Science Forum: Author-sourced capture of pathway knowledge in computable form using Biofactoid

  1. Jeffrey V Wong  Is a corresponding author
  2. Max Franz
  3. Metin Can Siper
  4. Dylan Fong
  5. Funda Durupinar
  6. Christian Dallago
  7. Augustin Luna
  8. John Giorgi
  9. Igor Rodchenkov
  10. Özgün Babur
  11. John A Bachman
  12. Benjamin M Gyori
  13. Emek Demir  Is a corresponding author
  14. Gary D Bader  Is a corresponding author
  15. Chris Sander  Is a corresponding author
  1. The Donnelly Centre, University of Toronto, Canada
  2. Computational Biology Program, Oregon Health and Science University, United States
  3. Computer Science Department, University of Massachusetts Boston, United States
  4. Department of Cell Biology, Harvard Medical School, United States
  5. Department of Systems Biology, Harvard Medical School, United States
  6. Department of Informatics, Technische Universität München, Germany
  7. Department of Data Sciences, Dana-Farber Cancer Institute, United States
  8. Broad Institute, Massachusetts Institute of Technology, Harvard University, United States
  9. Laboratory of Systems Pharmacology, Harvard Medical School, United States
  10. Department of Computer Science, Department of Molecular Genetics, University of Toronto, United States
  11. The Lunenfeld-Tanenbaum Research Institute, Mount Sinai Hospital, Canada
  12. Princess Margaret Cancer Centre, University Health Network, Canada

Abstract

Making the knowledge contained in scientific papers machine-readable and formally computable would allow researchers to take full advantage of this information by enabling integration with other knowledge sources to support data analysis and interpretation. Here we describe Biofactoid, a web-based platform that allows scientists to specify networks of interactions between genes, their products, and chemical compounds, and then translates this information into a representation suitable for computational analysis, search and discovery. We also report the results of a pilot study to encourage the wide adoption of Biofactoid by the scientific community.

Introduction

Biological pathways organize sets of molecular interactions and reactions that underlie cellular processes. This information can be used for experimental design, interpretation of genomics data (Khatri et al., 2012), understanding of disease mechanisms (Chinen et al., 2016; Santos et al., 2014), and identification of therapeutic targets (Mack et al., 2014).

To manage, visualize and interpret the large amount of available pathway information, researchers require computational tools such as pathway information resources, structured data representation standards and analysis software (Demir et al., 2010; Franz et al., 2016; Jassal et al., 2020; Pratt et al., 2015; Rodchenkov et al., 2020; Shannon et al., 2003). These computational tools can access the increasing amounts of pathway and interaction data (Bader et al., 2006) collected from organizations that either centrally curate the literature (Gene Ontology Consortium, 2015; Jassal et al., 2020) or crowdsource pathway information from the community of researchers (Slenter et al., 2018). Nevertheless, these existing efforts are not able to achieve wide coverage of the rapidly growing scientific corpus (over 1.3 million new PubMed articles per year; Bornmann and Mutz, 2015; Cordero et al., 2016), arguing for the development of more scalable and sustainable approaches (Attwood et al., 2015; Imker, 2018).

For reasons of efficiency and accuracy, structured knowledge of biomedical discoveries could be provided directly by the original authors of research reports, rather than post-hoc, through the curation efforts of knowledge base teams. By analogy, molecular 3D structures submitted directly by authors to the Protein DataBank (PDB) have become a key resource among structural biologists (Berman et al., 2000), and similar community practice led to direct submission of DNA sequences and transcriptomics information to public databases. Indeed, the importance and value of having such research outputs available in a community resource are underscored by the fact that data deposition is often a requirement of publication and funding. In contrast, there are few efforts and little technology to support direct submission by authors of biological pathway information and related knowledge in computable form.

Here we introduce Biofactoid (biofactoid.org), a web-based software system that empowers authors to capture and share structured human- and machine-readable summaries of molecular-level interactions that are supported by evidence in their publications. Without such a curation support tool, the onus would be on authors to handle a series of complex tasks involved in converting their knowledge to a computational form and depositing it in a suitable knowledge base. To overcome this and other significant barriers to computable knowledge acquisition and sharing, we developed Biofactoid to ease pathway curation and to rapidly generate expressive, structured representations with minimal user training. Structured knowledge newly acquired in this way becomes part of the global pool of pathway knowledge and can be shared in resources such as Pathway Commons (Rodchenkov et al., 2020), Network Data Exchange (NDEx) (Pratt et al., 2015), and STRING (Szklarczyk et al., 2017), to enhance information discovery and analysis.

