Challenges: design, investment, time, and failure rate
On average, from development to approval, it costs over 1 billion USD and takes over 10 years to put a new drug on the market. Drug design is difficult, costly, and time-consuming, with a success rate below 10%. Drugs cannot be developed fast enough to treat new or even existing diseases. Normally, drug design relies on experts and screening can have a high failure rate. The pharmaceutical industry needs faster, more reliable ways to design and discover effective drugs.
Huawei Cloud Pangu drug molecule model significantly improves drug design efficiency.
Huawei Cloud co-developed and -trained the Pangu drug molecule model with Shanghai Institute of Materia Medica, Chinese Academy of Sciences.
-
1. Brand-new deep learning network architecture for compound feature learning
The Pangu drug molecule model uses a unique encoder architecture to convert molecular structures into numerical values, facilitating accurate predictions of drug molecule structures and attributes. This helps ensure accuracy for downstream tasks.
The Pangu drug molecule model uses a unique encoder architecture to convert molecular structures into numerical values, facilitating accurate predictions of drug molecule structures and attributes. This helps ensure accuracy for downstream tasks.
-
2. Training of a hyperscale compound feature representation model
The Pangu drug molecule model has learned the chemical structure of 1.7 billion drug-like molecules in the market. Through unsupervised learning, this model can optimize such important metrics as reconstruction, validity, and uniqueness of molecular structures, significantly outperforming other methods used for similar purposes.
The Pangu drug molecule model has learned the chemical structure of 1.7 billion drug-like molecules in the market. Through unsupervised learning, this model can optimize such important metrics as reconstruction, validity, and uniqueness of molecular structures, significantly outperforming other methods used for similar purposes.
-
3. A library of 100 million new compounds
The molecule generator set up a library of 100 million novel drug-like small molecules, with a structural novelty of 99.68%. These molecules can be used to quickly generate new compounds with similar attributes, creating endless possibilities for new drug discovery. In addition, the Pangu molecular optimizer helps researchers optimize the chemical structure of drug molecules.
The molecule generator set up a library of 100 million novel drug-like small molecules, with a structural novelty of 99.68%. These molecules can be used to quickly generate new compounds with similar attributes, creating endless possibilities for new drug discovery. In addition, the Pangu molecular optimizer helps researchers optimize the chemical structure of drug molecules.
-
4. SOTA shows top performance in 20+ drug discovery tasks
The Pangu drug molecule model achieved industry-leading accuracy in over 20 tasks, including compound-target interaction prediction, compound ADMET attribute scoring, and molecule generation and optimization, covering the entire drug discovery chain.
The Pangu drug molecule model achieved industry-leading accuracy in over 20 tasks, including compound-target interaction prediction, compound ADMET attribute scoring, and molecule generation and optimization, covering the entire drug discovery chain.
Solution
Using EIHealth, Huawei's one-stop drug R&D platform, the Pangu drug molecule model has learned the chemical structure of 1.7 billion drug-like molecules. It creates in-depth representation of small-molecule compounds, matches drug molecules and target proteins, and predicts the biochemical attributes of new molecules, which allows it to quickly generate new drug molecules. Plus, it can optimize identified lead compounds in a targeted manner.
No. 1 Affiliated Hospital of the Medical School at Xi'an Jiaotong University designed new antimicrobial drugs using the Pangu drug molecule model. The cycle of lead compound development was reduced from several years to just one month.
"Huawei Cloud is a strategic partner of the Shanghai Institute of Materia Medica, Chinese Academy of Sciences. Our joint Pangu drug molecule model is one of the largest AI models specialized for drug discovery and research, with hundreds of millions of parameters. We believe this model will continue to make significant contributions to advance healthcare."
"Huawei Cloud is a strategic partner of the Shanghai Institute of Materia Medica, Chinese Academy of Sciences. Our joint Pangu drug molecule model is one of the largest AI models specialized for drug discovery and research, with hundreds of millions of parameters. We believe this model will continue to make significant contributions to advance healthcare."
"The Pangu drug molecule model has been a huge help for us in finding new broad-spectrum antimicrobial drugs while also reducing potentially toxic side effects. We're grateful for the support of the Huawei Cloud EIHealth team, and we wish them all the luck in continuing to improve the Pangu drug molecule model to better benefit the pharmaceutical industry."
"The Pangu drug molecule model has been a huge help for us in finding new broad-spectrum antimicrobial drugs while also reducing potentially toxic side effects. We're grateful for the support of the Huawei Cloud EIHealth team, and we wish them all the luck in continuing to improve the Pangu drug molecule model to better benefit the pharmaceutical industry."