-residue confidence metric (0-100). Higher values indicate more confident predictions.</p>\n<p><strong>PAE (Predicted Aligned Error):</strong> Matrix indicating confidence in relative positions between residue pairs. Low values (&lt;5 Å) suggest confident relative positioning.</p>\n<p><strong>Database Version:</strong> The REST API and the FTP <code>latest/</code> archives serve v6 (the response reports <code>latestVersion</code> / <code>allVersions</code>); the GCS/BigQuery datasets lag at v4. File URLs include a version suffix (e.g., <code>model_v6.cif</code>, while newer third-party models start at <code>_v1</code>) — read them from the prediction response rather than hardcoding the suffix.</p>\n<p><strong>Fragment Number:</strong> Large proteins may be split into fragments. Fragment number appears in AlphaFold ID (e.g., F1, F2).</p>\n<h2>Confidence Interpretation Guidelines</h2>\n<p><strong>pLDDT Thresholds:</strong></p>\n<ul>\n<li><strong>&gt;90</strong>: Very high confidence - suitable for detailed analysis</li>\n<li><strong>70-90</strong>: High confidence - generally reliable backbone structure</li>\n<li><strong>50-70</strong>: Low confidence - use with caution, flexible regions</li>\n<li><strong>&lt;50</strong>: Very low confidence - likely disordered or unreliable</li>\n</ul>\n<p><strong>PAE Guidelines:</strong></p>\n<ul>\n<li><strong>&lt;5 Å</strong>: Confident relative positioning of domains</li>\n<li><strong>5-10 Å</strong>: Moderate confidence in arrangement</li>\n<li><strong>&gt;15 Å</strong>: Uncertain relative positions, domains may be mobile</li>\n</ul>\n<h2>Resources</h2>\n<h3>references/code_examples.md</h3>\n<p>Worked, copy-paste Python recipes for every Core Capability: prediction\nretrieval (Biopython / REST / UniProt mapping), file downloads, pLDDT + PAE\nanalysis, GCP/BigQuery bulk access, mmCIF parsing, batch processing, and the\n3D-Beacons alternative.</p>\n<p>Load this when you need runnable code.</p>\n<h3>references/api_reference.md</h3>\n<p>Comprehensive API documentation covering:</p>\n<ul>\n<li>Complete REST API endpoint specifications</li>\n<li>File format details and data schemas</li>\n<li>Google Cloud dataset structure and access patterns</li>\n<li>Advanced query examples and batch processing strategies</li>\n<li>Rate limiting, caching, and best practices</li>\n<li>Troubleshooting common issues</li>\n</ul>\n<p>Consult this reference for detailed API information, bulk download strategies, or when working with large-scale datasets.</p>\n<h2>Important Notes</h2>\n<h3>Data Usage and Attribution</h3>\n<ul>\n<li>AlphaFold DB is freely available under CC-BY-4.0 license</li>\n<li>Cite: Jumper et al. (2021) Nature, plus the AFDB paper for the release you used — Varadi et al. (2024) NAR for v4, Bertoni et al. (2026) NAR (doi:10.1093/nar/gkaf1226) for v6</li>\n<li>Predictions are computational models, not experimental structures</li>\n<li>Always assess confidence metrics before downstream analysis</li>\n</ul>\n<h3>Version Management</h3>\n<ul>\n<li>REST API and FTP <code>latest/</code> serve v6 (<code>latestVersion</code>); GCS/BigQuery bulk datasets lag at v4</li>\n<li>Read file URLs from the <code>/prediction</code> response — never hardcode the <code>_v{N}</code> suffix</li>\n<li>Old <code>_v4</code> file URLs now 404; superseded versions are removed from <code>/files</code> (older releases remain on the FTP site under <code>v1/</code>–<code>v6/</code>)</li>\n<li>The v6 field renames (<code>entryId</code>→<code>modelEntityId</code>, <code>uniprotStart/End</code>→<code>sequenceStart/End</code>, <code>uniprotSequence</code>→<code>sequence</code>, <code>isReviewed</code>→<code>isUniProtReviewed</code>) passed their 2026-06-25 sunset; <code>paeImageUrl</code> is slated for removal — use <code>paeDocUrl</code></li>\n<li>Track which version a downloaded result came from</li>\n</ul>\n<h3>Data Quality Considerations</h3>\n<ul>\n<li>High pLDDT doesn't guarantee functional accuracy</li>\n<li>Low confidence regions may be disordered in vivo</li>\n<li>PAE indicates relative domain confidence, not absolute positioning</li>\n<li>Predictions lack ligands, post-translational modifications, and cofactors</li>\n<li>Default <code>/prediction</code> results are single chains. Precomputed complexes (≈1.7M high-confidence homodimers and ≈80k heterodimers, added March–May 2026) come back only with <code>include_complexes=true</code> or <code>/api/complex/{id}</code>; judge them by