OpenBind on Fragalysis: FatA and DENV-2/ZIKV RdRp fragment-screen structures: sample notebook

A worked notebook for OpenBind on Fragalysis: FatA and DENV-2/ZIKV RdRp fragment-screen structures, with 8 code cells and the output each one printed. On Scigantic it opens in a hosted JupyterLab with the dataset already mounted, so nothing is downloaded first.

Provider
OpenBind Consortium (Diamond Light Source XChem)
License
Hosted with OpenBind's permission; no published license (credit OpenBind and Diamond Light Source)
Size
68 objects, 4.46 GB in the bucket
Format
Parquet, JSON (in sampled files)
Stored at
Amazon S3, s3://scigantic-openbind-fragalysis (us-east-1)
Access
Free and public. No account with the data provider is needed.

The notebook

Start here. Click the grey cell below and press Shift+Enter. It reads one target's events table off the archive and takes about a second. Then change TARGET to Flavi_NS5_RdRp and run it again.

# Start here: click this cell, press Shift+Enter.
import os
from pathlib import Path
import duckdb

TARGET = "FatA"          # try "Flavi_NS5_RdRp", then run again

mount = Path(os.environ.get("SCIGANTIC_MOUNT_PATH") or Path.home() / "archive")
RELEASE = {"FatA": "v3.1-2026-07-28", "Flavi_NS5_RdRp": "v3.1-2026-09-01"}[TARGET]
events = mount / TARGET / RELEASE / "parquet" / "events.parquet"

n, crystals, compounds = duckdb.sql(
    f"SELECT count(*), count(distinct dataset), count(distinct smiles) FROM '{events}' WHERE NOT is_pdb_reference").fetchone()
print(f"{TARGET}: {n} ligand observations in {crystals} crystals, {compounds} distinct compounds")

best = duckdb.sql(f"""
    SELECT observation_id, smiles, resolution_a FROM '{events}'
    WHERE NOT is_pdb_reference ORDER BY resolution_a LIMIT 1""").fetchone()
print(f"best resolution: {best[2]:.2f} A, {best[0]}")
print(best[1])

OpenBind on Fragalysis: FatA and the dengue 2 / Zika polymerases

Crystallographic fragment screens that the OpenBind consortium published on Diamond Light Source's Fragalysis platform. FatA is a plant fatty acid thioesterase and a herbicide target. Flavi_NS5_RdRp holds the NS5 RNA polymerase of dengue virus 2 and Zika virus, two different proteins in one Fragalysis target.

Hosted with OpenBind's permission. OpenBind has not published a license for these datasets, so credit OpenBind and the Diamond Light Source XChem facility. There are no affinity measurements here, only structures.

The archive has one folder per target. Each holds parquet/ (four tables), raw/ (the upload tarballs exactly as Fragalysis serves them, with the electron-density maps), metadata/ and _SOURCE.json. This notebook reads events (one row per ligand observation), structures (the aligned structures as text) and sites (binding sites).

import duckdb, numpy as np, pandas as pd
import matplotlib.pyplot as plt

FATA = mount / "FatA" / "v3.1-2026-07-28" / "parquet"
FLAVI = mount / "Flavi_NS5_RdRp" / "v3.1-2026-09-01" / "parquet"
con = duckdb.connect()
for prefix, d in (("fata", FATA), ("flavi", FLAVI)):
    for f in sorted(d.glob("*.parquet")):
        con.execute(f"CREATE OR REPLACE VIEW {prefix}_{f.stem} AS SELECT * FROM read_parquet('{f}')")

for t in ("fata_events", "fata_structures", "flavi_events", "flavi_structures"):
    print(f"{t}: {con.sql(f'SELECT count(*) FROM {t}').fetchone()[0]:,} rows")

Output:

fata_events: 402 rows
fata_structures: 402 rows
flavi_events: 402 rows
flavi_structures: 402 rows

Observations, crystals and compounds

An observation is one ligand modelled in one crystal. A crystal can hold several ligand copies or sites, so there are more observations than crystals, and a compound can appear in several crystals. Flavi's 24 is_pdb_reference rows are public PDB structures used to align the others; they are left out of the compound counts.

con.sql("""
    SELECT target, CASE WHEN is_pdb_reference THEN 'PDB reference' ELSE coalesce(soakdb_ProteinName, '(no SoakDB row)') END AS protein,
           count(*) AS observations, count(distinct dataset) AS crystals, count(distinct smiles) AS compounds
    FROM (SELECT * FROM fata_events UNION ALL BY NAME SELECT * FROM flavi_events)
    GROUP BY 1, 2 ORDER BY 1, 2
""").df()

Output:

           target          protein  observations  crystals  compounds
0            FatA  (no SoakDB row)             4         3          3
1            FatA          FATAATA           398       163        162
2  Flavi_NS5_RdRp        DNV2_NS5A           120        96         95
3  Flavi_NS5_RdRp    PDB reference            24        20         16
4  Flavi_NS5_RdRp        Zika_NS5A           258       144        141

Fragments and follow-on compounds in FatA

FatA's SoakDB library names split the observations into fragment-screen hits (libraries DSIPoised and EnamineEF) and follow-on compounds (KateKFUs, plus four crystals with no library). Those counts match what OpenBind states for FatA.

con.sql("""
    SELECT CASE WHEN soakdb_LibraryName IN ('DSIPoised', 'EnamineEF') THEN 'fragment' ELSE 'follow-on' END AS kind,
           coalesce(soakdb_LibraryName, '(no library)') AS library,
           count(*) AS observations, count(distinct smiles) AS compounds
    FROM fata_events GROUP BY 1, 2 ORDER BY 1, 3 DESC
""").df()

