Collections

Curated dataset collections for common benchmarks and training sets. Collections provide a convenient way to fetch well-known datasets with a single function call.

Quick Start

from molfun.data.collections import fetch_collection, list_collections

# See what's available
print(list_collections())
# ["casp14", "casp15", "cameo", "pdbbind_refined", ...]

# Fetch a collection
dataset = fetch_collection("casp15", split="test")

# Check collection size before downloading
from molfun.data.collections import count_collection
n = count_collection("pdbbind_refined")
print(f"PDBbind refined set: {n} complexes")

Functions

fetch_collection

fetch_collection

fetch_collection(name: str, *, cache_dir: str | None = None, fmt: str = 'cif', max_structures: int | None = None, resolution_max: float | None = None, deduplicate: bool = False, identity: float = 0.3, workers: int = 4, progress: bool = False, storage_options: dict | None = None) -> list[str]

Fetch structures for a named collection.

Parameters:

Name	Type	Description	Default
name	str	Collection name (see list_collections()).	required
cache_dir	str | None	Override cache directory.	None
fmt	str	File format ("cif" or "pdb").	'cif'
max_structures	int | None	Override collection default max.	None
resolution_max	float | None	Override collection default resolution.	None
deduplicate	bool	If True, cluster by sequence and keep one representative.	False
identity	float	Sequence identity threshold for deduplication.	0.3
workers	int	Number of parallel download threads.	4
progress	bool	Show tqdm progress bar.	False
storage_options	dict | None	fsspec options for remote storage.	None

Returns:

Type	Description
list[str]	List of file paths to downloaded structures.

Download and return a curated dataset collection.

from molfun.data.collections import fetch_collection

# Structure prediction benchmark
ds = fetch_collection("casp15", split="test", cache_dir="./data")

# Affinity dataset
ds = fetch_collection("pdbbind_refined", version="2020")

Parameter	Type	Default	Description
name	str	required	Collection name
split	str \| None	None	Dataset split (collection-dependent)
version	str \| None	None	Dataset version
cache_dir	str \| Path	"~/.molfun/collections"	Download cache directory

Returns: Dataset appropriate for the collection type.

---

count_collection

count_collection

count_collection(name: str, resolution_max: float | None = None) -> int

Count how many structures are available in RCSB for a collection.

Makes a single lightweight HTTP request (no downloads).

Parameters:

Name	Type	Description	Default
name	str	Collection name (see list_collections()).	required
resolution_max	float | None	Override collection default resolution.	None

Returns:

Type	Description
int	Total number of matching structures in RCSB.

Return the number of entries in a collection without downloading it.

from molfun.data.collections import count_collection

n = count_collection("casp15")
print(f"CASP15 has {n} targets")

Parameter	Type	Description
name	str	Collection name

Returns: int

---

list_collections

list_collections

list_collections(tag: str | None = None) -> list[CollectionSpec]

List available collections, optionally filtered by tag.

Parameters:

Name	Type	Description	Default
tag	str | None	If given, return only collections containing this tag.	None

List all available collection names.

from molfun.data.collections import list_collections

for name in list_collections():
    n = count_collection(name)
    print(f"{name}: {n} entries")

Returns: list[str]

---

CollectionSpec

CollectionSpec dataclass

Defines a reusable protein collection query.

Dataclass describing a collection's metadata.

Field	Type	Description
name	str	Collection identifier
description	str	Human-readable description
task	str	Task type: "structure", "affinity", "property"
size	int	Number of entries
url	str	Source URL
citation	str	BibTeX citation key

---

Available Collections

Name	Task	Size	Description
casp14	structure	87	CASP14 free-modeling targets
casp15	structure	109	CASP15 free-modeling targets
cameo	structure	~500/yr	CAMEO weekly structure prediction targets
pdbbind_refined	affinity	~5,300	PDBbind refined set
pdbbind_core	affinity	~290	PDBbind core set (test benchmark)
pdbbind_general	affinity	~19,400	PDBbind general set