Authors can use Biofactoid to share structured information from research articles both as part of the publication process and outside of it. Personal email invitations sent to authors as part of a pilot outreach effort resulted in them contributing over 95 articles to Biofactoid from a range of journals; collectively these pathways have described a total of 426 entities (340 unique) from eight organisms (H. sapiens, M. musculus, R. norvegicus, D. rerio, S. cerevisiae, D. melanogaster, C. elegans, E. coli) involved in 287 interactions. The development of Biofactoid and related computational tools helps support human-computer communication and inference algorithms in an ecosystem in which scientific reasoning is increasingly assisted by broad and deep knowledge computation.

Results

Data sharing workflow

Biofactoid enables molecular-level detail of biological processes reported in articles to be shared in a structured format accessible to humans and computers. Interactions (binding, post-translational modification and transcription/translation) involving molecules of various types (proteins, nucleic acids, genes, or chemicals such as metabolites and drug compounds) can be represented. Users begin by entering the article title, or identifier (e.g. PubMed identifier or Digital Object Identifier). Next, article metadata, including authors, abstract and journal issue, are automatically retrieved. Next, users draw a network of biological entities and interactions using the Biofactoid curation tool, which has an easy-to-use interface similar to graphical illustration software (e.g. Microsoft Powerpoint, Adobe Illustrator) which will be familiar to users, but with the added ability to generate structured data from the author-drawn biological pathway. The major features of the Biofactoid curation tool include support for:

  1. Molecular entities. Genes, gene products and chemicals are created in the network using an “Add a gene or chemical” tool, which creates a node (circle) that users label (Figure 1A). Biofactoid automatically matches the label with a record in an external database: NCBI Gene for genes and their products and ChEBI for small molecules (Brown et al., 2015; Hastings et al., 2016). This match represents the top hit of a search based on the similarity between a user’s label and a database record’s list of common names and synonyms; for genes, organisms are given priority based upon the organism of genes previously added to the network. Users can update the match by selecting another organism (e.g. human or mouse p53) or update the automatically inferred gene product type (i.e. RNA or protein). Alternatively, users may assign an alternate database entry from a list of search results. The system supports human and selected model organisms (M. musculus, R. norvegicus, S. cerevisiae, D. melanogaster, E. coli, C. elegans, D. rerio, and A. thaliana). The system is fast (response times of 100 milliseconds or less) and can be easily extended, for instance, to support SARS-CoV and SARS-CoV-2, which we added in response to the burst of publications related to the COVID-19 global pandemic (Ostaszewski et al., 2020). The matching process is also accurate, as measured using tests that use entity names from research articles as queries and assessing the quality of the search result. In over 90% of the cases, the correct result was first among search hits and was among the top 10 in over 97% of cases. Thus, Biofactoid can offer an accurate database identifier match using only the author-provided entity name. This represents a major advance in usability compared to traditional gene and chemical name querying systems that only work with exact name string matching and are often slow or unreliable. In fact, we were forced to build our own system after trying to use major existing systems that turned out to have high failure rates for our use case.

  2. Interactions. A “Draw interaction” tool lets users link two nodes by clicking one and dragging to the other (Figure 1B). Edges can have type and direction: a pointed arrow indicates stimulation or activation; a ‘T-bar’ arrowhead indicates inhibition or repression; and an undirected line indicates any other interaction. The interaction mechanism can be refined by selecting from a list that includes binding, transcription/translation, or common forms of post-translational modifications.

  3. Complexes. Molecular complexes can be created by dragging genes or chemicals close to each other, resulting in a box that encloses them, which can also be labeled (Figure 1B).

  4. Context. A context input box that encourages authors to “List terms that describe the context (e.g. T cell, cancer, genome stability)”. Allowing authors to provide a list in free text form has the benefit of accommodating any terms or concepts they consider important for specifying the research context.

  5. Automatic saving and co-editing. Pathway representations in Biofactoid are automatically saved as changes are made and a live-sync capability enables multiple authors to collaboratively edit the same pathway, analogous to Google Docs.

