benchmol 0.0.1

A Multi-Modality Benchmarking Platform For Molecular Representation

A Multi-Modality Benchmarking Platform For Molecular Representation

Supporting ADMET, QSAR, Virtual Screening and More

Built on PyTorch, benchmol is an easy-to-use and extensible Python package for drug discovery.

News !

• [2024/09/17] benchmol is released on GitHub.

Features !

• Supports Property Prediction, ADMET, QSAR, Virtual Screening and More

• Supports 7 different modalities of molecules, including fingerprint, sequence, graph, geometry graph, image, geometry image, and video.

• Supports a large number of baselines for molecular data of different modalities.

• Two novel benchmarks: MBANet and StructNet.

Installation

From PyPi

conda create --name test_env python=3.9

pip install benchmol

• If the dependencies are not automatically installed, use the following command to initialize the dependency environment:

pip install -r requirements_pip.txt

Benchmarks

BenchMol provides two new benchmarks, MBANet and StructNet.

MBANet

| Name | Link | Description |

| ------ | ------------------------------------------------------------ | ------------------------------------------------------------ |

| MBANet | OneDrive | Complete MBANet data, including csv files of molecules and data of different modalities processed |

StructNet

| Name | Link | Description |

| --------------------------- | ------------------------------------------------------------ | ------------------------------------------------------------ |

| Raw Data | OneDrive | > 10 million molecules from CHEMBL 34 |

| Raw Data of StructNet | OneDrive | The original data of StructNet includes csv files of 60 datasets. |

| Processed Data of StructNet | OneDrive | Processed data, including geometry, image, video and other multi-modal data |

Tutorials

We provide examples of using benchmol, please see below:

Feature Extraction

The following shows a use case of extracting features from different modalities with benchmol:

| Description | Tutorial Links |

| ------------------------------------------------------------ | ------------------------------------------------------------ |

| Extracting Molecular Fingerprints | 1_extract_fp_features.ipynb |

| Extracting features from sequence using un-pretrained CHEM-BERT | 1_extract_sequence_features.ipynb |

| Extracting features from geometry image using IEM | 1_extract_geometry_image_features.ipynb |

| Extracting features from graph using GIN | 1_extract_graph_features.ipynb |

| Extracting features from molecular image using ImageMol | 1_extract_image_features.ipynb |

| Extracting features from video using VideoMol | 1_extract_video_features.ipynb |

Linear Probing

Use case for linear probing is provided with benchmol: 2_linear_probing.ipynb

Fine-tuning

Use case for fine-tuning is provided with benchmol:

| Description | Tutorial Links |

|------------------------------------------|-----------------------------------------------------------------------|

| Fine-tuning with sequence modality | 3_fine_tuning_sequence.ipynb |

| Fine-tuning with graph modality | 3_fine_tuning_graph.ipynb |

| Fine-tuning with geometry graph modality | 3_fine_tuning_geometry.ipynb |

| Fine-tuning with image modality | 3_fine_tuning_image.ipynb |

| Fine-tuning with geometry image modality | 3_fine_tuning_geometry_image.ipynb |

| Fine-tuning with geometry video modality | 3_fine_tuning_video.ipynb |

Releases

For more information on BenchMol versions, see the Releases page.

Reference

If you find our code or anything else helpful, please do not hesitate to cite the following relevant papers:

Acknowledge