Getting Started

pharmit is a web server that facilitates virtual screening: it enables users to search for small molecules based on their structural and chemical similarity to another small molecule, with the goal of identifying those that bind to a target molecule (typically a protein receptor or enzyme). The search can take as input either a small molecule, a set of pharmacophore features (which will be subsequently described in greater detail), or both a protein and a pharmacophore or a small molecule for which a putative binding pose is known. This help follows the 4pps example, which the user may choose to follow simultaneously by clicking the link, or by other search initiation methods described below.

Initiating a Search

There are three ways to initiate a search with pharmit: from a PDB accession code, from receptor/ligand files (including both a receptor and ligand as well as a ligand by itself), or by using a pharmacophore generated from an external source and uploaded to pharmit. Each method is described in detail below.

From PDB

A structure can be obtained directly from the PDB by entering its four-character PDB identifier in the first text box after "start from PDB" as shown in the figure below. To initiate the 4pps search using this method, the user may enter "4pps" in the box. A list of possible small molecules will be generated automatically in the second box, from which the user may choose a ligand of interest (ESE, in our case). Binding site waters may be excluded entirely, used as part of the ligand as optional pharmacophore features to include in the similarity search, or used as part of the receptor to identify which pharmacophore features of the ligand are relevant to binding (for our example, we will exclude them). To proceed with the search, the user should next choose "submit".

From receptor/ligand files

Alternatively the user may first choose "enter pharmit search" in order to upload structural files. After being redirected to the search page, choose "Load Features" to upload a small molecule structure in sdf, pdb, mol2, or xyz format. A receptor file in any of the same formats may optionally be provided as well; our example uses both the ligand and receptor. To initiate the 4pps search using this method, the user should load the attached receptor and ligand files as the receptor and features, respectively. These files were obtained directly from the 4pps PDB entry and are provided as examples of possible input formats. If a receptor is provided via the "Load Receptor" option, it should be one for which the binding pose of the provided ligand is known, or to which the ligand was previously docked. pharmit will not dock the two compounds, and since the receptor is used to identify which pharmacophore features of the ligand are relevant to binding, providing a pair of structures that are oriented arbitrarily will fail to identify a relevant pharmacophore. If a ligand structure is provided without a receptor, the query will proceed normally but the user will not be able to energy minimize the results of queries, since this calculation depends on the presence of a receptor. The second image below shows an example of uploading a ligand structure without a receptor.

From external sources

The user may also use an external program such as MOE, LigBuilder, LigandScout, or Pharmer to generate a pharmacophore query file and then upload that file to pharmit directly. It is provided in the same manner as a ligand structure file; that is, the user should click "enter pharmit search" on the main page, than click "Load Features" to upload a query file in MOE, LigBuilder, LigandScout, or Pharmer format. A receptor file may optionally also be uploaded via the "Load Receptor" button.

Visualization

After the desired structures are provided, pharmit will identify all pharmacophore features present in the ligand if a ligand structure rather than a pharmacophore query file was provided. If a receptor structure was provided, it will identify which of these features are relevant to the interaction between the protein-ligand pair using distance cutoffs between interacting features and will display only these interacting features. Next, it will center the features and use a set of default visualization options to display the provided structures, including the electrostatic surface of the receptor. If the user is dissatisfied with the default visualization scheme, they may scroll down to the "Visualization" menu in the left sidebar and toggle the options as desired. In particular, if the electrostatic surface of the receptor is obscuring the pharmacophore features, receptor surface opacity may be reduced and at the lowest setting becomes entirely transparent. In the graphical display window, the left mouse button may be used to rotate the scene and the right mouse button or center wheel may be used to zoom. Clicking on spheres - both pharmacophore spheres and shape constraint spheres - toggles between a solid and wire display. In order to maximize the viewable graphical display area, the sidebar may be hidden by pressing the left-pointing arrowhead at the top right of the sidebar.