The Biofactoid curation tool.

Curation in Biofactoid involves drawing a network of relationships between genes or chemicals. (A) Genes and chemicals are represented by circles (nodes, highlighted in blue) where users provide a label, the type of gene product, and the organism. A search engine matches the label to a corresponding record from a database of genes or chemicals. (B) Relationships are represented by connecting genes, chemicals and/or rectangular complexes (shown in grey) with plain lines (when neither activation nor repression occur), arrows (to indicate activation), or ‘T-bars’ (to represent repression). Users select the mechanism that best describes the interaction. Complexes are represented as genes and/or chemicals enclosed by a box (e.g. the grey box labelled "Activated ras").

Once complete, the pathway data is validated by pressing a "Submit" button. At this point, the user may address any potential quality issues flagged by the system (e.g. unlabelled nodes, empty document) before confirming their submission. Submitted pathways are shared publicly with the research community and users retain the ability to edit their pathway via the link they received at the start of the workflow. Authors who wish to have their submissions removed from Biofactoid can do so by contacting the support team (support@biofactoid.org).

Data sharing and exploration

When research findings are shared through Biofactoid in a structured and computable manner, curated knowledge is connected to, and becomes part of, a collective pool of computable knowledge that the community can access in different ways. Visitors to the Biofactoid website (biofactoid.org) may browse recently added articles, and a graphical abstract of each new submission is posted to Twitter (twitter.com/biofactoid). Each entry is automatically linked to its associated authors, article information and structured data and presented in an interactive Biofactoid Explorer web app (Figure 2A).

Users can select any entity in the pathway diagram to see more information about genes (via NCBI Gene database links) and chemicals (via ChEBI database links). To support attribution, each author of an article is automatically linked to their profile in the Open Researcher and Contributor ID database (ORCID; orcid.org). Finally, a set of “Recommended articles” are generated to help the user explore similar knowledge in the literature.

To generate these recommendations, we first retrieve an article’s references, articles it is cited by and “Similar articles” from PubMed using the NCBI Entrez Programming Utilities service (Wheeler et al., 2006). These are combined with articles that mention or provide evidence for interactions involving any of the molecular entities in the Biofactoid entry, which are accessed from curated pathway and interaction databases as well as resources that aggregate interaction information directly from the biomedical literature using natural language processing (NLP) tools (e.g. REACH, INDRA) (Gyori et al., 2017; Valenzuela-Escárcega et al., 2018). Recommended articles are ranked by determining how similar their titles and abstracts are to that of the Biofactoid article, using a deep-learning-based method we developed (Giorgi et al., 2021). The list of recommendations is context-sensitive in that articles are shown relevant to the part of the pathway selected by a user.

Biofactoid data is connected to information sources and establishes a bridge between related pathways.

(A) The Biofactoid Explorer is an interactive web app that publicly presents each author-curated entry alongside their article. Yellow arrows indicate how curated information is connected to outside knowledge bases. A “Network overview” (left) displays information about the article and pathway as a whole; a “Network item view” (right) displays information for a selected item (e.g. interaction, protein). (B) Biofactoid data establishes a bridge between related pathways described by structured biological knowledge from distinct data sources. Two author-curated interactions submitted to Biofactoid (red edges) bridge previously distinct pathways from the Reactome Pathway Database involved in mitochondrial biogenesis (left) and provide a new, more direct regulatory route between two mitochondrial genes (right). Pathway and interaction information was provided by Pathway Commons (pathwaycommons.org), a web resource that provides a single point of access for multiple public interaction and pathway databases. Details regarding the generation of these networks can be found in the section “Visualization of network data across sources” in Materials and methods.

Beyond the “Explorer”, Biofactoid data represented in the formal BioPAX ontology is interoperable with external pathway and interaction knowledge using technologies for BioPAX processing and analysis, including Paxtools (Demir et al., 2013) and cPath2 (Cerami et al., 2006). This permits author-contributed information to be connected to existing information, and leverages tools for searching, visualizing and analyzing data across different sources.

