AlphaKnot: server to analyze entanglement in structures predicted by AlphaFold methods

Abstract

AlphaKnot is a server that measures entanglement in AlphaFold-solved protein models while considering pLDDT confidence values. AlphaKnot has two main functions: (i) providing researchers with a webserver for analyzing knotting in their own AlphaFold predictions and (ii) providing a database of knotting in AlphaFold predictions from the 21 proteomes for which models have been published prior to 2022. The knotting is defined in a probabilistic fashion. The knotting complexity of proteins is presented in the form of a matrix diagram which shows users the knot type for the entire polypeptide chain and for each of its subchains. The dominant knot types as well as the computed locations of the knot cores (i.e. minimal portions of protein backbones that form a given knot type) are shown for each protein structure. Based mainly on the pLDDT confidence values, entanglements are classified as Knots, Unsure, and Artifacts. The database portion of the server can be used, for example, to examine protein geometry and entanglement-function correlations, as a reference set for protein modeling, and for facilitating evolutional studies. The AlphaKnot server can be found at https://alphaknot.cent.uw.edu.pl/.

Graphical Abstract

Graphical representation of AlphaKnot pipeline for topology recognition of protein structures predicted by AlphaFold.

Figure 1.

Submission options. The user can choose from several options which determine what methods will be used to describe the entanglement in the structure, and also determine the accuracy of the calculations. The calculation time on the server is directly related to the selected parameters (and to the length and complexity of the structure). For a single structure, computation time varies from a few seconds to several hours.

Figure 2.

Example page with the output from a submission to AlphaKnot. Here the multiple chains option was chosen and four proteins with at least 40% sequence similarity were analyzed: I1L257, Q2JIZ5, P93527, B4YB07 (UniProtKB ID). (A) Job status table. The table shows details of the job, information about the methods used, the chosen parameters for the job, and the main results. (B) Structure choice. For multiple chain uploads, the user can choose the structure to view. Here I1L257 is selected, which forms the knot 4₁. (C) Information and Artifact check. The tables present basic information on the topology of the structure and results of some automatic tests on the quality of the structure and its knotted region. These results can help the user to recognize if the knotted topology may be an artifact—any information in red suggests that the user should be wary of the results. (D) The knot map (if full matrix was computed). The user can choose the cutoff—knot frequency required to show it on the matrix. (E) View of the structure. The user can choose the coloring method and select the fragment of the chain which is colored. In the figure the knot core region is colored by the pLDDT values, which is helpful in assessing whether the knot is an artifact. (F) Knot types and model sequence. The table contains detailed information about all knots formed by fragments of the chain (if full matrix was computed). By selecting a knot, the corresponding sequence is highlighted. (G) Comparison of the location and types of knots in each (knotted) structure. The color of the bar corresponds to the type of knot. The user can see on the chart that three of the four structures (besides Q2JIZ5) form the knot 4₁.

Figure 3.

Features that can be searched for knotted proteins in Advanced Search in the AlphaKnot database.

Figure 4.

Protein (UniProtKB ID: Q54P92) with very high pLDDT value, but whose topology (knot with 10 crossings K10₁₄₆, shown with schematic) suggests that the prediction is not reliable.