Pharmacophore Representations

Pharmacophores are used to define molecular similarity and perform structural alignment. A pharmacophore describes the spatial arrangement of the essential features of an interaction. Each pharmacophore query feature includes both a type and a radius. The pharmacophore type derives from the underlying chemistry of functional groups and in the context of pharmit may be defined as an aromatic, hydrogen donor, hydrogen acceptor, hydrophobic, negative ion, or positive ion. The set of pharmacophore features chosen to represent a particular ligand and their geometric orientation is used to identify other molecules in a database that also possess that configuration of features. The radius of a pharmacophore query feature determines how closely a molecule in the database must match the configuration of the query. To be considered similar, the pharmacophore feature points of a molecule in the database must overlap the corresponding pharmacophore feature spheres in the query. The likelihood of this occurring increases with larger feature radius.

A list of pharmacophore features is provided in the "Pharmacophore" menu in the sidebar. Pharmacophore features can be toggled "on" or "off" to include them in or remove them from the search. Pressing the "x" to the right of the pharmacophore removes it from the menu entirely. By pressing the arrow to the left of a given pharmacophore, the user can vary the type, radius, and location of a pharmacophore, as well as optionally restrict the number of atoms associated with a hydrophobic pharmacophore and the phi or psi angle associated with pharmacophore features, such as hydrogen bonds, that have an associated direction.

At the bottom of the "Pharmacophore" menu, there are buttons to sort the pharmacophore features and to add new features. Sorting places all active pharmacophores at the top of the list and organizes them by type. Pharmacophore features added using the "Add" button are initialized generically to hydrophobic pharmacophores of radius 1.0 located at (0,0,0), and must be set to appropriate types and locations by the user.

Shape Constraints

Users have the option of using shape as an additional constraint on the similarity search. The ligand's surface or a set of user-defined spheres may be used as an inclusive constraint, the receptor's surface or a set of user-defined spheres may be used as an exclusive constraint, or both inclusive and exclusive shapes may be used. Multiple spheres may be defined for both inclusive and exclusive search by choosing the "Spheres" option in the dropdown menu and pressing "Add". This will display a menu allowing the user to set the position and size of the sphere. Pressing "Add" again will enable the user to continue adding spheres. The shape constraint is applied differently depending on whether it is used to search the database prior to pharmacophore filtering or to filter results returned by a pharmacophore search. In the former case, a shape search returns molecules in the search database that contain the entire included shape and do not overlap any part of the excluded shape. In the latter case, the results of the pharmacophore search are filtered by shape to ensure that at least one heavy atom falls within the inclusive shape and no heavy atom centers fall within any exclusive shape. In either case, the results of the initial search are aligned to that search for the secondary filtering - thus for a given molecule a pose may exist that allows it to meet the secondary filter, but it will not be returned if the aligned pose resulting from the initial search does not meet those constraints. Our recommendation is that users define both inclusive and exclusive shape constraints using the ligand and receptor surfaces and use them as secondary filtering on the results of a pharmacophore search. The order of application of pharmacophore and shape constraints can be swapped using the second button from the top of the left sidebar, which switches between "Pharmacophore Search → Shape Filter" and "Shape Search → Pharmacophore Filter".

Hit Reduction and Feasibility Screening

Additional hit reduction and screening options are provided in the "Filters" menu. The three options for hit reduction include restricting the maximum hits returned for every configuration, for every molecule, and overall. The maximum hits per configuration constraint is applied greedily, returning the first n results found without prioritizing any other aspects of those results. The other two constraints are applied after the results are sorted, giving priority to those of higher rank (lower RMSD for pharmacophore searches and higher similarity scores for shape searches). There are seven options given for hit screening. These include constraints on molecular weight, the number of rotatable bonds, logP (a measure of lipophilicity), polar surface area (indicative of ability to permeate cell membranes), the number of aromatic groups, the number of hydrogen bond acceptors, and the number of hydrogen bond donors. Properties are computing using OpenBabel.


Database Selection

When the user is satisfied with the query and constraints, a database to search should be selected by pressing the arrow next to the first button on the left sidebar. The options include CHEMBL20, ChemDiv, MolPort, NCI Open Chemical Repository, and PubChem, as well as several publicly available user-contributed libraries and any public or private libraries the user has personally submitted using the "create" menu accessible from the pharmit main page. Once a database is selected, pressing the search button initiates the search.

