{"success":true,"data":{"author":"EMBL-EBI (EMDB)","bucketAccessType":"public","bucketListable":true,"bucketName":"ftp.ebi.ac.uk","category":"Structural Biology","cloudProvider":"api","createdAt":"2026-08-14T12:42:31.910Z","description":"EMDB is EMBL-EBI's public archive of 3D electron microscopy reconstructions: the density maps produced from cryo-EM, cryo-ET and volume-EM experiments, together with the half-maps, masks, FSC curves and metadata that document how each was determined. It is the processed counterpart to EMPIAR, which holds the raw images those maps were computed from, and the two cross-reference each other entry by entry. ~60,900 entries, each a separate structure determination keyed by an EMD accession (e.g. EMD-0339, the cannabinoid receptor CB1 bound to its G protein). Maps are CCP4/MRC volumes, typically tens of megabytes gzipped, so unlike EMPIAR an entry is small enough to open and work with directly. The whole tree is mounted read-only over EBI's public HTTPS endpoint; nothing is copied.","displayTitle":"EMDB: Electron Microscopy Data Bank","dockerImageUrl":"531775725781.dkr.ecr.us-east-1.amazonaws.com/scigantic-notebook-cryoem:latest","doi":"10.1093/nar/gkad1019","fuseMountBrowsable":false,"httpUrl":"https://ftp.ebi.ac.uk/pub/databases/emdb/structures/","id":"b1f4c8e2-3d7a-5c91-84b6-2e9f7a3c5d18","license":"CC0 1.0 — EMBL-EBI places all primary and derived EMDB data in the public domain","referenceUrl":"https://www.ebi.ac.uk/emdb/","region":"us-east-1","schemaCard":{"accessHints":[{"language":"python","snippet":"# START HERE. Finding the right entry is the hard part of EMDB, not reading it.\nfrom scigantic_emdb import EmdbCatalog\ncat = EmdbCatalog()\n\n# Free text over title, sample name, organism, method and accessions, with\n# cryo-EM synonyms (so \"cryoET\" reaches \"cryo-ET\"). What the user literally\n# typed always ranks first.\ncat.search(\"spliceosome\", max_res=3.5)\ncat.search(\"GPCR\", organism=\"Homo sapiens\", has_pdb=True, sort=\"resolution\")\ncat.search(\"ribosome\", method=\"helical reconstruction\", min_year=2023)\n\n# Filters: max_res/min_res (Angstrom), organism, method, has_pdb,\n# has_uniprot, year/min_year/max_year, sort=relevance|resolution|year|id.\ncat.entry(\"EMD-22962\")      # one row as a dict\ncat.coverage()              # what the index actually holds, measured"},{"language":"python","snippet":"# Molecular filters. max_chain_kda is the LARGEST SINGLE PROTEIN CHAIN,\n# which is the one to use for \"a receptor under 100 kDa\": the assembled\n# complex carries the G protein and any nanobodies, so filtering on\n# complex weight finds the wrong structures.\ncat.search(\"GPCR\", max_chain_kda=100, max_res=3.5, has_half_maps=True)\ncat.search(\"ribosome\", complex_kda_max=3000, has_mask=True)\ncat.search(\"protease\", ligand=\"ATP\", has_model=True)\ncat.search(\"capsid\", microscope=\"KRIOS\", min_year=2023)\n\n# These come from the deposition and are NOT universal (weights sit near\n# 58%, masks lower). A record missing the filtered field is excluded, so\n# check cat.coverage()[\"enrichment\"][\"field_fill_pct\"] before telling a\n# user a count is the whole picture."},{"language":"python","snippet":"# Acquisition and reconstruction fields, read from the DEPOSIT. Use them to find data\n# recorded the way yours was, and to see what a published map really used.