Fragment sets¶

A v1 database can hold several fragment sets in one file. Each set is a separate frequency column on every radiusN table, so the same deduplicated envs and frags tables are shared while each set records how often a fragment occurs within that set.

This lets one database describe, for example, how common a fragment is in ChEMBL versus in a focused in-house library, and lets you switch between those views at generation time.

Build with a single set¶

cremdb_create -i input.smi -o fragments.db -s chembl

Every fragment occurrence is counted in the chembl column.

Build with set-membership files¶

Pass one or more membership files to -s/--set-name. Each existing file is treated as a membership list, and its basename (without extension) becomes the set column name:

cremdb_create -i input_with_ids.smi -o fragments.db \
  -s set1_ids.txt set2_ids.txt all_set

• set1_ids.txt → column set1_ids
• set2_ids.txt → column set2_ids
• all_set is not an existing file, so it becomes a default set containing all molecules.

A fragment occurrence contributes to a set's count only when the source molecule's ID belongs to that set.

Membership file format¶

One molecule ID per line:

mol_0001
mol_0007
mol_0042

For membership to work, the input SMILES file must carry IDs in the second column:

CCO         mol_0001
c1ccccc1    mol_0002

Inspect the sets in a database¶

cremdb_get_set_names -i fragments.db

prints the set columns per radius table, e.g.:

radius1: ['chembl']
radius2: ['chembl']
radius3: ['chembl']

Equivalently, with SQLite:

PRAGMA table_info(radius3);

Any column other than env_id, core_smi_id, core_num_atoms, dist2, and is_ring_closure is a fragment-set frequency column.

Use a set at generation time¶

Choose the set(s) with set_names and the threshold with min_freq:

from rdkit import Chem
from crem.crem import mutate_mol

m = Chem.MolFromSmiles("c1ccccc1N")

# Fragments seen at least 5 times in set1_ids
res = list(mutate_mol(m, db_name="fragments.db", set_names="set1_ids", min_freq=5))

set_names accepts a single column name or a list. When several sets are named, a fragment is included if any of them meets min_freq (OR logic). With set_names=None (default), all set columns are considered. Naming a column that does not exist raises a ValueError listing the available set names.

# Frequent in either set
res = list(mutate_mol(m, db_name="fragments.db",
                      set_names=["set1_ids", "set2_ids"], min_freq=3))