Architecture of PubPharm

PubPharm is a discovery system based on the open-source software Qcovery/VuFind. These systems are modern library catalogues based on search engine technology. The data is integrated into a single search index. The aim of the Qcovery network is to provide a discovery system that can be easily adapted to different use cases so that the adaptation effort through programming is reduced. The backend, i.e. the central search index K10plus-Zentral, is located at the Head Office of the GBV Common Library Network in Göttingen. The PubPharm architecture can be divided into three layers: data sources, data management and frontend.

• Data sources: PubPharm relies on bibliographic data as the basis of the search and norm data (from the DrugBank database and the Medical Subject Headings) for search support, e.g., for generating search suggestions.
• Data management: The data from the various sources are transferred to an index (Apache Solr Cloud) via the infrastructure of the Head Office of the GBV Common Library Network. The data sources include, among others, the metadata of the biomedical database Medline (PubMed), pharmacy and chemistry collections of the Braunschweig University Library, study registers, preprint archives and journals licensed by the FID Pharmazie. Therefore, a query in PubPharm can identify matching journal articles, preprints, clinical studies and books or dissertations (from collections of most German university libraries and international dissertations), which reduces the need for individual searches in several databases.
• Frontend: The search platform is based on the open-source software Qcovery/VuFind and can be accessed via the new PubPharm portal (based on TYPO3). All other services (e.g., social media channels) are also integrated into this portal.

Full-text access

• The search index contains metadata on each record. It provides information such as title, author(s), year of publication, document type. Metadata do not (usually) contain full texts. For electronic sources (e-books, e-articles), a link to the full text is often listed in the metadata, or information is provided which can be used to generate such a link in the result list. Access to the full text depends on many factors, such as whether it is open access or licensed by the local institution (see next section).
• Sometimes, the metadata do not contain information on the full-text link. In these cases, you can navigate via the journal homepage, volume and issue to get to the desired article.

Availability check

• PubPharm displays whether free full-text access is available depending on the individual location. An availability check based on the IP address of the search query is used for this purpose.
• The availability check proceeds stepwise. In the initial step, it is determined whether the publication is open access or freely accessible. In the next step, the local license for the source is checked. A link resolver is used to determine the IP authorisation for IP addresses belonging to German, Austrian or Swiss universities. It is important to be in your university network to access full-text versions. If this is not the case, it is recommended to prove whether your university offers a VPN service. The next step is to check whether the hit has been licensed by the FID Pharmazie (FID-licences). If direct full-text access is not possible, a link to the commercial document delivery service "subito" (https://www.subito-doc.de/) is offered (if available). Alternatively, a request can be sent to the FID Pharmazie. The FID Pharmazie will then determine whether and how access to the full text is possible.

Refine the search by narrowing and sorting the results (see arrow 1)

• Change sort order
  By default search results are sorted by publication date (newest hits first). Sorting by relevance (most relevant article first), oldest hits first, author (in alphabetically ascending order) or journal title is also possible (see arrow 2). The number of displayed hits on a page can be expanded from 10 to 20 or 50 hits.
• Sorting by relevance
  Sorting by relevance is made possible by a specialised algorithm. High weighting is given to topics (keywords), the main author and the title. Low weighting has, e.g., full text, the publisher, ISBN and ISSN.

Filter functions
A large number of results can be refined by different filters, e.g., for media type, journal title, year of publication, subject or language (see arrow 4). Furthermore, checkbox filters are offered to restrict the search to clinical trials (yes/no) from the ClinicalTrials.gov registry and the International Clinical Trials Registry Platform (ICTRP) of the WHO, patents (yes/no), systematic reviews (yes/no) and current publications with a focus on COVID-19/SARS-CoV-2 (see arrow 3). Using the filter "Systematic Reviews", hits are obtained that have "systematic" and "review" in their title. Via additional checkbox filters, systematic reviews, clinical trials, and patents can be excluded from the result list. The different filters can be (logically) combined with each other and are linked with a logical AND.
Searches and filter settings can be recalled in the search history and saved in the personal account (see How can I use the list of favourites and the export function?).

You can see by means of the icon which source is listed (e.g., journal article, see arrow 1). In addition, the title, further information on the respective source (e.g., journal title, year of publication, etc.), the first three authors and the result of the availability check (full-text availability, see also What are the technical details behind PubPharm?) are displayed.