\ncat.search(method=\"single-particle reconstruction\", voltage_kv=300, detector=\"K3\",\n           has_raw_data=True, max_raw_gb=400, sort=\"year\")\ncat.search(\"ribosome\", min_dose=20, min_particles=50000, max_res=3.0)\ncat.search(\"apoferritin\", symmetry=\"O\")\n# min_chain_copies / max_chain_copies is the oligomeric order of a single protein.\n\n# One entry, to see what its deposit says about how it was recorded and processed:\ne = cat.entry(\"EMD-0731\")\n[e[k] for k in (\"voltage_kv\", \"detector\", \"dose_e_a2\", \"particles_used\", \"symmetry\")]\n# -> [200.0, 'FEI FALCON III (4k x 4k)', 50.0, 89513.0, 'C3']"},{"language":"python","snippet":"# Datasets whose deposited maps show MORE THAN ONE assembly state (a monomer and a\n# trimer, different copy numbers or point groups). One raw dataset often yields\n# several maps, so the evidence sits across sibling records, not in any one of them.\nm = cat.datasets_with_multiple_states(max_raw_gb=400)\nm[[\"empiar_id\", \"n_states\", \"copies\", \"symmetries\", \"title_states\", \"raw_size_gb\", \"example_title\"]]\n# A HEURISTIC that ranks candidates for a person to read, not a verdict: one state\n# deposited with and without symmetry (C1 and D2) also trips it."},{"language":"python","snippet":"# Gallery. EBI already publishes an isosurface rendering for ~97% of\n# entries, so this costs no map reads and copies nothing — the card\n# points at EBI's public URL.\nhits = cat.search(\"spliceosome\", max_res=4.0)\ncat.gallery(hits.head(8))\n\nfrom scigantic_emdb import image_url\nimage_url(cat.entry(\"EMD-22962\"))   # \"\" when the entry has no rendering"},{"language":"python","snippet":"# Read and look at a map — two calls rather than the usual six.\nfrom scigantic_emdb import load_map, slices, entry_files\nvol, meta = load_map(\"EMD-22962\")   # decompresses to /tmp, caches, opens\nprint(meta[\"shape\"], meta[\"voxel_size_a\"], meta[\"size_mb\"])\nslices(vol)                          # central XY/XZ/YZ sections\n\n# What this entry actually deposited. Half-maps/masks/FSC are per-entry,\n# so check rather than assuming — this returns only what exists.\nentry_files(\"EMD-22962\")             # {\"map\": [...], \"other\": [...], ...}"},{"language":"python","snippet":"# \"Which of these can I actually reprocess?\" is ONE search, not a join.\ncat.search(\"GPCR\", max_res=3.0, has_raw_data=True, sort=\"raw_size\")\n\n# sort=\"raw_size\" puts the cheapest first. Cryo-EM raw data is enormous:\n# of the 25 sub-3A GPCR structures with public movies, NONE is under 1 TB\n# and the cheapest is 1,331 GB. Scope it to what the user can actually pull:\ncat.search(\"ribosome\", has_raw_data=True, max_raw_gb=2000)\n\n# Only 8.2% of EMDB (5,007 of 60,895) has public raw data, because most\n# groups never deposit it. Say so rather than implying the rest is missing.\ncat.with_empiar(hits)   # full pairing: emdb_id, empiar_id, raw_size_gb"},{"language":"python","snippet":"# A map is small enough to just read. Files are gzipped CCP4/MRC.\nimport os, gzip, shutil, mrcfile, numpy as np\nbase = os.environ[\"SCIGANTIC_MOUNT_PATH\"]        # /mnt/http-archive/data\nacc  = \"EMD-22962\"; n = acc.split(\"-\")[1]\nsrc  = f\"{base}/{acc}/map/emd_{n}.map.gz\"\n# Decompress to local disk first — mrcfile needs random access, and the\n# mount is a lazy HTTP range reader, so streaming a gzip through it is slow.\nwith gzip.open(src, \"rb\") as fh, open(f\"/tmp/{acc}.map\", \"wb\") as out:\n    shutil.copyfileobj(fh, out)\nwith mrcfile.open(f\"/tmp/{acc}.map\", permissive=True) as m:\n    vol = m.data           # 3D numpy array\n    print(vol.shape, m.voxel_size)"},{"language":"python","snippet":"# Central slices — the fastest way to see what a map contains.