PrankWeb allows users to predict and visualize the protein ligand binding sites and contrast these with both highly conserved areas and actual ligand binding sites.
To carry out the prediction, users can either upload a PDB file or provide a PDB ID, in which case PrankWeb will download and store the corresponding PDB file from the PDB database (30 (link)). In addition to selecting what protein to analyze, users can also specify whether evolutionary conservation should be included in the prediction process, which in turn determines which of the two pre-trained models will be used.
Conservation scores are calculated using the Jensen-Divergence method (31 (link)) from a multiple sequence alignment (MSA) file, which can come from three sources: (i) users can specify their own alignment file, (ii) if a protein’s PDB code is provided, PrankWeb uses MSA from the HSSP (32 (link)) database or (iii) where no MSA is provided and no MSA is found in HSSP, the MSA is computed using PrankWeb’s own conservation pipeline, which utilizes UniProt (33 ), PSI-Blast (34 (link)), MUSCLE (35 (link)) and CD-HIT (36 (link)). This process is depicted in Figure 2 and described in detail in the Supplementary Material.
After specification of the input, the submitted data is sent via a REST API to the server, which then starts the prediction pipeline. The user is provided with a URL address from which progress of the prediction process can be tracked and results inspected once the process finishes.
On the results page, PrankWeb utilizes LiteMol for visualization of 3D structural information and Protael for sequence visualization. Figure 1 displays the predicted binding sites of dasatinib (a drug used for treatment of chronic myelogenous leukemia) bound to the kinase domain of human LCK (PDB ID 3AD5). The sequence and structure plugins are synchronized so that the user can easily locate a sequence position in the structure and vice versa. The sequence view comprises predicted pockets, computed conservation and binding sites (if present in the PDB file). The side panel displays information about the identified pockets and a toolbar allowing the user to (i) download all inputs and calculated results, (ii) share the results page link or (iii) switch between visualization modes. PrankWeb comes with three predefined 3D model renderings (protein surface, cartoon and atoms) and the predicted binding sites and conservation scores are color coded. Conservation is displayed in grayscale (darker denoting more conserved residues) and binding sites are color-highlighted. When the conservation score is not available, the protein surface is white. If conservation analysis is chosen, the user can contrast the positions of putative active sites with conservation scores of the respective positions. In cases where the preset modes do not suffice, one can completely customize the 3D visualization using LiteMol’s advanced user interface or the PyMOL visualization script for offline inspection.
PrankWeb consists of a Java backend, REST API and a Typescript frontend, the backend being based on the WildFly (37 ) web server and the P2Rank application, while the frontend uses the Protael, LiteMol and Bootstrap.js libraries to provide an interactive user interface on top of the REST API. All source code is available under Apache License 2.0 at GitHub (https://github.com/jendelel/PrankWebApp). The GitHub website also includes documentation for developers on how to use our REST API and how to deploy their own version of the server.