AI DeepMind has solved the 50-year-old problem of protein folding Some AI applications are as impressive as they are essentially useless, but others can bring about a breakthrough in treating serious diseases, and this is exactly what we are dealing with.

Alphabet's DeepMind entity, and more specifically its artificial intelligence, has just presented its unique skills. It is a system that is capable of predicting the 3D structures of unique proteins, thereby overcoming a problem that has been blocking biologists for over 50 years. By understanding the 3D shapes of different proteins, scientists are better able to understand how they work and how they cause disease, which in turn leads to the development of more effective drugs. In addition, as protein is a central component of the chemical processes of all living organisms, better mapping of 3D structures can benefit many fields of biological research.

All because while modern scientific tools such as structural X-ray or cryoelectron microscopy allow researchers to study these structures with unprecedented precision, they still rely on trial and error. Proteins are made of one-dimensional chains of amino acids, which then fold into the final 3D structure, which, however, is a surprise for scientists. Or at least they were, because thanks to the artificial intelligence of DeepMind they will be able to predict what the end result will look like - it's the so-called the problem of protein folding, which has been bothering researchers since the beginning of the seventies of the last century.

Mainly because there are so many possible 3D configurations that scientists believe it would take more time than the universe exists to establish them all using today's methods. And this is where AlphaFold comes into play, which was designed for this very purpose - the first version was presented in 2018 and although it was already very effective at that time, now it has planetwatching many significant improvements. The system was trained on approximately 170,000 publicly available protein structures and a huge database of unknown protein structures, which allowed it to achieve a score of 92.4% in terms of effectiveness in the widely used Global Distance test.

'The amazingly accurate AlphaFold models have allowed us to solve the protein structure we've been stuck with for nearly a decade, restarting our efforts to understand how signals are transmitted between cell membranes,' explains Professor Andrei Lupas of the Max Planck Institute. Overall, AI will help scientists identify malfunctioning proteins and the causes that lead to certain diseases, opening up new avenues for drug development and prompt treatment. They can also help in the development of enzymes for the degradation of plastic garbage or future pandemics, they predict the structure of new viruses. In addition, they will reveal the secrets of the current protein structures that have been a mystery to us so far.