# MolGraph
**Graph Neural Networks** with **TensorFlow** and **Keras**. Focused on **Molecular Machine Learning**.
<img src="https://github.com/akensert/molgraph/blob/main/media/molgraph.jpg" alt="molgraph" width="800">
## Highlights
Build a Graph Neural Network with Keras' [Sequential](https://www.tensorflow.org/api_docs/python/tf/keras/Sequential) API:
```python
from molgraph import GraphTensor
from molgraph import layers
from tensorflow import keras
model = keras.Sequential([
layers.GINConv(units=32),
layers.GINConv(units=32),
layers.Readout(),
keras.layers.Dense(units=1),
])
output = model(
GraphTensor(node_feature=[[4.], [2.]], edge_src=[0], edge_dst=[1])
)
```
## Paper
See [arXiv](https://arxiv.org/abs/2208.09944)
## Documentation
See [readthedocs](https://molgraph.readthedocs.io/en/latest/)
## Implementations
- **Graph tensor** ([GraphTensor](http://github.com/akensert/molgraph/tree/main/molgraph/tensors/graph_tensor.py))
- A composite tensor holding graph data.
- Has a ragged state (multiple graphs) and a non-ragged state (single disjoint graph).
- Can conveniently go between both states (merge(), separate()).
- Can propagate node states (features) based on edges (propagate()).
- Can add, update and remove graph data (update(), remove()).
- Compatible with TensorFlow's APIs (including Keras). For instance, graph data (encoded as a GraphTensor) can now seamlessly be used with keras.Sequential, keras.Functional, tf.data.Dataset, and tf.saved_model APIs.
- **Layers**
- **Convolutional**
- GCNConv ([GCNConv](http://github.com/akensert/molgraph/tree/main/molgraph/layers/convolutional/gcn_conv.py))
- GINConv ([GINConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/convolutional/gin_conv.py))
- GCNIIConv ([GCNIIConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/convolutional/gcnii_conv.py))
- GraphSageConv ([GraphSageConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/convolutional/graph_sage_conv.py))
- **Attentional**
- GATConv ([GATConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/attentional/gat_conv.py))
- GATv2Conv ([GATv2Conv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/attentional/gatv2_conv.py))
- GTConv ([GTConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/attentional/gt_conv.py))
- GMMConv ([GMMConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/attentional/gmm_conv.py))
- GatedGCNConv ([GatedGCNConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/attentional/gated_gcn_conv.py))
- AttentiveFPConv ([AttentiveFPConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/attentional/attentive_fp_conv.py))
- **Message-passing**
- MPNNConv ([MPNNConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/message_passing/mpnn_conv.py))
- EdgeConv ([EdgeConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/message_passing/edge_conv.py))
- **Distance-geometric**
- DTNNConv ([DTNNConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/geometric/dtnn_conv.py))
- GCFConv ([GCFConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/geometric/gcf_conv.py))
- **Pre- and post-processing**
- In addition to the aforementioned GNN layers, there are also several other layers which improves model-building. See [readout/](https://github.com/akensert/molgraph/tree/main/molgraph/layers/readout), [preprocessing/](https://github.com/akensert/molgraph/tree/main/molgraph/layers/preprocessing), [postprocessing/](https://github.com/akensert/molgraph/tree/main/molgraph/layers/postprocessing), [positional_encoding/](https://github.com/akensert/molgraph/tree/main/molgraph/layers/positional_encoding).
- **Models**
- Although model building is easy with MolGraph, there are some built-in GNN [models](https://github.com/akensert/molgraph/tree/main/molgraph/models):
- **GIN**
- **MPNN**
- **DMPNN**
- And models for improved interpretability of GNNs:
- **SaliencyMapping**
- **IntegratedSaliencyMapping**
- **SmoothGradSaliencyMapping**
- **GradientActivationMapping** (Recommended)
## Requirements/dependencies
- **Python** (version ~= 3.10)
- **TensorFlow** (version ~= 2.15.0)
- **RDKit** (version ~= 2022.3.5)
- **Pandas** (version ~= 1.0.3)
- **IPython** (version ~= 8.12.0)
> MolGraph should work with the more recent TensorFlow and RDKit versions. If not, try installing earlier versions of TensorFlow and RDKit.
