Stanford University graduate students, led by an associate professor, have developed an AI that calculates the exact three-dimensional shape of molecules.
Determining the three-dimensional shape of biological molecules is the most difficult problem in modern biology. Companies and research institutes spend millions of dollars to understand how a particular molecular structure looks and interacts, but this does not always lead to results.
The authors of the new work used machine learning methods and developed an approach that solves this problem, as it predicts their exact structure. The researchers note that their approach can be applied even to molecules and structures that are most difficult to determine experimentally.
The graduate students’ algorithm predicts the exact molecular structures and from this, it is possible to understand how they work in different fields: this will help both in basic biological research and in drug development.
Proteins are machines on a molecular scale that perform all sorts of functions. In order to do some kind of action, they can bind to other proteins. If you know that a pair of proteins is involved in the disease, and also understand how they both interact, then you can try to make a drug that will hit the same point.
Stefan Eismann, PhD student at Stanford University
AI has learned to find the fundamental concepts of proteins that are key to forming molecular structure. And, what is important in the opinion of the authors, it was not loaded in advance with data on specific proteins, this could make the algorithm biased towards them and confuse in global analysis.
Therefore, the algorithm finds data and characteristics of proteins that scientists did not know about before.
During the experiment, the algorithm successfully coped with proteins and graduate students tested it on RNA molecules. As a result, the AI managed the puzzle in every case, although it was not designed specifically for RNA structures.