\nimport matplotlib.pyplot as plt\nz, y, x = [s // 2 for s in vol.shape]\nfig, ax = plt.subplots(1, 3, figsize=(12, 4))\nfor a, (img, t) in zip(ax, [(vol[z], \"XY\"), (vol[:, y], \"XZ\"), (vol[:, :, x], \"YZ\")]):\n    a.imshow(img, cmap=\"gray\"); a.set_title(t); a.axis(\"off\")"},{"language":"python","snippet":"# Half-maps + mask, when deposited — what you need for an FSC.\nh1 = f\"{base}/{acc}/other/emd_{n}_half_map_1.map.gz\"\nh2 = f\"{base}/{acc}/other/emd_{n}_half_map_2.map.gz\"\nmk = f\"{base}/{acc}/masks/emd_{n}_msk_1.map\"\n# Not every entry has them. Check before assuming:\nprint(os.path.isdir(f\"{base}/{acc}/other\"), os.path.isdir(f\"{base}/{acc}/masks\"))"},{"language":"bash","snippet":"# EMDB REST API — metadata without touching the mount (sample, resolution, cross-refs):\ncurl https://www.ebi.ac.uk/emdb/api/entry/EMD-0339\n# Search EMDB: https://www.ebi.ac.uk/emdb/search"}],"caveats":["USE THE CATALOG TO FIND THINGS. `from scigantic_emdb import EmdbCatalog` then EmdbCatalog().search(...). This is the single most important thing about this archive: it has ~60,900 entries and no other way to locate one by science rather than by accession. If a user asks for \"cryo-EM structures of X\" or \"the best resolution Y\", that is a catalog search, not a directory walk and not a guess from memory. Never invent an EMD accession — look it up, then read it.","MOLECULAR FIELDS ARE PARTIAL, AND THAT CHANGES HOW YOU REPORT A COUNT. max_chain_kda, complex_kda, ligands and has_mask come from what each group chose to deposit, not from every entry: weights sit near 58% and masks lower, while microscope, box and map_mb are ~100%. A record missing the filtered field is EXCLUDED, so \"12 structures match\" means twelve among those that deposited a weight, not twelve in EMDB. Say so, and read the real numbers from cat.coverage()[\"enrichment\"][\"field_fill_pct\"].","max_chain_kda IS THE LARGEST SINGLE PROTEIN CHAIN, NOT THE ASSEMBLY. This is the filter for \"a receptor under 100 kDa\". complex_kda is the whole complex and is almost always heavier, because the receptor is solved bound to its G protein and often a nanobody. Using complex_kda for a per-molecule question returns the wrong structures. Same field name and meaning as EmpiarCatalog.","STATE COVERAGE FROM cat.coverage(), NEVER FROM MEMORY. The catalog records how many released entries exist and how many it indexes, measured when it was built. Quote those numbers. Do not tell a user the catalog covers \"everything\" or \"all ~60,900 entries\" unless coverage() says so. A previous archive card promised a catalog over \"all ~3,000 entries\" when it held 12, and the assistant repeated that claim to a scientist who then could not find anything.","SEARCH WIDENS ITSELF WHEN A QUERY IS THIN, AND SAYS SO. A query matching fewer than 25 entries falls back to synonym expansion, so \"cannabinoid receptor\" reaches related GPCR structures. cat.last_query_expanded is True when that happened — mention it, because the extra rows are related structures rather than literal matches. A well-served query (\"rhodopsin\", 40 hits) is NOT widened and returns only real rhodopsin.","RAW DATA IS A FILTER, NOT A JOIN. has_raw_data=True answers \"can I reprocess this\" in one search, and max_raw_gb scopes it to what the user can actually download; sort=\"raw_size\" puts the cheapest first. Only 8.2% of EMDB has public raw movies (5,007 of 60,895) because most groups never deposit them — say that rather than implying the rest is missing. The