The Promise and Limitations of AlphaFold for Single Mutation Predictibility

Category Artificial Intelligence

tldr #

AlphaFold, a revolutionary AI developed by DeepMind, has revolutionized the field of structural bioinformatics. However, a recent study has highlighted AlphaFold's limitations in being used to accurately predict the effect of single mutations on protein stability. Despite these limitations, AlphaFold is still a powerful tool for researching protein structure and functioning.

content #

Proteins are vital components of the human body, performing both structural and functional roles. With the advent of AlphaFold in 2020, machine learning enthusiasts proclaimed the advent of a new age in structural biology, where the fold of a protein can be accurately predicted from its amino acid sequence.

In April 2023, Russian researchers at Skoltech attempted to take this one step further, applying AlphaFold to another central task of structural bioinformatics: predicting the impact of single mutations on protein stability.

AlphaFold was developed by the DeepMind team, a branch of Google.

The study’s principal investigator, Assistant Professor Dmitry Ivankov of Skoltech Bio said their research was motivated by a wish to determine the feasibility of using AlphaFold to predict the impact of single mutations on protein stability and function. To test this, Ivankov’s team compared AlphaFold’s predictions of the effects of single mutations on protein stability with known experimental findings. After testing several different datasets, the team found that AlphaFold’s single mutation stability predictions contradicted known experimental findings, leading them to conclude that AlphaFold was unable to accurately predict the effects of single mutations on protein stability.

AlphaFold is the first AI to successfully predict protein structures from amino acid sequences.

Despite the results of the study, Ivankov emphasized that AlphaFold’s creators never actually claimed that the AI was applicable to other tasks besides predicting protein structures based on their amino acid sequences. But some machine learning enthusiasts were quick to prophesy the end of structural bioinformatics as we know it, claiming that AlphaFold’s success was a sure sign they were simply dealing with the first generation of a revolutionary new tool in predicting protein structure.

AlphaFold is not able to accurately predict the impact of co-location mutations or cascade effects.

This enthusiasm was understandably quashed when this research was conducted a few months ago, showing that AlphaFold’s applications were more limited than some had hoped, and that structural bioinformatics still has a future. Despite this, Ivankov believes that AlphaFold can still be a powerful tool in structural biology, allowing researchers to quickly and accurately predict protein structures from amino acid sequences and allowing for detailed exploration of potential protein structures and their associated functionalities.

AlphaFold has not yet been tested on recently discovered proteins.

Overall, this study serves as a reminder that, while AlphaFold is an incredible breakthrough in structural bioinformatics and will undoubtedly revolutionize the field, it does have its limitations and should not be seen as a one-stop-shop for predicting protein structure, stability and even function.

hashtags #
worddensity #