interface metrics (ipTM, pDockQ) as well as pLDDT. For a complex that is not in the DB, fold it yourself (<code>alterlab-alphafold</code>)</li>\n</ul>\n<h3>Performance Tips</h3>\n<ul>\n<li>Use Biopython for simple single-protein access</li>\n<li>Use the FTP proteome tars or Google Cloud for bulk downloads (much faster than individual files)</li>\n<li>Cache downloaded files locally to avoid repeated downloads</li>\n<li>BigQuery free tier: 1 TB processed data per month</li>\n<li>Consider network bandwidth for large-scale downloads</li>\n</ul>\n<h2>Additional Resources</h2>\n<ul>\n<li><strong>AlphaFold DB Website:</strong> <a href=\"https://alphafold.ebi.ac.uk/\">https://alphafold.ebi.ac.uk/</a></li>\n<li><strong>API Documentation:</strong> <a href=\"https://alphafold.ebi.ac.uk/api-docs\">https://alphafold.ebi.ac.uk/api-docs</a> (machine-readable spec: <a href=\"https://alphafold.ebi.ac.uk/api/openapi.json\">https://alphafold.ebi.ac.uk/api/openapi.json</a>)</li>\n<li><strong>Release notes / FTP changelog:</strong> <a href=\"https://www.ebi.ac.uk/pdbe/news/alphafold-database-release-notes\">https://www.ebi.ac.uk/pdbe/news/alphafold-database-release-notes</a>, <a href=\"https://ftp.ebi.ac.uk/pub/databases/alphafold/CHANGELOG.txt\">https://ftp.ebi.ac.uk/pub/databases/alphafold/CHANGELOG.txt</a></li>\n<li><strong>Google Cloud Dataset:</strong> <a href=\"https://cloud.google.com/blog/products/ai-machine-learning/alphafold-protein-structure-database\">https://cloud.google.com/blog/products/ai-machine-learning/alphafold-protein-structure-database</a></li>\n<li><strong>3D-Beacons API:</strong> <a href=\"https://www.ebi.ac.uk/pdbe/pdbe-kb/3dbeacons/\">https://www.ebi.ac.uk/pdbe/pdbe-kb/3dbeacons/</a></li>\n<li><strong>AlphaFold Papers:</strong>\n<ul>\n<li>Nature (2021): <a href=\"https://doi.org/10.1038/s41586-021-03819-2\">https://doi.org/10.1038/s41586-021-03819-2</a></li>\n<li>Nucleic Acids Research (2024): <a href=\"https://doi.org/10.1093/nar/gkad1011\">https://doi.org/10.1093/nar/gkad1011</a></li>\n</ul>\n</li>\n<li><strong>Biopython Documentation:</strong> <a href=\"https://biopython.org/docs/dev/api/Bio.PDB.alphafold_db.html\">https://biopython.org/docs/dev/api/Bio.PDB.alphafold_db.html</a></li>\n<li><strong>GitHub Repository:</strong> <a href=\"https://github.com/google-deepmind/alphafold\">https://github.com/google-deepmind/alphafold</a></li>\n</ul>\n<h2>Scripts</h2>\n<p><code>scripts/query_alphafold.py</code> — runnable helper for the AlphaFold REST API (no key):</p>\n<pre><code>python scripts/query_alphafold.py prediction P00520\npython scripts/query_alphafold.py confidence P00520 --summary\npython scripts/query_alphafold.py download P00520 --fmt cif -o ./structures\n</code></pre>\n","files":[{"path":"evals/evals.json","sizeBytes":5934,"isText":true},{"path":"references/api_reference.md","sizeBytes":18393,"isText":true},{"path":"references/code_examples.md","sizeBytes":12549,"isText":true},{"path":"scripts/query_alphafold.py","sizeBytes":3714,"isText":true},{"path":"SKILL.md","sizeBytes":13423,"isText":true}],"reviewScore":null,"reviewSummary":null,"trust":{"provenance":"trusted-source-unreviewed","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow.","bodySource":null},"bodyLocked":false,"purchaseUrl":null,"sourceUrl":null,"report":{"provenance":"trusted-source-unreviewed","screen":{"ran":true,"outcome":"clean","suspicious":0,"notes":0,"hiddenCharacters":false},"virusScan":{"engine":"clamav","status":"clean","scannedAt":"2026-09-23T18:59:18.128631Z","sha256":"CC33370EFA20BEDD9704EB3FBF863AD46D5CA53250A4170EE8706AE66E2BC066","sizeBytes":22051},"review":null,"source":{"repositoryUrl":"https://github.com/AlterLab-IEU/AlterLab-Academic-Skills","path":"skills/databases/alterlab-alphafold-db","license":"MIT","commit":"e4836c08a20da195a11f30f203a8cf23ec30aa95","subtreeSha":"FF4ABC64789CCEECEDC310083B8AD7B97FC5381C9068BA7A6A9F130B519BBAF1","lastSyncedAt":"2026-09-23T18:56:52.297238Z"},"reviewedAt":"2026-09-23T19:03:30.133027Z","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow."},"install":[{"target":"skills-cli","command":"npx skills add https://github.com/AlterLab-IEU/AlterLab-Academic-Skills/tree/main/skills/databases/alterlab-alphafold-db"},{"target":"claude-code","command":"claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install alterlab-ieu-alterlab-academic-skills@llmmart"},{"target":"git","command":"git clone https://github.com/AlterLab-IEU/AlterLab-Academic-Skills.git"}]}