A team led by Mila researcher Jian Tang launches TorchDrug, an open-source platform for drug discovery

Jian Tang, Mila faculty member and Assistant Professor at HEC Montréal, as well as Mila students and collaborators from Tsinghua University, Peking University, and Shanghai Jiao Tong University, today announced the launch of TorchDrug—an open-source machine learning platform aimed at making AI drug discovery software and libraries freely available to the research community. TorchDrug is one of the first known open-source platforms allowing ML practitioners to use, contribute, and apply fundamental algorithms to drug discovery applications, with minimal efforts on domain knowledge and trivial details.

Initiated by Professor Tang’s research group MilaGraph, the creators of TorchDrug integrated techniques ranging from graph machine learning (graph neural networks, geometric deep learning & knowledge graphs) and deep generative models to reinforcement learning. Its primary features include minimal domain knowledge, datasets and building blocks, comprehensive benchmarks, and scalable training and inference models. Current tutorials include property prediction, pretrained molecular representations, and de novo molecular design and optimization and more. 

Over the past decade, many pharmaceutical and biotech companies have refocused their strategies to harness the potential of artificial intelligence (AI) for drug discovery and development. AI can offer a higher level of precision to the complex and lengthy discovery process in drug development, which ultimately leads to faster development timelines and significantly reduced costs. 

While AI for drug discovery continues to live up to the hype, according to Professor Tang, the future of drug discovery relies on fostering a rich open-source community. “We hope this platform can accelerate the process of drug discovery by bringing researchers from both the machine learning and biomedicine communities together and become the leading open-source platform for machine learning for drug discovery in the future.”

The team plans to further extend the platform to 3D structure modeling and generation with geometric deep learning methods. 

Click here to learn more about TorchDrug.