For example, researchers can ask the question: “Which known pathways are related to an interaction in Biofactoid?” As an example answer to this question, a search of Pathway Commons shows how two interactions made computable by authors using Biofactoid (i.e. “SENP1 activates SIRT3”, Wang et al., 2019a; “SIRT5 activates UCP1”, Wang et al., 2019b) unify previously disconnected pathways for mitochondrial biogenesis from the Reactome Pathway Database (Jassal et al., 2020; Figure 2B; see the section "Visualization of network data across sources" in Materials and methods). Researchers can also ask: “How are two genes related?” As another query example, a search of Pathway Commons shows how an interaction contributed by an author using Biofactoid (“SENP1 activates SIRT3”, Wang et al., 2019b) provides a new, more direct regulatory pathway linking the mitochondrial proteins SENP1 and SOD2. Thus, Biofactoid directly contributes to building a more complete and comprehensive collection of biological pathways.

User testing and pilot study involving authors and journals

We tested the Biofactoid software over multiple iterations with researchers from the authors’ institutions, journal editors and finally pathway authors, updating the software each time based on user feedback. This process supported two major goals: to improve the user experience of the software for authors, who are typically unfamiliar with curation and structured data concepts, and to develop a model for integrating Biofactoid into the publication process with journals. We performed user testing, with a user-centred design process, of a prototype of the Biofactoid curation tool (Norman, 2001; Norman and Draper, 1986). We improved our software based on this feedback and then repeated this process a second time (see the section "User testing and pilot study" in Materials and methods). We also gathered more informal feedback that was used to improve the system.

Once Biofactoid software achieved a sufficient level of sophistication and completeness, satisfying many of these initial users, we engaged journal editors to determine the feasibility of using Biofactoid to capture information from authors. Our pilot study consisted of three phases (Figure 3). Phase I introduced Biofactoid to journal editors via an “author-simulation”, which began with a mock email invitation asking the editor to use Biofactoid to curate a selected article, and ending when a pathway from that article was input into the system by the editor using the Biofactoid web-based user interface. Phase II involved sending email invitations to a small number (N = 15) of authors from a single journal, with multiple successful responses, proving that authors can use the system without any direct support from the Biofactoid team or journal. Phase III measured engagement rate by emailing 260 authors of articles with content suitable for Biofactoid from 16 selected journals (Table 1).

Roughly 8.5% of these authors successfully shared their research in Biofactoid, proving that many authors can and will use Biofactoid (Figure 2A). We also found that 10% of articles across selected journals are suitable for inclusion in Biofactoid based on the current set of supported concepts (Table 1). In addition to proving that Biofactoid can be independently used by authors to capture pathway knowledge, these results indicate a willingness among authors in the research community to use the system. These also demonstrate that the information captured by authors may not be present in any existing pathway database and helps connect previously unconnected entities and pathways in these databases (Figure 2B). We also observed that pilot study authors invariably chose to represent only their novel findings, rather than previously known pathway information found in their background sections.

Biofactoid pilot study.

A three-phase pilot tested the feasibility of Biofactoid and involved journal editors and authors of research articles. Phase I and II involved editors and authors whose articles were recently published. In Phase III, 2,065 published articles were screened and authors of suitable articles were invited to Biofactoid. The articles screened were from 16 journals (Table 1).

Table 1
Prevalence of articles with pathway knowledge suitable for Biofactoid.
ISSNJournal*Coverage² [Vol. (Issue)]Articles screenedHits% Hits
2211–1247Cell Reports30(1) - 32(11)95310910.3
1097–4164Molecular Cell73(1) - 79(6)7258510.5
1549–5477Genes & Development34(1-2) - 34(17-18)931513.9
1476–4679Nature Cell Biology22(4) - 22(9)841010.6
1083–351 XJournal of Biological Chemistry295(31) - 295(37)210219.1
Total--2065240-
Weighted Average----10.4
  1. *

    Only journals in which at least 80 ‘hits’ were identified were included. A ‘hit’ is an article that provides direct evidence for a molecular interaction that can be captured by Biofactoid. The ability of Biofactoid to capture an interaction depends upon the type of bioentities, the relationship types and organisms described in the article. In total, articles from 16 journals were screened: EMBO; Molecular and Cellular Biology; Cell; Cancer Cell; iScience; J Biol Chem; Cell Metabolism; Science; Nature Genetics; Science Signaling; Science Advances; Immunity; Cell Reports; Molecular Cell; Genes & Development; Nature Cell Biology. ²Coverage indicates the span of journal issues that were included. Only primary research articles from each issue were screened.