In addition, information on citation statistics and the social media impact of the source provided by the Dimensions database and Altmetric is linked (see arrow 2 and arrow 3). These data are available for each source in PubPharm that has a digital object identifier (DOI) and has been cited or mentioned in social media at least once. With each new search, the citation information is dynamically queried from the respective provider and is thus up-to-date on a daily basis. Clicking the Dimensions citation icon will take you to their webpage listing the citing articles. Altmetric provides information on the social media resonance of the searched article.

Favourites lists

• By clicking the favourites button (star, see arrow 1), individual results can be added to the favourites list (see arrow 2). From this list, it is possible to export all or selected titles, print them, send them as a link list by email or save them in a permanent list in the user account (free account) (see arrow 3).
• You can basically create favourites lists in a browser session without logging in. A personal account is required to save and later edit these lists.

Personal account

• You can create a free personal account to save watch lists (favourites lists) and search queries via the search history.
• The account can be created under "My Account" and "Create New Account" (see arrow 1 and arrow 2).
• Entries in the favourites list can be transferred to a new or existing list via "Save". Hits of a list can be sorted by "Title", "Author", "Newest first" and "Oldest first".
• Selected entries of a list can be sent by email or exported for, e.g., a literature management program.

What can I use the export function for?

• Results can be exported, e.g., for a literature management program (e.g., Citavi, EndNote, Mendeley or Zotero) (see arrow 1). A literature management program or reference manager contains a database into which (meta) data of publications can be imported and managed. The link to a word processing program makes it possible to insert references and literature lists in a citation style.
• BibTex, EndNote, EndNoteWeb, RefWorks, MARC, MARCXML and RIS are offered as export formats. In addition for each result, a corresponding ID, such as DOI and COinS, are offered in the interface for browser extensions of reference management tools.

How can I be automatically notified about newly indexed hits?

• An RSS feed for a search query can be called up and subscribed below the result list (see arrow 2). The feed shows newly added entries with the title and year of publication. To read the RSS feed, you need a so-called feed reader. These tools are available free of charge.

The advanced search can be used for more selective and complex queries (see arrow 1). It can be defined which index fields (metadata) for which search term should be searched (see arrow 4). For example, you can search for all publications of an author in a specific journal.

A video tutorial is available to introduce the advanced search (see arrow 6).

The following index fields can be selected in the advanced search:

• Standard search: Search in fields with main contents: Author(s), title, keywords, abstract, table of contents, publisher, ISBN/ISSN.
• Title: Search within the title (e.g., titles of articles and books).
• Journal Title: Search within the title of journals. Articles published in the particular journal will also be returned.
• Person: Search for authors (including co-authors and other contributors).
• Publisher: Search for the publisher (for books and journals).
• Topic: It searches for publications on a specific topic. Each article in PubPharm is automatically assigned to specific topics that result from phrases and keywords. These words are available as part of the metadata for the selected article. In most cases, vocabulary words are used (such as Medical Subject Headings, MeSH). All data records from the Medline database in the search index should be provided with corresponding entries from MeSH. Other records may also contain both controlled vocabulary entries and free keywords.
• Full text: The search term is searched in the full text. It should be noted, that full texts are not available in the database for all publications.
• Table of contents: Search within the table of contents.
• ISBN/ISSN: Search for the ISBN of books or ISSN of journals.
• Year(s) of publication: Search for the publication year.
• Index (all fields): In addition to the standard search, additional fields, such as URL, are parsed. This search function provides the most comprehensive search.

Search Fields

Within the search fields (see arrow 2 and arrow 3), you can combine search words with the operators and search functions supported by PubPharm (see How can I search more specifically?). All search fields of a search group can be combined with boolean operators.

• AND Returns only the hits that contain all words or phrases from this search group.
• OR Returns the hits that contain one word or all words or phrases from this search group.
• NOT Returns all hits that contain words or phrases not specified in this search group. This search condition works only when searching in at least two search fields.

Search Groups

Some search queries are more complicated, and searching across different fields is not enough. With search groups you can group search fields: You can use "AND" or "OR" to link the search groups (see arrow 5).

Example

You want to search for titles on bisoprolol and captopril for the treatment of hypertension:

• Add the terms "bisoprolol" and "hypertension" to the input fields of the first search group and leave the "search operator" for the group set to "AND".
• Add another search group and enter the terms "captopril" and "hypertension" there, and leave the "search operator" for the group on "AND".
• Set the search operator for the combination of search groups to "OR".

The structure search complements the text-based search as a unique function of PubPharm. It is based on data from the PubChem database. More information can be found in this section under Background information: What is behind the structure search?

For an introduction to the structure search, a video tutorial is available.

