---
license: mit
task_categories:
  - other
tags:
  - bioinformatics
  - cheminformatics
  - drug-discovery
  - virtual-screening
  - reverse-screening
  - chemberta
  - rdkit
pretty_name: ReverseLigQ Dataset
---

# ReverseLigQ dataset (Hugging Face)

This repository contains the **ReverseLigQ** dataset files in a simplified layout designed to be loaded with a `LigandStore`/`Representation` interface (compound representations) plus organism-level auxiliary tables (ReverseLigQ metadata).

## Directory layout

```
compound_data/
  pdb_chembl/
    ligands.parquet
    reps/
      chemberta_zinc_base_768.dat
      chemberta_zinc_base_768.meta.json
      morgan_1024_r2.dat
      morgan_1024_r2.meta.json

merged_databases/
  binding_data_merged.parquet
  uncurated_binding_data.parquet
  ligs_smiles_merged.parquet

rev_ligq/
  fam_prot_dict.pkl
  ligand_lists.pkl
  ligs_fams_curated.pkl
  ligs_fams_possible.pkl
  prot_descriptions.pkl
```

## Compound data (`compound_data/pdb_chembl/`)

### `ligands.parquet`
Canonical ligand index table with a **dense integer index** (`lig_idx`) used to align all representations on disk.

Typical columns:
- `chem_comp_id`: unified ligand ID (PDB CCD or ChEMBL)
- `smiles`: canonical SMILES
- `inchikey`: optional (may be missing)
- `lig_idx`: dense index **0..N-1** (row order for the `.dat` matrices)

### Representations (`reps/`)
Each representation is stored as:
- `<rep_name>.dat`: memory-mapped matrix on disk
- `<rep_name>.meta.json`: metadata (dtype, dim, packed_bits, etc.)

Available representations:
- `chemberta_zinc_base_768`: ChemBERTa embeddings (dim=768), dense float matrix.
- `morgan_1024_r2`: Morgan fingerprints (1024 bits, radius=2), stored with `packed_bits=true`.

## Detailed binding data (merged_databases/)
Known binding data from PDB and ChEMBL.

## Organism-specific tables (`rev_ligq/`)

These files provide organism-level ligand lists, Pfam-based protein families, and optional protein descriptions used to project ligand-level similarity hits into candidate protein targets.

- `ligand_lists.pkl`: dict `{organism_key (str): [chem_comp_id, ...]}`
- `ligs_fams_curated.pkl`: dict `{chem_comp_id: [pfam_id, ...]}` (curated evidence)
- `ligs_fams_possible.pkl`: dict `{chem_comp_id: [pfam_id, ...]}` (possible/uncurated evidence)
- `fam_prot_dict.pkl`: nested dict `{organism_key: {pfam_id: [uniprot_id, ...]}}`
- `prot_descriptions.pkl`: protein descriptions (when available)

### Organism keys

ReverseLigQ integrates multiple organisms, each identified by an integer key:

| Key | Organism |
|---:|---|
| 1 | *Bartonella bacilliformis* |
| 2 | *Klebsiella pneumoniae* |
| 3 | *Mycobacterium tuberculosis* |
| 4 | *Trypanosoma cruzi* |
| 5 | *Staphylococcus aureus* RF122 |
| 6 | *Streptococcus uberis* 0140J |
| 7 | *Enterococcus faecium* |
| 8 | *Escherichia coli* MG1655 |
| 9 | *Streptococcus agalactiae* NEM316 |
| 10 | *Pseudomonas syringae* |
| 11 | DENV (Dengue virus) |
| 12 | SARS-CoV-2 |
| 13 | *Homo sapiens* |

## Citation

If you use these datasets, please cite:

Schottlender G, Prieto JM, Palumbo MC, Castello FA, Serral F, Sosa EJ, Turjanski AG, Martí MA and Fernández Do Porto D (2022). *From drugs to targets: Reverse engineering the virtual screening process on a proteomic scale.* Front. Drug. Discov. 2:969983. doi: 10.3389/fddsv.2022.969983