## Installation
For **GPU** users:
<pre>
pip install molgraph[gpu]
</pre>
For **CPU** users:
<pre>
pip install molgraph
</pre>
Now run your first program with **MolGraph**:
```python
from tensorflow import keras
from molgraph import chemistry
from molgraph import layers
from molgraph import models
# Obtain dataset, specifically ESOL
qm7 = chemistry.datasets.get('esol')
# Define molecular graph encoder
atom_encoder = chemistry.Featurizer([
chemistry.features.Symbol(),
chemistry.features.Hybridization(),
# ...
])
bond_encoder = chemistry.Featurizer([
chemistry.features.BondType(),
# ...
])
encoder = chemistry.MolecularGraphEncoder(atom_encoder, bond_encoder)
# Obtain graphs and associated labels
x_train = encoder(qm7['train']['x'])
y_train = qm7['train']['y']
x_test = encoder(qm7['test']['x'])
y_test = qm7['test']['y']
# Build model via Keras API
gnn_model = keras.Sequential([
layers.GATConv(units=32, name='gat_conv_1'),
layers.GATConv(units=32, name='gat_conv_2'),
layers.Readout(),
keras.layers.Dense(units=1024, activation='relu'),
keras.layers.Dense(units=y_train.shape[-1])
])
# Compile, fit and evaluate
gnn_model.compile(optimizer='adam', loss='mae')
gnn_model.fit(x_train, y_train, epochs=50)
scores = gnn_model.evaluate(x_test, y_test)
# Compute gradient activation maps
gam_model = models.GradientActivationMapping(
model=gnn_model, layer_names=['gat_conv_1', 'gat_conv_2'])
maps = gam_model(x_train.separate())
```
## Changelog
For a detailed list of changes, see the [CHANGELOG.md](https://github.com/akensert/molgraph/blob/main/CHANGELOG.md).
Raw data
{
"_id": null,
"home_page": "https://github.com/akensert/molgraph",
"name": "molgraph",
"maintainer": "",
"docs_url": null,
"requires_python": ">=3.10",
"maintainer_email": "",
"keywords": "graph-neural-networks,deep-learning,machine-learning,molecular-machine-learning,molecular-graphs,graphs,cheminformatics,chemometrics,bioinformatics,chemistry,biology,biochemistry",
"author": "Alexander Kensert",
"author_email": "alexander.kensert@gmail.com",
"download_url": "https://files.pythonhosted.org/packages/63/98/d0955d28e9e05f29b3a5c0c8270ef08929c61dfb65f939873c79c7bda3d1/molgraph-0.6.5.tar.gz",
"platform": null,
"description": "# MolGraph\n\n**Graph Neural Networks** with **TensorFlow** and **Keras**. Focused on **Molecular Machine Learning**.\n\n<img src=\"https://github.com/akensert/molgraph/blob/main/media/molgraph.jpg\" alt=\"molgraph\" width=\"800\">\n\n## Highlights\n\nBuild a Graph Neural Network with Keras' [Sequential](https://www.tensorflow.org/api_docs/python/tf/keras/Sequential) API:\n\n```python\nfrom molgraph import GraphTensor\nfrom molgraph import layers\nfrom tensorflow import keras\n\nmodel = keras.Sequential([\n layers.GINConv(units=32),\n layers.GINConv(units=32),\n layers.Readout(),\n keras.layers.Dense(units=1),\n])\noutput = model(\n GraphTensor(node_feature=[[4.], [2.]], edge_src=[0], edge_dst=[1])\n)\n```\n\n## Paper\nSee [arXiv](https://arxiv.org/abs/2208.09944)\n\n## Documentation\nSee [readthedocs](https://molgraph.readthedocs.io/en/latest/)\n\n## Implementations\n\n- **Graph tensor** ([GraphTensor](http://github.com/akensert/molgraph/tree/main/molgraph/tensors/graph_tensor.py))\n - A composite tensor holding graph data.\n - Has a ragged state (multiple graphs) and a non-ragged state (single disjoint graph).\n - Can conveniently go between both states (merge(), separate()).\n - Can propagate node states (features) based on edges (propagate()).