In the structure search, chemical compounds can be searched via a structure editor based on their molecular structure, via the international non-proprietary name or via import of SMILES/InChI code. Furthermore, substructure and similarity searches are possible starting from an identified substance. Various links (e.g., to PubChem and the DrugBank database) can be used to obtain further (drug) information. For an identified substance, publications can be searched in PubPharm.

Start the search

The following search options are available:

• Draw a compound in the editor and search with "Search by Structure". Information about the editor can be found on the help page of the open-source Epam Ketcher used in PubPharm.
• Enter the name of a substance and start the search with "Search by Name". The identified compound will be loaded into the editor and can be modified there if necessary.
• Enter the InChI or SMILES code of a compound and start the search with "Import Structure". The structure is loaded into the editor and can be edited if necessary.

Result visualisation

If the compound is identified in the database, a short overview is displayed. The result contains information from PubChem and, if available, a short description from DrugBank. In addition, links to the entries in PubChem, DrugBank and the ChEMBL database (if available), as well as to the search interface of the European Patent Office are displayed for each result. "Search in PubPharm" can be used to search directly in PubPharm for publications on the substance, whereby the first 5 synonyms from PubChem are included in the search.

Further search options

Starting from an identified compound, the structure search can be continued:

• Substructure search: compounds are searched that contain the substance as a substructure.
• Similarity search: A search for structurally similar compounds is performed.

The search results are displayed as a list. Each result can be used again as a starting compound for new searches. Via "Load into Editor" the structure can be loaded into the editor and modified there.

If a structure was entered in the editor for which no matching substance is available in the PubChem database, the hit is marked with "?". Based on this compound, a substructure or similarity search can be performed.

Background info: What is behind the structure search?

The structure search is based on data and open-source tools of the PubChem database. The search results are determined and prepared for display via the PUG REST API. The substance search (based on the structure or name search) and substructure and similarity searches are performed via the PubChem PUG REST API. The PubChem search parameters and, where necessary, PubChem fingerprints are used.

When searching for active substances and diseases/symptoms, suggestion lists of context-similar "related substances", "related diseases" and "related genes" are generated if available (i.e. a certain number of references must be available). For example, when searching for the inhibitor of dipeptidyl peptidase-4 sitagliptin, other drugs used to treat diabetes mellitus, e.g., metformin and exenatide, are listed among the related substances. Another example is the listing of kinase inhibitors when searching for vemurafenib in PubPharm. For more complex queries with several search terms (e.g. "diabetes mellitus AND (exenatide OR metformin"), the suggestion lists are displayed for the first substance or disease occurring in the search (here: diabetes mellitus). Further information can be obtained via links to the UniProt and KEGG (Kyoto Encyclopedia of Genes and Genomes) databases. In addition, the search can be continued in the structure search, the Drug Overviews or in the Narrative Service, depending on the information required.
This suggestion function was developed by the Institute of Information Systems and is based on machine learning technologies (further information). Text mining processes are used to link bioactive compounds with corresponding diseases and genes. The number of scientific publications linking active substances, diseases and genes is counted. The most frequent associations are displayed in the suggestion lists. A click on "Drug Overviews" or "Narrative Service" leads to a targeted search for this literature.

The Narrative Service (a unique PubPharm characteristic) enables a precise and structured literature search by defining the relationships between different pharmaceutical concepts (e.g., active ingredients, dosage forms, plant families, targets, (laboratory) methods, etc.) in the search query. To do this, enter your information need (research question) as a subject-predicate-object relationship in the central input line. In addition, structured literature overviews can be generated using variables on pharmaceutically relevant concepts instead of concrete subjects or objects.

Various video tutorials are available as an introduction to the Narrative Service. More information can be found here.

Searching in the Narrative Service

Query Builder

• Step 1: As the first step, the query must be formulated. The Narrative Service enables the targeted search for interactions between pharmaceutical concepts such as active substances, diseases and targets. The concepts (here metformin and disease) must be entered in the input mask (see arrow 1), which is supported by corresponding suggestions (auto-completion). You can also get an overview of the concepts and variables/placeholders via the "Browse" button. Afterwards, the interaction you are looking for (here treats) must be selected.
• Step 2: You can start the search by clicking "Search" (see arrow 2). Alternatively, you can add the desired interaction to the current query via "Add". All interactions must be present in the corresponding document hits (logical AND).  
• Step 3: After a successful search, you can view the results (see arrow 3). In the example search, the search was for diseases that are treated with metformin. In this case, the answer is a nested list: 4546 documents deal with the treatment of diabetes mellitus and 4118 with the treatment of diabetes mellitus type 2. By clicking on such a group, the individual results can be viewed.
  An important note: All displayed associations were automatically extracted from the literature. Unfortunately, possible incorrect extractions cannot be excluded.
• Step 4: If a document appears to be particularly relevant, you can click on the coloured graph to display the content of the document (see arrow 4). The Narrative Service then visualises the title, abstract, authors and other metadata. Detected concepts are highlighted in colour in the texts, giving you a quick overview of the content. On the right-hand side, a graph structure is visualised that shows interactions between concepts that can be extracted from the corresponding document. With a click on the document ID, you are redirected to PubPharm. Another function is the so-called "Provenance" button. A click on this button shows you the text passage from the corresponding document that supports the searched interaction.
• Step 5: The Narrative Service has various filter functions (see arrow 5). For example, the results can be limited to the date of publication. You can also use a free text field to filter the document hits by searching for word phrases in the titles. For example, entering child and pressing the enter key would limit the results to documents that contain "child" as a phrase in the title.