Discussion

Biofactoid focuses on the capture of published pathway information in computational form to augment discovery, attribution and communication of scientific knowledge. In developing our strategy, we have considered how technology can best be used to aid authors. The result is a generic approach that can be extended to capture other types of biological knowledge, at the source of knowledge creation, in computable form. To be successful, Biofactoid development has focused on providing key elements of efficiency, incentives and better technology for human-computer interaction.

The computable knowledge capture model proposed here includes formal knowledge representation using ontologies, easy to use curation support software that links molecules to corresponding database identifiers (normalization), submission to widely available knowledge resources, a connection with the publication process and author attribution. This model expands on prior work defining digital abstracts and building crowdsourcing efforts for pathway data and software (Ceol et al., 2008; Gerstein et al., 2007; Leitner et al., 2010; Liechti et al., 2017; Slenter et al., 2018; Table 2). In principle, Biofactoid technology can be extended to support any network of entities or concepts and their relationships (i.e. a knowledge graph) collaboratively built by a community of knowledge generators.

Table 2
Comparison of non-centrally curated biocuration projects.

Comparison of projects that support community curation of pathway and interaction knowledge as their primary concern.

ProjectScopeSourceIntegrated curation toolAutomatic entity recognitionSingle-article orientedRef.
BiofactoidPathwayAuthorThis study.
Structured Digital Abstract (FEBS Letters)Protein-ProteinAuthor--Ceol et al., 2008; Leitner et al., 2010; Gerstein et al., 2007
WikiPathwaysPathwayAnyone---Slenter et al., 2018
SourceDataFigureAuthor-Liechti et al., 2017

A key assumption underlying Biofactoid is that scalable knowledge capture can be achieved by community curation. However, crowd-sourcing of biological research findings presents unique challenges that must be considered if widespread use is to be realised. First, the biological research literature contains a wide array of biological entities and concepts, with descriptions that range from detailed to abstract. As support for more biological entity types are added, a curation system can become more complex, increasing the barrier to entry and making it more likely users will require training to use it effectively. Thus, a successful strategy should support wide coverage of biological concepts while minimising complexity.

In a related issue, biological researchers are accustomed to communicating pathway concepts using visual depictions that employ symbols and conventions, which, although comprehensible to the community, are often ad hoc, abstract and potentially ambiguous (e.g. “A activates B”). Thus, a successful curation system should accommodate ways that biologists articulate concepts, while also retaining the ability to represent the information in a rigorous and computable manner.

A second major challenge to achieving uptake is raising awareness and communicating the goals of the project within the research community. Pathway information is qualitatively distinct from items that researchers are accustomed to depositing (e.g. DNA sequencing data), as pathways are models that represent the outcome or interpretation of multiple experiments. Ideally, potential contributors should be able to independently determine whether their articles are suitable for Biofactoid.

A third challenge to realising wide uptake is incentivizing authors to contribute the knowledge in their article. In particular, authors will be more likely to contribute if the information they enter will be useful, either directly to themselves and/or others in their community. Scalable and sustainable growth is more likely to be achieved when researchers directly use and benefit from Biofactoid information and then decide to contribute themselves.

We designed Biofactoid to enable researchers to express details of their research quickly and effectively, with minimal instruction. To preserve ease of use, we have purposely limited both the scope of biological concepts and the granularity of detail that can be captured, though this can be expanded in the future. With regard to entities, only major model organisms and SARS-CoV(–2) are supported; we have not included more abstract concepts like cellular processes (e.g. apoptosis) or fine details like reaction stoichiometry, molecular sub-regions (e.g. promoter region) and modification (i.e. amino acid position, mutations). With regard to interactions, the system does not explicitly support biological concepts that involve spatial (e.g. cellular location) or temporal considerations (e.g. ordered recruitment of proteins).

The aforementioned details are supported by the underlying BioPAX ontology and additional concepts (e.g. global biological context, that is, the environment in which pathway components act) can be supported by referencing other controlled vocabularies or databases from the BioPAX model. While the current set of features constrains the scope of capturable research facts, it enables Biofactoid to be accessible by a wide audience. Nevertheless, future work will expand the range of biological concepts supported.

