January 2, 2019
ATOM was officially established in October 2017 by founding members, GSK, Lawrence Livermore National Laboratory, Frederick National Laboratory for Cancer Research, and the University of California, San Francisco. The ATOM Technical Team, comprised of experts in machine learning, data science, pharmaceutical sciences, cancer biology, biophysics, and engineering, has been working diligently to build our computational and experimental infrastructure, develop algorithms, and reach R&D milestones.
At the ATOM 1-Year Anniversary event this past October, over 60 delegates from ATOM member organizations convened at ATOM headquarters. Team leaders shared the major accomplishments achieved which ranged from benchmarking ATOM DeepChem models for demonstrated performance gains, to performing proof-of-principle multi-parameter simulations of the active learning process.
Highlights from the Team’s Year-1 Achievements
Data and modeling:
Combined private and curated datasets, and identified diversity and data gaps.
Made descriptor sets of chemical features for more than 2 million compounds
Automated a framework for model creation and tracking
Novel hybrid model development:
Showed that hybrid models perform better than typical molecule descriptors
Pharmacokinetic (PK) and safety data-driven modeling:
Built baseline models for PK parameters and liability assays with a focus on heart and liver toxicity
Benchmarked ATOM DeepChem models and demonstrated performance gains
Generated safety data sets to fill data gaps
Active learning integrated loop:
Performed proof-of-principle multi-parameter simulations of active learning process
Additional technical updates will be featured on our News page periodically, so check back soon!