Skoltech Bio scientists tested AlphaFold, the artificial intelligence program that solved the core problem of structural bioinformatics by predicting protein structures, in another field challenge. The team asked AlphaFold to predict the impact of single mutations on protein stability and the results contradicted the experimental results, suggesting that artificial intelligence is not a panacea for structural bioinformatics. The authors also refuted claims by some AlphaFold enthusiasts that the program had mastered ultimate protein physics and should perform beyond the task for which it was designed. The results have been reported in a PLOS One study.
Skoltech Bio scientists tested AlphaFold to predict the impact of single mutations on protein stability, and the AI program’s predictions contradicted the experimental results, refuting claims that it had mastered ultimate protein physics.
A high school bioinformatics boot camp at Skoltech has turned into a venue for the latest chapter in the ongoing competition between humans and AI in science. Having previously solved a key 50-year-old structural bioinformatics problem, the innovative AlphaFold AI program proved inapplicable to another challenge facing researchers in this field. This discovery is reported in a PLOS One study, whose authors refute the claims of some AlphaFold enthusiasts that DeepMinds AI has mastered ultimate protein physics and is the be-all and end-all of structural bioinformatics.
Structural bioinformatics is a branch of science that explores the structures of proteins,
” data-gt-translate-attributes=”[{” attribute=””>RNA, 
Poster of the project Playing With AlphaFold2 at the School of Molecular and Theoretical Biology held by Skoltech online in 2021. Credit: Dmitry Ivankov/Skoltech
After 50 years, the problem was resolved by AlfaFold, an artificial intelligence program created by Googles DeepMind, whose predecessors earlier made headlines by achieving superhuman performance in chess, the game of go, and the video game StarCraft II.
This milestone achievement led to speculations that the neural network must have somehow internalized the underlying physics of proteins and should work beyond the task it was designed for. Some people, even in the structural bioinformatics community, expected that the AI would soon give the definitive answers to that disciplines remaining questions and consign it to the history of science.
We decided to settle this and put AlphaFold to work on another central task of structural bioinformatics: predicting the impact of single mutations on protein stability. That means you choose a certain known protein and introduce exactly one mutation, the smallest change possible. And you want to know whether the resulting mutant is more stable or less stable and to what extent. AlphaFold was clearly unable to do this, as evidenced by its predictions contradicting the known experimental findings, the studys principal investigator, Assistant Professor Dmitry Ivankov of Skoltech Bio, commented.
Asked about the role of the high school students taking part in the project, the researcher said they were involved in mutation data processing, writing scripts for handling prediction results, visualizing the structures specified by AlphaFold, and basically fooling around with the online version of the AI.
Ivankov emphasized that AlphaFolds creators never actually claimed that the AI was applicable to other tasks besides predicting protein structures based on their amino
This point is supported by two previously voiced reservations regarding the AIs knowledge of physics. First, AlphaFold predicts some structures with side groups dangling in a way that suggests a zinc ion to be bound to them. However, the programs input is limited to the proteins amino acid sequence, so the only reason why the invisible zinc is there is that the AI was trained on analogous protein structures bound to this ion. Without the zinc, the predicted side group orientation contradicts physics. Second, AlphaFold can predict a solitary protein structure that looks sort of like a spiral and is indeed accurate provided that it is interlaced with two other such chains. Without them, the prediction is physically unsound. So rather than rely on physics, the program must be simply reproducing a shape it isolated from a compound structure.
Interestingly, this research grew out of a playful project featuring the participants of the School of Molecular and Theoretical Biology. We called it Games With AlphaFold. The moment AlphaFold became openly accessible, our lab installed it on the Zhores supercomputer. One of the games involved comparing the known mutation effects with what AlphaFold predicts for the original and the mutant proteins. This led to a study, in which high schoolers got the chance to simultaneously experience a supercomputer and advanced artificial intelligence, the studys lead author, Skoltech PhD student Marina Pak, commented.
Reference: Using AlphaFold to predict the impact of single mutations on protein stability and function by Marina A. Pak, Karina A. Markhieva, Mariia S. Novikova, Dmitry S. Petrov, Ilya S. Vorobyev, Ekaterina S. Maksimova, Fyodor A. Kondrashov and Dmitry N. Ivankov, 16 March 2023, PLOS One.
DOI: 10.1371/journal.pone.0282689
The study reported in this story was co-authored by Skoltech scientists, their colleagues from the Institute of Science and Technology Austria and Okinawa Institute of Science and Technology, Japan, and high school students who currently study at Ural Federal University and the Peoples Friendship University of Russia, and Armand Hammer United World College of the American West.
