molfeat - the hub for all your molecular featurizers
Molfeat is a hub of molecular featurizers. It supports a wide variety of out-of-the-box molecular featurizers and can be easily extended to include your own custom featurizers.
- π Fast, with a simple and efficient API.
- π Unify pre-trained molecular embeddings and hand-crafted featurizers in a single package.
- β Easily add your own featurizers through plugins.
- π Benefit from increased performance through a trouble-free caching system.
Visit our website at https://molfeat.datamol.io.
Use mamba:
mamba install -c conda-forge molfeat
Tips: You can replace mamba
by conda
.
Note: We highly recommend using a Conda Python distribution to install Molfeat. The package is also pip installable if you need it: pip install molfeat
.
Not all featurizers in the Molfeat core package are supported by default. Some featurizers require additional dependencies. If you try to use a featurizer that requires additional dependencies, Molfeat will raise an error and tell you which dependencies are missing and how to install them.
- To install
dgl
: runmamba install -c dglteam dgl
- To install
dgllife
: runmamba install -c conda-forge dgllife
- To install
fcd_torch
: runmamba install -c conda-forge fcd_torch
- To install
pyg
: runmamba install -c conda-forge pytorch_geometric
- To install
graphormer-pretrained
: runmamba install -c conda-forge graphormer-pretrained
- To install
map4
: see https://github.com/reymond-group/map4 - To install
bio-embeddings
: runmamba install -c conda-forge 'bio-embeddings >=0.2.2'
If you install Molfeat using pip, there are optional dependencies that can be installed with the main package. For example, pip install "molfeat[all]"
allows installing all the compatible optional dependencies for small molecule featurization. There are other options such as molfeat[dgl]
, molfeat[graphormer]
, molfeat[transformer]
, molfeat[viz]
, and molfeat[fcd]
. See the optional-dependencies for more information.
The functionality of Molfeat can be extended through plugins. The use of a plugin system ensures that the core package remains easy to install and as light as possible, while making it easy to extend its functionality with plug-and-play components. Additionally, it ensures that plugins can be developed independently from the core package, removing the bottleneck of a central party that reviews and approves new plugins. Consult the molfeat documentation for more details on how to create your own plugins.
However, this does imply that the installation of a plugin is plugin-dependent: please consult the relevant documentation to learn more.
import datamol as dm
from molfeat.calc import FPCalculator
from molfeat.trans import MoleculeTransformer
from molfeat.store.modelstore import ModelStore
# Load some dummy data
data = dm.data.freesolv().sample(100).smiles.values
# Featurize a single molecule
calc = FPCalculator("ecfp")
calc(data[0])
# Define a parallelized featurization pipeline
mol_transf = MoleculeTransformer(calc, n_jobs=-1)
mol_transf(data)
# Easily save and load featurizers
mol_transf.to_state_yaml_file("state_dict.yml")
mol_transf = MoleculeTransformer.from_state_yaml_file("state_dict.yml")
mol_transf(data)
# List all available featurizers
store = ModelStore()
store.available_models
# Find a featurizer and learn how to use it
model_card = store.search(name="ChemBERTa-77M-MLM")[0]
model_card.usage()
Please cite Molfeat if you use it in your research: .
See developers for a comprehensive guide on how to contribute to Molfeat
See the latest changelogs at CHANGELOG.rst.
- @cwognum
- @maclandrol
- @hadim
Under the Apache-2.0 license. See LICENSE.