TorchDrug: a python library for drug design

Hi! After a short break, I’m back with more posts! Today, let’s discuss TorchDrug, a freshly-released open-source python library.

Recently, a group of scientists from MILA, a Canadian research institute, has made TorchDrug publicly available. The authors say the target audience for this tool is machine learning professionals who want to quickly build a model or two with a view to developing new drug molecules.

The authors point out the following advantages of their tool:

• It does not require domain knowledge in chemistry
• Provides pre-processed data sets for model training and benchmarking
• Provides the user with pre-compiled building blocks of complex neural networks (in fact, ready-made layers) from which you can easily aseemble a custom model
• Makes it easy to distribute the computations across any number of CPUs and GPUs.

At the moment, TorchDrug can be used to predict molecular properties (QSAR), extract molecular representations (embeddings), do de novo molecular design and optimization, work with chemical reactions and retrosynthesis, and fill in missing links between nodes in a given biomedical graph (Biomedical Knowledge Graph Reasoning). The authors aim at adding support for protein representation learning soon.

All the code, tutorials and documentation of TorchDrug can be found at https://torchdrug.ai/.

That’s all for today’s short review of a new chemoinformatics tool! Did you like the tool? Let me know on Medium, Twitter or Instagram.

Thank you for your kind attention and feedback. See you next time. Cheers!