linked datasets are LARGE: of the 25 sub-3A GPCR structures with public movies, NONE is under 1 TB and the cheapest is 1,331 GB. Never imply Scigantic will fetch a multi-terabyte EMPIAR set; for that, point the user at EMPIAR bulk channels (Globus preferred). EMDB itself does not record the EMPIAR link — it comes from EMPIAR cross-references, so it exists only where EMPIAR listed it.","ACQUISITION FIELDS COME FROM THE DEPOSIT, AND THEY ARE NOT EQUALLY COMPLETE. voltage_kv, detector, dose_e_a2 and particles_used are filled for 94-100% of entries and are filters (voltage_kv=, detector=, min/max_dose, min/max_particles). recording_mode, n_movies, cs_mm and recon_software are present for only 19-57% and are NOT filters: a record without them is unknown, never negative. dose_e_a2 is whatever the depositor typed: 11% of values are under 5, which are almost certainly per-frame or per-second, so use min_dose=20 when you mean a total dose. pixel_a is the pixel size of the deposited MAP, which can differ from the acquisition pixel when the map was rebinned. Read current fill rates from cat.coverage()[\"enrichment\"][\"field_fill_pct\"], not from this card.","NEVER COPY ACQUISITION SETTINGS FROM ONE DATASET TO ANOTHER. A user reprocessing a public entry needs THAT entry's voltage, pixel size, dose and Cs, taken from its own deposit (cat.entry(id)) and its movie header, and compared with each other. Nothing fails when the voltage is wrong: the CTF fit is just worse and the map stalls short. A processing run of EMPIAR-10581 used 300 kV, but its deposit (EMD-0731, a Talos Arctica) records 200 kV, and the median CTF max resolution improved from 4.6 to 3.8 A once corrected. scigantic_headers.compare_acquisition() cross-checks values from several sources and reports the disagreements before compute is spent.","MORE THAN ONE ASSEMBLY STATE: cat.datasets_with_multiple_states() looks across the sibling maps of each EMPIAR dataset for different copy numbers, point groups, or oligomer words in the titles (monomer, trimer). It ranks candidates for a person to read. It misses datasets whose states the depositor never named, and one state deposited with and without symmetry (C1 and D2) also trips it, so read the titles before telling a user a dataset contains several oligomeric states.","DO NOT LIST THE ROOT. There are ~60,900 entries and the autoindex is ~12 MB of HTML; a single os.listdir on the mount root takes ~23s and returns 60,900 names. Address entries directly by accession (EMD-0339) or find them via the EMDB REST API / the EMPIAR cross-reference. The archive is fuseMountBrowsable:false for this reason — the mount is at $SCIGANTIC_MOUNT_PATH, not symlinked into the JupyterLab file browser.","GZIPPED, AND THAT MATTERS. Maps ship as .map.gz. mrcfile needs random access, so decompress to local disk (/tmp) before opening rather than streaming the gzip through the FUSE mount. A 20-60 MB map decompresses in seconds; do not try to be clever about it.","RELATIONSHIP TO EMPIAR: EMDB is the reconstruction, EMPIAR is the raw data behind it. If the user wants to reprocess, they want EMPIAR; if they want the final volume, a mask, half-maps, or to compare published structures, they want EMDB. Entries cross-reference both ways, and the Scigantic EMPIAR catalog already carries emdb_ids per entry.","NOT EVERY ENTRY HAS EVERYTHING. map/, header/ and images/ are essentially universal; masks/, other/ (half-maps), fsc/ and metadata/ are per-deposition. Check with os.path.isdir before reading, and do not tell the user