In December 2017 Chinese pharma tech startup AccutarBio raised US$15 million from IDG Capital, YITU Tech, and ZhenFund. This was one of the country’s largest pharma tech funding rounds, and signaled AI’s strong promise and potential in early-stage drug discovery.
“Drug discovery is never subjective,” AccutarBio CEO and Co-founder Dr. Jie Fan tells Synced. “A person may have a good knowledge of thousands of types of drugs. But there is no way they know ten thousand or even one hundred thousand drugs. In contrast, machines can do a much better job than humans.” Dr. Fan also believes that hybrid AI can accelerate drug discovery for targeted therapy and provide cancer patients with further alternative treatment plans.
Dr. Fan graduated from Fudan University with a BA and received his Master’s in Biostatistics from the University of California Berkeley in 2004. He completed his PhD under Dr. Nikola Pavletich, studying the crystal structure of DNA binding proteins. Fan then worked as a postdoc researcher under 1999 Nobel Prize laureate Günter Blobel, focusing on the structural analysis of nuclear pore complexes (NPCs), which are large protein complexes that allow transport of molecules across the nuclear membrane.
AccutarBio continues to widely collaborate with academics for cross-interdisciplinary research, and the company has deployed labs in both Shanghai and New York.
In the research paper Chemi-net: a Graph Convolutional Network for Accurate Drug Property Prediction, AccutarBio extended the use of traditional statistical learning methods to create a multi-layer DNN architecture called “Chemi-Net”, which can predict ADME (absorption, distribution, metabolism, and excretion) properties of molecular compounds. ADME study is crucial and should be done at an early stage in drug discovery as it helps researchers understand the transport of molecules in organisms and can efficiently eliminate weak drug candidates, increasing success rate in drug trials and shortening drug discovery timelines.
Chemi-Net was tested on 5 ADME endpoints — human microsomal clearance, human CYP450 inhibition, aqueous equilibrium solubility, pregnane X receptor induction, and bioavailability — with 13 industrial grade datasets selected for predictive model development, involving more than 250,000 data points. Both single-task and multi-task Chemi-Net exhibited dramatic predictive accuracy improvements over benchmarks. Researchers expect Chemi-Net’s significantly increased ADME prediction accuracy to greatly accelerate efficiency and success rates in drug discovery.
Following on the success of their AI-powered drug discovery model, AccutarBio plans to further develop and promote Chemi-Net with the aim of revolutionizing traditional experiment-based and experience-based drug development.
AI has demonstrated its abilities in healthcare applications in medical and pharmaceutical fields. The tech’s capability for ‘learning’ meaningful features from large datasets has been widely used in clinical situations involving computer-aided image detection and diagnosis (e.g., assisted image-based early cancer screening), implementation and maintenance of electronic health records, and continuous patient monitoring.
AccutarBio strategic investor YITU Tech says “Asymmetric information in multidisciplinary fields has brought huge barriers to technology development. We hope our knowledge and work in AI will be able to assist AccutarBio and make great advances in biology. We believe artificial intelligence will make significant difference on the current status of biological research and biopharmaceuticals through our in-depth cooperation with AccutarBio. And collaboration with YITU for Medical AI-powered research in clinical applications will bring more profound value.”
Localization: Tingting Cao| Editor: Meghan Han, Michael Sarazen