Saving and Loading Sessions

The two buttons at the bottom of the left sidebar enable users to save and load pharmit sessions. Saving a session generates a file documenting the receptor, features, visualization state, shape constraints, and any additional features the user specified. This file can then be loaded to recreate the state of the interface at the time the file was created.

Managing Search Results

When a database search is initiated, a right sidebar opens to display the search results. This sidebar may be hidden by pressing the right-facing arrow at the top left of the sidebar, or it may be closed by pressing the "x" in the upper right corner. Results may be sorted in increasing or decreasing order based on RMSD, mass, or number of rotatable bonds; the sorting method is determined by pressing on the name of desired sorting method, and pressing the name again to toggle between increasing and decreasing order. Lavender arrowheads to the right of the sorting method names indicate whether sorting is being performed in increasing (up) or decreasing (down) order.

The query aligned pose may be visualized by clicking on the corresponding row in the results panel. Hovering over a row reveals alternative identifiers for the compounds in other databases hosted by pharmit such as PubChem. A selected compound has the option of initiating a new query using the compound as the query ligand.
Saving and Minimizing Results

There are two buttons available at the bottom of the right sidebar. The "Save" button allows the results to be saved in their current state; this generates a single file in sdf format that contains all the structures that were returned by the query. The structures are sorted in the output in the same order that they are sorted in the sidebar. Pressing the "Minimize" button before saving allows the user to assess the favorability of the interaction between the protein and the ligands returned by the search. This feature energy minimizes the results using smina, a fork of AutoDock Vina, using the default Vina scoring function. Using this scoring function, a more negative score indicates that the corresponding interaction is more favorable according to the scoring function's model of protein-ligand interaction; thus large positive scores indicate unfavorable interactions, and large negative scores indicate more favorable interactions. If the user chooses to minimize the results, they may be sorted by score or RMSD of the minimized pose from the starting pose (which was aligned to the query) by pressing the name of the sorting method. Additional filters may also be applied by setting values for maximum score, maximum minimized RMSD, and requesting that only one pose be retained for each conformer. Once the filters have been set as desired, the user should press "Apply" to enforce them. Pressing "Save" after minimizing the results generates a single sdf file containing the minimized structures that includes their minimized scores; it is often useful to rank the structures in order of increasing score so that the best results can be reviewed first. We additionally recommend that results be filtered by mRMSD (e.g. 2Å) to eliminate poses with significant deviations from the original query.

User-submitted Databases

Users may submit databases to be used for searches using the "create" area on the main page.
If a user chooses to proceed as a guest, no private databases may be submitted and public databases the user submits have a maximum size of 10,000 conformers. If the user creates an account and logs in, the user may create one private database with at most 1,000,000 conformers and public databases the user submits may have at most 10,000,000 conformers. The database files should be in SMILES or sdf format. Conformers will be automatically generated from SMILES strings, but the structures contained in the sdf files will be used directly and should therefore be created using a high quality conformer generator. SMILES files should contain a single molecule on each line and may include an optional name after the SMILES string. File sizes are capped at 200MB, so submitting a compressed file is recommended. The number of permitted conformers in a database may be extended by emailing a short justification to dkoes@pitt.edu.

Benchmarking Databases

Databases can be formatted to enable automatic benchmarking of pharmacophore queries. To enable this feature, active compounds must have the word "active" embedded in their name. When a search is performed, Pharmit will compute and display the enrichment factor (EF) and F1 score of the results, as shown below. When calculating these metrics, only the number of unique molecules are used. That is, the number of conformations of a molecule returned by the query isn't considered, only if at least one conformation is returned. Several benchmarking databases, derived from the DUD-E dataset, are available as contributed libraries (search for benchmark).

Related Software

pharmit is completely open-source with all the code required to setup and host the web application available through SourceForge. pharmit integrates several software packages actively developed by David Koes and his research group, including Pharmer, an efficient molecular search algorithm based on pharmacophores; smina, a fork of AutoDock Vina that performs pose scoring, small molecule docking, and structural minimization and permits custom scoring function development; ShapeDB, software that performs a similarity search based on molecular shape; and 3Dmol.js, a highly optimized molecular visualization package utilizing WebGL.

Additional Help

For additional help with pharmit please post to the discussion board or contact us directly.