a half-map exists without looking.","FORMATS: map/ and masks/ are CCP4/MRC volumes (mrcfile reads them; gemmi also handles CCP4). header/ is XML, metadata/ is mmCIF. All preinstalled in this image.","LANGUAGE: answer in the language the user writes in, and keep it there for the rest of the conversation. Cryo-EM groups at KEK use this archive in Japanese. Keep code, identifiers and library names in English; translate the prose around them."],"format":"volumetric","generatedAt":"2026-10-28T16:48:20.417Z","layout":{"example":"EMD-22962/map/emd_22962.map.gz  →  26 MB gzipped CCP4 volume (apo CGRP receptor, 3.15 Å)","pattern":"EMD-<n>/{map,header,images,metadata,masks,other,fsc}/ — map/emd_<n>.map.gz is the primary volume (CCP4/MRC, gzipped); header/emd-<n>.xml is the full metadata; other/ holds half_map_1/2 when deposited; masks/ holds emd_<n>_msk_*.map; images/ holds the published rendering."},"library":{"name":"scigantic_emdb","repo":"https://github.com/Scigantic/scigantic-emdb","summary":"scigantic_emdb is preinstalled and is how you FIND an entry. EmdbCatalog() searches every released EMDB structure by protein, organism, resolution, method and PDB cross-reference, in memory, with no live reads. Use it before anything else: without it the only way to locate a structure is to already know its accession, because listing the mount root takes ~23s and returns 60,900 opaque directory names. load_map(acc) and slices(vol) cover reading and looking at one. mrcfile and gemmi are also present and are the right tools for anything past that — scigantic_emdb wraps them, it does not replace them. cat.coverage() reports exactly how many entries the index holds and how many released entries exist; quote THAT rather than any number in this card, and never describe the catalog as complete without checking it."},"mountPath":"/mnt/http-archive/data","overview":"EMDB (EMBL-EBI Electron Microscopy Data Bank) — ~60,900 entries, each a 3D density map determined by cryo-EM, cryo-ET or volume-EM, plus the half-maps, masks, FSC curves and metadata documenting how it was obtained. It is the PROCESSED counterpart to EMPIAR: EMPIAR holds the raw movies/micrographs, EMDB holds the reconstruction computed from them, and entries cross-reference each other (an EMPIAR entry lists its EMD accessions; an EMDB entry lists its EMPIAR source). Each entry is keyed by an EMD accession, e.g. EMD-0339. The whole tree is FUSE-mounted read-only at $SCIGANTIC_MOUNT_PATH over EBI's HTTPS endpoint via rclone — nothing is copied. Unlike EMPIAR, an entry is SMALL: a map is typically 20-60 MB gzipped, so opening one directly is entirely practical.","sampleObjects":["EMD-0339/map/emd_0339.map.gz — 51 MB, cannabinoid receptor CB1 + G protein, 3.0 Å","EMD-22962/other/emd_22962_half_map_1.map.gz — half-map 1 of 2, for FSC work","EMD-22962/masks/emd_22962_msk_1.map — deposited mask","EMD-2275/header/emd-2275.xml — full structure-determination metadata"],"starterCell":"import os\nbase = os.environ[\"SCIGANTIC_MOUNT_PATH\"]\nprint(sorted(os.listdir(f\"{base}/EMD-0339\")))  # map, header, images, metadata","starterSamplePath":"EMD-0339/map/emd_0339.map.gz","totalObjects":"~60,900 entries"},"sourceType":"http","summary":"EMDB is EMBL-EBI's public archive of 3D electron microscopy reconstructions: the density maps produced from cryo-EM, cryo-ET and volume-EM experiments, together with the…","title":"EMDB: Electron Microscopy Data Bank","updatedAt":"2026-10-28T16:48:20.777Z","visibility":"public","websiteUrl":"https://www.ebi.ac.uk/emdb/"}}