Output:

        kind       library  observations  compounds
0  follow-on      KateKFUs            87         34
1  follow-on  (no library)             4          3
2   fragment     DSIPoised           197         86
3   fragment     EnamineEF           114         42

Resolution and alignment quality

Resolution and R-factors come from each crystal's PDB header. chain_align_rmsd_a is how well each chain aligned onto the target's reference. The Zika chains are aligned onto a dengue 2 based reference, so their alignment RMSD is much larger.

con.sql("""
    SELECT target, CASE WHEN is_pdb_reference THEN 'PDB reference' ELSE coalesce(soakdb_ProteinName, '(no SoakDB row)') END AS protein, count(*) AS observations,
           round(min(resolution_a), 2) AS res_min, round(median(resolution_a), 2) AS res_median, round(max(resolution_a), 2) AS res_max,
           round(median(r_free), 3) AS r_free_median, round(median(chain_align_rmsd_a), 2) AS align_rmsd_median
    FROM (SELECT * FROM fata_events UNION ALL BY NAME SELECT * FROM flavi_events)
    GROUP BY 1, 2 ORDER BY 1, 2
""").df()

Output:

           target          protein  observations  res_min  res_median  \
0            FatA  (no SoakDB row)             4     1.76        1.86   
1            FatA          FATAATA           398     1.37        1.82   
2  Flavi_NS5_RdRp        DNV2_NS5A           120     1.49        1.75   
3  Flavi_NS5_RdRp    PDB reference            24     1.40        2.10   
4  Flavi_NS5_RdRp        Zika_NS5A           258     1.43        1.84   

   res_max  r_free_median  align_rmsd_median  
0     1.89          0.313               0.43  
1     2.87          0.282               0.34  
2     2.43          0.243               0.87  
3     3.00          0.224               2.63  
4     2.49          0.271               4.99

Where the ligands bind

The aligner groups observations into canonical binding sites. Most FatA observations share one site.

con.sql("""
    SELECT canonical_site_id, count(*) AS observations, count(distinct dataset) AS crystals,
           count(*) FILTER (WHERE soakdb_LibraryName IN ('DSIPoised', 'EnamineEF')) AS fragment_obs
    FROM fata_events GROUP BY 1 ORDER BY 2 DESC
""").df()

Output:

         canonical_site_id  observations  crystals  fragment_obs
0  FATAATA-x4223+A+459+0+1           308       138           223
1  FATAATA-x1147+A+360+0+1            42        21            42
2  FATAATA-x1985+A+360+0+1            39        19            39
3  FATAATA-x1285+B+403+0+1             5         4             5
4  FATAATA-x4024+A+503+0+1             2         1             0
5  FATAATA-x4223+A+461+0+1             2         1             0
6  FATAATA-x4223+A+462+0+1             2         1             0
7  FATAATA-x1142+A+360+0+1             2         1             2

Ten follow-on compounds from FatA

Drawn in 2D from the stored ligand SDF, one row per compound (the best-resolved crystal).

from rdkit import Chem
from rdkit.Chem import Draw, AllChem

pick = con.sql("""
    SELECT DISTINCT ON (smiles) observation_id, smiles, resolution_a
    FROM fata_events WHERE soakdb_LibraryName = 'KateKFUs'
    ORDER BY smiles, resolution_a
""").df().sort_values("resolution_a").head(10)
sdf = con.sql("SELECT observation_id, ligand_sdf FROM fata_structures").df().set_index("observation_id").ligand_sdf
mols, legends = [], []
for r in pick.itertuples():
    m = Chem.MolFromMolBlock(sdf[r.observation_id])
    AllChem.Compute2DCoords(m)            # 2D drawing only; the stored pose stays 3D
    mols.append(m); legends.append(f"{r.observation_id.split('_')[0][-5:]}  {r.resolution_a:.2f} A")
Draw.MolsToGridImage(mols, molsPerRow=5, subImgSize=(230, 190), legends=legends)

One structure, read

structures.parquet holds each aligned structure as text, so no file is opened. gemmi parses the best-resolved FatA follow-on, and the pocket is the set of protein residues within 4 A of the ligand.

import gemmi

obs = pick.iloc[0].observation_id
ev = con.sql(f"SELECT chain, ligand_res_seq, resolution_a, canonical_site_id FROM fata_events WHERE observation_id = '{obs}'").fetchone()
text = con.sql(f"SELECT aligned_pdb FROM fata_structures WHERE observation_id = '{obs}'").fetchone()[0]
st = gemmi.read_pdb_string(text)
model = st[0]
lig = next(r for ch in model if ch.name == ev[0] for r in ch if r.name == "LIG" and r.seqid.num == ev[1])
ns = gemmi.NeighborSearch(model, st.cell, 5).populate()
pocket = set()
for a in lig:
    for mark in ns.find_atoms(a.pos, "\0", radius=4.0):
        cra = mark.to_cra(model)
        if cra.residue.het_flag != "H":
            pocket.add((cra.chain.name, cra.residue.seqid.num, cra.residue.name))
print(f"{obs}: resolution {ev[2]:.2f} A, ligand {len(lig)} atoms, site {ev[3]}")
print(f"{len(pocket)} protein residues within 4 A:", ", ".join(f"{n}{i}" for _, i, n in sorted(pocket)))

Output:

FATAATA-x4034_A_459_0_1_FATAATA-x4223+A+459+0+1: resolution 1.37 A, ligand 21 atoms, site FATAATA-x4223+A+459+0+1
15 protein residues within 4 A: HIS117, ALA118, VAL121, PHE123, ALA129, TRP141, CYS166, GLN167, SER168, THR174, ARG176, TRP196, ARG214, TYR217, CYS221

Things to know when using this data

you split data.

the electron-density maps.

Run it

Open this notebook to run it, see the dataset page for the full description, or browse the full catalog.