Distributing curation effort among authors has the potential to be efficient but also accurate because of their first-hand knowledge of the research. To ensure authors can use Biofactoid effectively we developed an automatic entity recognition tool that accurately matches the names of genes and chemicals provided by authors to an entity database record. We also flag potential issues with author input at submission so they can be corrected immediately (e.g. unlabelled entities, genes from multiple organisms). Finally, we evaluated our system through user testing to improve the speed and accuracy with which users can curate interactions described in text. In the future, we will develop strategies to monitor the accuracy of pathway data post hoc and identify and amend inaccuracies we discover.

To monitor data accuracy, we will seek out experts to independently rate pathway items, which is an approach that has been successfully applied to problems in computer-aided diagnostics and healthcare (McHugh, 2012). Specifically, we will recruit at least two experienced curators of pathway knowledge to assign quantitative ratings for a sample of Biofactoid pathways with respect to accuracy of database identifiers for entities, accuracy of interactions and completeness (i.e. missing/extraneous elements). Curator scores will be compared to estimate variability. We will also enable users to take action when they identify inaccurate information by contacting the article author to ask them to edit and correct the record. Failing this, we will enable users to directly report an issue with any particular pathway to the Biofactoid support team and will reserve the right to lock editing of pathway items, mark them as disputed or remove them from the collection.

A convenient time to capture knowledge is during publication, when authors - the primary source of knowledge - are most aware of the details of their report and when they are typically asked to deposit other types of data (e.g. DNA sequencing or gene expression data). As with any innovative approach, introducing the system to editors, journals and their respective authors requires an investment of effort for all parties. It will likely be necessary to directly engage individual journals, to familiarize them with the system, and accommodate new requirements they may have. There are many ways to offer Biofactoid to authors, including, for instance: integrating it with the peer-review process, requiring it as a condition of publication or recommending it to authors post-publication-acceptance.

As a demonstration, our pilot study engaged publishers to establish requirements for using Biofactoid within their publication pipeline, ideally as a condition for acceptance, and to drive greater awareness of the system among the wider readership. The pilot study also revealed that directly inviting authors of suitable articles to contribute, even after publication, is a viable engagement strategy. Ideally, we would invite all authors who publish suitable content to use Biofactoid, however, the rapid pace of publication makes it impractical for the Biofactoid team to manually screen articles, arguing for the development of systems (e.g. using NLP) to automate article triage. If successful, this approach could be used to regularly identify relevant articles from the pool of all new entries indexed by PubMed.

Another approach to reach users is to notify those whose research articles are referenced by information curated in Biofactoid. For this purpose, we have developed a system that automatically notifies authors when their research is linked to papers curated in Biofactoid (e.g. by citations or because of related content) so that they may further explore this information and curate their own pathway knowledge. More generally, we intend to engage the research community by cultivating partnerships with knowledge database organizations including those that curate information about articles (PubMed), biomolecules (e.g. UniProt, NCBI, ChEBI), pathways and interactions (e.g. Reactome, STRING), model organisms (e.g. Saccharomyces Genome Database [Lang et al., 2018]) and researchers (e.g. ORCID).

To improve the efficiency and utility of Biofactoid, we are developing NLP technology to support authors in using Biofactoid, as well as to enable the representation of pathways in textual form (Giorgi et al., 2019; Giorgi and Bader, 2018; Giorgi and Bader, 2020; Valenzuela-Escárcega et al., 2018). To better accommodate the way individual users prefer to communicate, the system will accept both graphical and textual entry of pathway information as well as automated conversion between these two forms. Assistants will support users to rapidly compose their network, enabling them to add new information from a list of interactions identified in their article by NLP technology (Gyori et al., 2017). Further development of NLP methods for the reliable extraction of pathway information from the publication full-text, combined with development of new tools for curation and quality control, will help realize broad and accurate coverage of pathway knowledge in computable form.

We are also improving the search and exploration functions of the system, ensuring Biofactoid information is connected to other useful knowledge, and that related information is easily accessible starting from a Biofactoid entry. In the future, we envision that information entered by an author in Biofactoid will serve as a custom query that can be used to regularly notify the author of new information (e.g. from other publications) related to their interests, such as interacting molecules and phenotypes. This work provides a basis for the development of new technologies to make scientific knowledge more computable and accessible and help researchers identify information within the rapidly growing scientific corpus.

