The most accurate simulations of objects made from tens of millions of atoms are run on world-class supercomputers with the help of artificial intelligence.
Existing simulations that describe in detail how atoms behave, interact and evolve are limited to small molecules due to the computational power required. Techniques exist to simulate a much larger number of atoms over time, but these rely on approximations and are not accurate enough to extract many detailed features of the molecule in question.
Now Boris Kosinski and his colleagues at Harvard University have developed a tool called Allegro that can use artificial intelligence to accurately simulate systems containing tens of millions of atoms.
Kosinski and his team used Perl Matter, the world’s eighth most powerful supercomputer, to simulate the 44 million atoms in the protein shell of HIV. They also simulated other common biomolecules such as cellulose, a protein lacking in hemophiliacs, and the prevalent tobacco plant virus.
“Anything that is essentially made of atoms can be simulated with very high accuracy using these methods, and now at large scale,” says Kosinski. “This is one demonstration, he said, but by no means is it limited to this area.” The system could also be used for many problems in materials science, such as research in batteries, catalysts and semiconductors. says he.
To be able to simulate such a large number of particles, the researchers used a type of AI called a neural network to simulate interactions between atoms that are symmetric from all angles, a principle called equal variance. Calculated.
“When we develop networks that very fundamentally contain these symmetries … accuracy and other things we care about, such as the stability of our simulations and the speed with which machine learning models learn when we are taught more data, The characteristics are greatly improved,” said team member Albert Musaelian, also of Harvard University.
“This is a masterpiece of programming and demonstrates that these machine learning possibilities are scalable,” said Gábor Csányi of the University of Cambridge.
However, biochemists already have tools that are sufficiently accurate and can run much faster, so simulations of biomolecules like this are more of a practical boost for researchers than they are a large-scale implementation of the tools. It’s just a demonstration that the system works, he says. This could be useful in atomic-rich materials that undergo shocks and extreme forces on very short timescales, Tsani says, such as planetary cores.
topic:
