TmAlphaFold database: membrane localization and evaluation of AlphaFold2 predicted alpha-helical transmembrane protein structures

Abstract

AI-driven protein structure prediction, most notably AlphaFold2 (AF2) opens new frontiers for almost all fields of structural biology. As traditional structure prediction methods for transmembrane proteins were both complicated and error prone, AF2 is a great help to the community. Complementing the relatively meager number of experimental structures, AF2 provides 3D predictions for thousands of new alpha-helical membrane proteins. However, the lack of reliable structural templates and the fact that AF2 was not trained to handle phase boundaries also necessitates a delicate assessment of structural correctness. In our new database, Transmembrane AlphaFold database (TmAlphaFold database), we apply TMDET, a simple geometry-based method to visualize the likeliest position of the membrane plane. In addition, we calculate several parameters to evaluate the location of the protein into the membrane. This also allows TmAlphaFold database to show whether the predicted 3D structure is realistic or not. The TmAlphaFold database is available at https://tmalphafold.ttk.hu/.

TMvisDB: resource for transmembrane protein annotation and 3D visualization

“…The database and web application TMvisDB offers a straightforward search functionality and visualization interface. While several accurate structure prediction methods have been made available over the last year [14][15][16][17], we chose to enhance TMvisDB sequence annotations with AlphaFold2 [12] predictions that have been shown to perform well in structural analysis of transmembrane proteins (TMPs) [52], and have, therefore, been successfully applied as input by resources such as the TmAlphaFold database that collects alpha-helical TMPs [34].…”

“…While TMbed also predicts signal peptides (SP), performing on par with expert methods such as SignalP6.0 in binary classification, it does not distinguish different SP types [10,35]. TMvisDB is the first resource to gather such a large number of perresidue transmembrane topology annotations and, compared to existing TMP databases [8,34], expands the exploration space of easily browsable TMPs by two to four orders of magnitude.…”

“…While TMbed also predicts signal peptides (SP), performing on par with expert methods such as SignalP6.0 in binary classification, it does not distinguish different SP types [10, 35]. TMvisDB is the first resource to gather such a large number of per-residue transmembrane topology annotations and, compared to existing TMP databases [8, 34], expands the exploration space of easily browsable TMPs by two to four orders of magnitude. Providing interactive visualizations of TMP annotations with their respective AlphaFold2 structures, the intuitive interface of tmvis.predictprotein.org enables novices and experts in fields such as microbiology, drug development and bioinformatics to investigate small- and large-scale hypotheses on TMPs.…”

MembraneFold: Visualising transmembrane protein structure and topology

“…While these approaches are capable of accurately determining the membrane-buried residues, they require the user to specify the topology of the N-terminus, as they cannot differentiate between outside and inside of the membrane. Furthermore, they can suffer from AlphaFold's spurious folding of disordered regions and signal peptides that can violate membrane constraints [14].…”

“…However, given the nature of expert curation, they offer no easy way for the user to directly investigate a transmembrane protein from the sequence only. Another resource is TmAlphaFold [14] that has more than 200 thousand AlphaFold structures predicted to be alpha helical transmembrane proteins by CCTOP [7] with membrane placement predicted by TMDET [13].…”