DeepMind, the UK-based but Google-owned leader in AI research, has published research in the Nature journal that offers potentially revolutionary new insight into quantum chemistry. The AI scientists working at DeepMind have built a “neural network” system able to accurately predict the chemical energy stored in individual molecules.
The significance of that is that a molecule’s chemical energy, which it is now understood is determined by its structure, determines how the group of atoms that make up a particular molecule will behave in specific environments and if a particular reaction can occur or not. The breakthrough achieved by the AI system promises a potential revolution in the development of new manmade materials and drugs discovery.
Scientists will be able to use the DeepMind model to predict how matter will behave at nanoscale, the level of neutrons and electrons, more quickly and accurately simulate chemistry experiments on computers. While high school chemistry textbooks often simplify matters and represent electrons as moving around an atom’s nucleus in tidy orbit the reality is the way they behave is more akin to swarming in clusters.
Scientists researching quantum chemistry attempt to predict the probability of an electron’s position at a particular time. That allows for the creation of something like a 3D heat map predicting where electrons are likely to be and when and showing their probable density at a particular moment in time. That electron density model can be used to determine a molecule’s energy level.
The DeepMind research team behind the findings published in Nature commented:
“Solving some of the major challenges of the 21st century, such as producing clean electricity or developing high temperature superconductors, will require us to design new materials with specific properties. To do this on a computer requires the simulation of electrons, the subatomic particles that govern how atoms bond to form molecules and are also responsible for the flow of electricity in solids.”
The relationship between electron density and a molecule’s chemical energy has been known since the 1960s and used in chemistry, biology and materials science. The theoretical chemist Walter Kohn’s work on the topic won him a 1998 Nobel prize. However, for half a century the bottleneck has been that nobody had found a way to work out a reliable formula to accurately predict what is known as the “density functional”.
DeepMind’s approach was to first show its neural network system the electron density of molecules whose energy levels are already known. Trained on these known density functionals, the system was subsequently able to make far better prediction of how other molecules, whose energy levels weren’t yet known, would behave, allowing it to predict their energies and better describe a range of chemical reaction.
It is hoped the new knowledge will allow scientists to design new synthetic molecules which could be used in new materials or drugs. DeepMind research scientist James Kirkpatrick expands:
“The quantum world is a little like a dark room. At the moment, people proceed by trial and error. These computational tools, they are a torch that you can shine and predict what is around you.”