Keyword Search

If the interaction between search words is still unclear, a search can alternatively be started via "Keyword Search". With this search option, the pharmaceutical concepts (e.g., active ingredients, diseases and targets) can be entered in the input mask without selecting a predicate. Each concept (search word) is entered individually and confirmed with the Enter key or the "Add" button. After "clicking" on "Search", possible interaction patterns with different predicates are suggested. "Clicking" on the desired interaction pattern starts the literature search.

Below you can see an example search with the concepts metformin and diabetes mellitus.

The Drug Overviews service (a unique PubPharm characteristic) is an extension of the Narrative Service. In this service, you can search specifically for an active substance and receive (I) existing information from the ChEMBL database (including pkA and logP values, molecular mass, etc.) and (II) extracted associations from the literature (including interactions with other active substances, information on targets and laboratory methods as well as indications and dosage forms).

An introduction to the Drug Overviews is available here (video tutorial). Information on the respective active substance is also visualised in the Drug Overviews in network views. A video tutorial introducing this feature is available here. More information on the Drug Overviews can be found here.

Searching in the Drug Overviews

Step 1: The Drug Overviews are substance-centred overviews generated from the current research literature. To start a search, enter a desired active substance in the input mask (see arrow 1). An auto-completion supports your input.

Step 2: Review of the fact sheet. The Drug Overviews first summarise known properties of an active substance such as the molecular formula, the mass, calculated logP and pKA values (see arrow 2). Also displayed is the InChI code, a link to the corresponding entry in the ChEMBL database and a link to the PubPharm structure search. 

Step 3: Exploration of interactions and associations. On the left side, you will see a navigation bar (see arrow 3) that will take you to the various extracted information in the Drug Overviews at any time.
An important note: All displayed associations were automatically extracted from the literature. Unfortunately, possible incorrect extractions cannot be excluded.

Drug-Target-Disease Networks

The Drug Overviews visualise extracted associations to other drugs, diseases and targets in a so-called Drug-Target-Disease Network (see arrow 4). Here, the respective concepts are represented as nodes and their associations as edges. The numbers on the edges indicate how many documents from the literature support the respective association. Clicking on an edge triggers a corresponding search in the Narrative Service for literature. If you click on a node in the network, the associations between the clicked node and the already existing network are displayed. If you want to hide diseases, e.g., you can do so in the upper left area of the networks.

Drug aspects

Further down in the Drug Overviews, various imformation of the active substances are compiled (see arrow 5). For example, the first box shows possible indications of the active substance. In our example, the first group is diabetes mellitus, which is supported by 4546 documents from the literature (see the small blue box). The Roman numeral "IV" indicates that the treatment of this disease with metformin is in clinical phase 4 (based on ClinicalTrials.gov). Clicking on the number will then direct you to the appropriate source: corresponding literature search in the Narrative Service or an entry in ClinicalTrials.gov. 

Other aspects include dosage forms (administration), interactions with targets, laboratory methods, species, human target groups ("HealthStatus"), drug associations and interactions, as well as adverse effects and effects on tissue.

There is a search bar in the upper right area of each box that you can use to filter in the box (e.g., "Diabetes"). You can also adjust the sorting in the individual boxes. Below the last box, the latest publications on the active substance are displayed.

• The web design of PubPharm is responsive. That means the website is displayed well on different devices - from smartphones to desktop PC.
• No matter from which device you are searching, all tools and functions, e.g., the service Drug Overviews, filters, favourites lists and export functions, can be used. The only exception is the structure search.
• Please note that the layout of the individual elements may differ on different devices. In the desktop view the info boxes are arranged horizontally, while in the smartphone view they are found in a vertical order.