The costs and process of developing anti-cancer drugs has been an extreme challenge for decades. Today one company, AccutarBio, is harnessing the power of AI to accelerate drug discovery and reform the current “hit-to-lead” drug discovery scheme. The company recently received $15 million in funding (including money from Chinese AI/facial recognition company YITU) and is now partnering with Amgen.
AccutarBio is proud of the dramatic improvements the company’s hybrid approach (combining computation design and experimental validation) has made in radically speeding up the drug discovery process. The company has achieved thus far:
- A data-driven atom-based scoring function is learned from 100,000 protein crystal structures containing information of >100 million amino acid side chains.
- A dynamic deep neural network specifically designed for chemical informatics.
The video below demonstrates how the company’s Orbital docking and virtual screen works.
AccutarBio uses AI technology to explain the physical and chemical nature of biological systems to accelerate drug discoveries. This drastically reduces the traditional drug screening process of extensive experimentation at the cost of billions of dollars so that work that took two years previously can be done in a matter of hours.
Overall network architecture
Artificial Intelligence for Drug Discovery
Accutarbio’s AI platform outperforms current standard approach in multiple tasks in drug discovery including: drug pocket prediction, drug-target complex conformation prediction, drug-target binding affinity prediction and drug property (ADME) prediction etc. The company’s virtual screen could potentially replace the need for costly experiment based screens; the docking pose prediction potentially saves the efforts for crystallization; and Chemi-Net could guide compound optimization/sampling more efficiently. Most importantly, the company proposes an integrated use the platform in pre-clinical research that will greatly improve drug discovery efficiency amounting to potentially saving 80% of the cost and accelerate the discovery cycle significantly.
At present, the screening of target drugs is often completed through a large number of experiments, and this process is often measured in years. For pharmaceutical companies with patents that are generally only valid for 17 years, it is common that R&D takes about 12-13 years. The annual economic benefits brought about by early completion of experiments are likely to be in the hundreds of millions or even billions of dollars. AccutarBio found that much pre-clinical research can be replaced in the form of AI algorithms, and the time spent can also be reduced from years to months, days, or even hours. AccutarBio has trained an AI algorithm based on more than 100,000 crystallographic data to accurately predict the target-binding potential of chemical compounds, screen lead compounds, and even in silicon design the compound for some biological function.
Dr. Fan Jie, founder of AccutarBio, predicts that through this platform, R&D that otherwise takes two years or so can be completed in a matter of hours. AccutarBio has used this platform for drug design. AccutarBio has obtained proof of concept results for multiple targets. Some of the leading compounds have been validated by different experiments including crystallography and animal studies.
Current Research
AccutarBio researchers have a number of papers on the arXiv pre-print server. Specifically:
- Chemi-Net: A molecular graph convolutional network for accurate drug property prediction – The research results showed that the deep neural network method used improved current methods by a large margin.
- Prediction of amino acid side chain conformation using a deep neural network – A deep neural network based architecture was constructed to predict amino acid side chain conformation with unprecedented accuracy.
Sign up for the free insideAI News newsletter.
