Central to much of our work is a special‑purpose supercomputer called Anton, which we designed and constructed to vastly accelerate the process of molecular dynamics (MD) simulation.
MD simulation is a computational technique in which the physical forces acting on each atom in a biomolecular system are repeatedly calculated and all resulting atomic movements are tracked in detail over a period of time. This provides the researcher with what amounts to a “computational microscope,” revealing structural changes and intermolecular interactions that cannot be observed using laboratory experiments.
Although the potential power of MD simulations for biomedical research and drug discovery had been recognized for decades, the enormous computational demands of such simulations had historically placed severe constraints on their length. Prior to the development of Anton, even the world’s fastest supercomputers were unable to simulate periods of biological time over which many of the most scientifically and pharmaceutically important phenomena take place.
Three generations of Anton machines have thus far been developed, each of which has executed biological MD simulations roughly 100 times faster than the fastest general‑purpose supercomputers of its day. In recognition of their extraordinary performance and their significance for scientific and pharmaceutical research, Anton systems have twice been awarded the Gordon Bell Prize, widely regarded as the highest honor in the field of high‑performance computing.
We have assembled a number of Anton machines, one of which we make available without cost for non‑commercial research use by U.S. universities and other not‑for‑profit institutions. All other Anton systems are used for in‑house research and development activities at D. E. Shaw Research, providing us with a unique and powerful engine for scientific investigation and drug discovery.
Among the other technological pillars of our work are methods that lie within the field of artificial intelligence, and in particular, machine learning. At D. E. Shaw Research, we use machine learning techniques both as independent computational tools and as key elements of certain Anton‑enabled research and drug discovery workflows.
By way of example, Anton is sometimes used to generate data that is in turn used to train a neural network, while in other cases, machine learning methods are used to dramatically expand the number of potential drug compounds that can be evaluated using a given amount of Anton time. We also use machine learning techniques as part of an ongoing process of progressively enhancing the accuracy of the physics‑based models we use in Anton simulations.
While Anton‑based MD simulations and machine learning methods play particularly important roles in our work, we make use of a number of other technologies and algorithms that we have developed over the years for various computational problems in the fields of biology, chemistry, physics, and pharmacology. Our scientific research and drug discovery activities are also enabled by a specialized set of internally developed software tools, and by an advanced computational and data management infrastructure tailored to our typical workflows.