\n - Can add, update and remove graph data (update(), remove()).\n - Compatible with TensorFlow's APIs (including Keras). For instance, graph data (encoded as a GraphTensor) can now seamlessly be used with keras.Sequential, keras.Functional, tf.data.Dataset, and tf.saved_model APIs.\n- **Layers**\n - **Convolutional**\n - GCNConv ([GCNConv](http://github.com/akensert/molgraph/tree/main/molgraph/layers/convolutional/gcn_conv.py))\n - GINConv ([GINConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/convolutional/gin_conv.py))\n - GCNIIConv ([GCNIIConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/convolutional/gcnii_conv.py))\n - GraphSageConv ([GraphSageConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/convolutional/graph_sage_conv.py))\n - **Attentional**\n - GATConv ([GATConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/attentional/gat_conv.py))\n - GATv2Conv ([GATv2Conv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/attentional/gatv2_conv.py))\n - GTConv ([GTConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/attentional/gt_conv.py))\n - GMMConv ([GMMConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/attentional/gmm_conv.py))\n - GatedGCNConv ([GatedGCNConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/attentional/gated_gcn_conv.py))\n - AttentiveFPConv ([AttentiveFPConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/attentional/attentive_fp_conv.py))\n - **Message-passing**\n - MPNNConv ([MPNNConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/message_passing/mpnn_conv.py))\n - EdgeConv ([EdgeConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/message_passing/edge_conv.py))\n - **Distance-geometric**\n - DTNNConv ([DTNNConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/geometric/dtnn_conv.py))\n - GCFConv ([GCFConv](https://github.com/akensert/molgraph/tree/main/molgraph/layers/geometric/gcf_conv.py))\n - **Pre- and post-processing**\n - In addition to the aforementioned GNN layers, there are also several other layers which improves model-building. See [readout/](https://github.com/akensert/molgraph/tree/main/molgraph/layers/readout), [preprocessing/](https://github.com/akensert/molgraph/tree/main/molgraph/layers/preprocessing), [postprocessing/](https://github.com/akensert/molgraph/tree/main/molgraph/layers/postprocessing), [positional_encoding/](https://github.com/akensert/molgraph/tree/main/molgraph/layers/positional_encoding).\n- **Models**\n - Although model building is easy with MolGraph, there are some built-in GNN [models](https://github.com/akensert/molgraph/tree/main/molgraph/models):\n - **GIN**\n - **MPNN**\n - **DMPNN**\n - And models for improved interpretability of GNNs:\n - **SaliencyMapping**\n - **IntegratedSaliencyMapping**\n - **SmoothGradSaliencyMapping**\n - **GradientActivationMapping** (Recommended)\n\n## Requirements/dependencies\n- **Python** (version ~= 3.10)\n - **TensorFlow** (version ~= 2.15.0)\n - **RDKit** (version ~= 2022.3.5)\n - **Pandas** (version ~= 1.0.3)\n - **IPython** (version ~= 8.12.0)\n\n> MolGraph should work with the more recent TensorFlow and RDKit versions. If not, try installing earlier versions of TensorFlow and RDKit.