AlphaFold-Aided Protein Explosion Imaging : Structure determination through protein explosion imaging of AlphaFold-generated putative conformations

Sammanfattning

The introduction of the deep learning-based model AlphaFold has made protein structure prediction from amino acid sequences widely accessible. Nonetheless, experimental methods are still needed to confirm the predicted structure of unknown proteins, and the same amino acid sequence may lead to a variety of predicted conformations. While AlphaFold is useful for the study of protein structures, it does not by itself constitute an imaging technique. Therefore, we suggest combining AlphaFold with Protein Explosion Imaging (PXI) to obtain a method capable of definitively determining the structure of a protein. In a practical setting, AlphaFold-aided PXI would amount to exploding proteins of unknown structure in experiments and comparing these to simulated explosions of AlphaFold predictions — ultimately identifying which AlphaFold prediction corresponds to the actual structure. As proof-of-concept for the method, we have sourced a variety of AlphaFold predictions and experimental structures for three proteins. By simulating X-ray laser induced explosions of all protein conformations using the software package MOLDSTRUCT and projecting ion trajectories onto a spherical virtual detector, explosion images are obtained. Through dimensionality reduction of the images, we show that structural similarity translates to similarities in explosion images and the theoretical possibility of identifying which AlphaFold prediction best resembles an examined structure. While the method is promising, difficulties arise in identifying the structural features that govern similarity in the exploded state. Further analysis of the relationship between structure and explosion image, combined with future experiments and improved simulation configurations, could establish this method as a powerful tool for protein structure determination.