Cheminformatics¶

← Back to Tool Reference

These tools work on the chemistry of small molecules: converting molecule files between formats, building 3-D coordinates, and computing descriptors. They read a compounds stream (the SMILES/codes a Ligand or CompoundLibrary carries) and may also read or write a structures stream (the actual coordinate files). See The Ligand Contract for why chemistry and coordinates live in two separate streams.

OpenBabel¶

Converts a molecule stream between chemical file formats, optionally adding hydrogens (pH-aware) and generating 3-D coordinates. This is the standard way to turn a Ligand (which carries only SMILES) into a docking-ready 3-D file: OpenBabel(compounds=lig, convert_3d="sdf").

References: https://github.com/openbabel/openbabel

Environment: biopipelines (on Colab, install the openbabel pip extra).

Parameters: - compounds: DataStream | StandardizedOutput = None — Source compounds (SMILES). Mutually exclusive with structures. - structures: DataStream | StandardizedOutput = None — Source coordinate files. Mutually exclusive with compounds. - convert_3d: str = None — Target 3-D coordinate format ("sdf", "mol2", "mol", "pdb", "pdbqt", "xyz"). Produces a structures stream. - convert_1d: str | List[str] = None — Line/notation format(s) ("smi", "inchi", "cml"). Added as columns on the compounds stream. - add_hydrogens: bool = False — Add explicit hydrogens. - pH: float = None — Add hydrogens at the given pH (implies add_hydrogens=True). - gen3d: bool = False — Generate 3-D coordinates from a SMILES (compounds input only). - gen3d_quality: str = "medium" — Embedding effort: "fastest", "fast", "medium", "better", "best". - minimize: bool = False — Force-field geometry minimization after embedding. - ff: str = "MMFF94" — Force field for minimization ("MMFF94", "MMFF94s", "UFF", "GAFF", "Ghemical"). - minimize_steps: int = 500 — Maximum minimization iterations. - use_structure_template: bool = True — When converting from a structures input, use the input coordinates/connectivity as a bond-order template rather than re-perceiving from scratch.

Streams: - structures — coordinate files (when convert_3d or structures input is used). - compounds — chemistry passthrough (with extra columns when convert_1d is used).

Example:

from biopipelines.openbabel import OpenBabel
from biopipelines.entities import Ligand

# Turn a SMILES-only Ligand into a 3-D SDF for docking-adjacent tools
aspirin = Ligand("aspirin")
sdf = OpenBabel(compounds=aspirin, convert_3d="sdf")
# sdf.streams.structures -> the SDF; sdf.streams.compounds -> chemistry passthrough

# Protonate a ligand at physiological pH
protonated = OpenBabel(compounds=aspirin, convert_3d="sdf", pH=7.4)

RDKit¶

Computes per-compound cheminformatics descriptors from SMILES — molecular weight, logP, TPSA, hydrogen-bond donors/acceptors, rotatable bonds, QED, fraction sp³, and more. Useful for filtering or annotating a compound library before screening.

References: https://github.com/rdkit/rdkit

Environment: biopipelines (RDKit is pinned there — no extra installation).

Parameters: - compounds: DataStream | StandardizedOutput (required) — Input compounds; SMILES are read from the compounds map_table. - descriptors: List[str] = None — Subset of descriptor names to compute. None computes the default wide set. - morgan_fp: bool = False — Also emit a Morgan fingerprint per compound.

Tables: - descriptors:

Example:

from biopipelines.rdkit_descriptors import RDKit
from biopipelines.compound_library import CompoundLibrary
from biopipelines.panda import Panda

library = CompoundLibrary("my_library.csv")
desc = RDKit(compounds=library)

# Lipinski-style filter downstream
druglike = Panda(
    tables=desc.tables.descriptors,
    operations=[Panda.filter("MW < 500 and logP < 5 and HBD <= 5 and HBA <= 10")],
)