\n\n## Installation\n\nFor **GPU** users:\n<pre>\npip install molgraph[gpu]\n</pre>\n\nFor **CPU** users:\n<pre>\npip install molgraph\n</pre>\n\nNow run your first program with **MolGraph**:\n\n```python\nfrom tensorflow import keras\nfrom molgraph import chemistry\nfrom molgraph import layers\nfrom molgraph import models\n\n# Obtain dataset, specifically ESOL\nqm7 = chemistry.datasets.get('esol')\n\n# Define molecular graph encoder\natom_encoder = chemistry.Featurizer([\n chemistry.features.Symbol(),\n chemistry.features.Hybridization(),\n # ...\n])\n\nbond_encoder = chemistry.Featurizer([\n chemistry.features.BondType(),\n # ...\n])\n\nencoder = chemistry.MolecularGraphEncoder(atom_encoder, bond_encoder)\n\n# Obtain graphs and associated labels\nx_train = encoder(qm7['train']['x'])\ny_train = qm7['train']['y']\n\nx_test = encoder(qm7['test']['x'])\ny_test = qm7['test']['y']\n\n# Build model via Keras API\ngnn_model = keras.Sequential([\n layers.GATConv(units=32, name='gat_conv_1'),\n layers.GATConv(units=32, name='gat_conv_2'),\n layers.Readout(),\n keras.layers.Dense(units=1024, activation='relu'),\n keras.layers.Dense(units=y_train.shape[-1])\n])\n\n# Compile, fit and evaluate\ngnn_model.compile(optimizer='adam', loss='mae')\ngnn_model.fit(x_train, y_train, epochs=50)\nscores = gnn_model.evaluate(x_test, y_test)\n\n# Compute gradient activation maps\ngam_model = models.GradientActivationMapping(\n model=gnn_model, layer_names=['gat_conv_1', 'gat_conv_2'])\n\nmaps = gam_model(x_train.separate())\n```\n\n## Changelog\nFor a detailed list of changes, see the [CHANGELOG.md](https://github.com/akensert/molgraph/blob/main/CHANGELOG.md).\n",
"bugtrack_url": null,
"license": "MIT",
"summary": "Graph Neural Networks for Molecular Machine Learning",
"version": "0.6.5",
"project_urls": {
"Homepage": "https://github.com/akensert/molgraph"
},
"split_keywords": [
"graph-neural-networks",
"deep-learning",
"machine-learning",
"molecular-machine-learning",
"molecular-graphs",
"graphs",
"cheminformatics",
"chemometrics",
"bioinformatics",
"chemistry",
"biology",
"biochemistry"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "da23e521768e871977dc58e52a7d1e70648110962580e0fe3cc06043bd2af1b1",
"md5": "51b17c46d64d79affb3d8ee7bb2f702c",
"sha256": "f35dcc2ea30c43be63246bc2f6f4891b30cebb125066e72de7cf5a5ef8ef732f"
},
"downloads": -1,
"filename": "molgraph-0.6.5-py3-none-any.whl",
"has_sig": false,
"md5_digest": "51b17c46d64d79affb3d8ee7bb2f702c",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.10",
"size": 197555,
"upload_time": "2024-02-09T14:18:04",
"upload_time_iso_8601": "2024-02-09T14:18:04.808094Z",
"url": "https://files.pythonhosted.org/packages/da/23/e521768e871977dc58e52a7d1e70648110962580e0fe3cc06043bd2af1b1/molgraph-0.6.5-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "6398d0955d28e9e05f29b3a5c0c8270ef08929c61dfb65f939873c79c7bda3d1",
"md5": "4eb025301cd2b120de31fa1825154e21",
"sha256": "9096a60e87528c98c15ac05ee8ae5a8763bd1cc12604360f6323746245283eab"
},
"downloads": -1,
"filename": "molgraph-0.6.5.tar.gz",
"has_sig": false,
"md5_digest": "4eb025301cd2b120de31fa1825154e21",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.10",
"size": 110458,
"upload_time": "2024-02-09T14:18:08",
"upload_time_iso_8601": "2024-02-09T14:18:08.805041Z",
"url": "https://files.pythonhosted.org/packages/63/98/d0955d28e9e05f29b3a5c0c8270ef08929c61dfb65f939873c79c7bda3d1/molgraph-0.6.5.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-02-09 14:18:08",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "akensert",
"github_project": "molgraph",
"travis_ci": false,
"coveralls": false,
"github_actions": true,
"lcname": "molgraph"
}