Materials and methods

Software implementation

Request a detailed protocol

Biofactoid is written in JavaScript. The backend server uses a microservice architecture, with Node.js, Express and RethinkDB. Client-server data synchronization, supporting automatic saving and concurrent editing, uses websockets and a model similar to differential synchronization (Fraser, 2009). The front end uses React and Cytoscape.js (Franz et al., 2016), for network drawing and is optimized for desktop and mobile devices. An administrative dashboard, as well as user curation workflow automation features (e.g. automatic email generation) are integrated into the Biofactoid system to aid system scalability.

Visualization of network data across sources

Request a detailed protocol

The network visualizations which summarize a search of known pathways and interactions related to interactions captured by Biofactoid (Figure 2B) were created by combining data available in Biofactoid and Pathway Commons (version 12; Rodchenkov et al., 2020). Data for the network with subheading “New connections among mitochondrial pathways” was obtained by searching pathwaycommons.org for the genes SENP1, SIRT3, UCP1 and SIRT5, which are part of the Biofactoid entries displayed in Figure 2A. In the search results, the Reactome pathways entitled “SUMO is proteolytically processed”, “Transcriptional activation of mitochondrial biogenesis” and “The fatty acid cycling model” were each downloaded as a Simple Interaction Format (SIF) file. Also in the search results, the “Interactions” panel was selected and the network was downloaded as a SIF file. Using Cytoscape (cytoscape.org) (Shannon et al., 2003), each SIF file was imported as a new network and the four individual networks were merged. The Biofactoid interactions from Figure 2A were added to this merged network which was manually laid out and styled to achieve the final result. Data underlying the network with figure subheading “New regulatory route from SENP1 to SOD2” was obtained by querying the Pathway Commons SIF Graph web service (http://www.pathwaycommons.org/sifgraph/swagger-ui.html) to identify regulatory paths between SENP1 and SOD2, in which the maximum number of interactions separating the two genes (i.e. limit) is 3. This resulting SIF file was imported into Cytoscape and then manually laid out and styled to achieve the final result.

User testing

Request a detailed protocol

Two rounds of user-centered testing were performed, once following the development of the initial prototype of the Biofactoid curation tool, and once again after further development of the tool. Test participants were drawn from volunteers who responded to individual email invitations. In each round of testing, participants were provided with a one-page handout that provided an overview of the project, brief instructions on how to use the curation tool, followed by one or more text excerpts describing a biological interaction that they were asked to input. Ten researchers participated, including faculty members (n = 2), graduate students (n = 3) and research staff (n = 5). Participants used the software on an individual, in-person basis from a provided computer, during which screen and voice were recorded. All participants provided written consent to volunteer and have their session recorded. We also elicited user feedback both in-person and by email for a version of Biofactoid able to support the entire "Data sharing workflow", including: user onboarding material presented on the biofactoid.org homepage; a system to identify articles based on information entered by authors (i.e. title); the curation tool; and the Biofactoid Explorer. Those providing feedback for the "Data sharing workflow" included three curators from biological databases (Reactome and UniProt); participants were asked to input an article of their choosing, starting at the homepage.

Pilot study

Request a detailed protocol

A group of two editors from Molecular Cell and Cell Systems volunteered to evaluate Biofactoid in response to a direct invitation. A pilot study was designed with three phases. Phase I introduced the software to the journal editors to determine whether it met basic requirements for capturing pathway knowledge in articles in their journals and identify any issues precluding interaction with authors. To do this, we simulated the author’s experience using the Biofactoid curation tool to input pathway information. First, an editor and a member of the Biofactoid team selected a recently published article from the Molecular Cell, "Online Now" (https://www.cell.com/molecular-cell/newarticles) website, based on a single criterion: the article reported evidence directly supporting at least one interaction type supported by Biofactoid. Next, editors were sent a mock email invitation addressed to the selected article’s corresponding author. Invitations contained brief instructions on system use and a link to the Biofactoid curation tool, which had the article metadata already filled for the corresponding Biofactoid document. Finally, an editor was observed using the curation tool to enter the pathway information contained in the article remotely via video-conferencing software. A feedback session was carried out with two editors immediately following the simulation and a subsequent video-conference was scheduled to capture additional recommendations for the software and author workflow.

Phase II validated the Biofactoid software and the workflow with authors. First, a set of suitable articles that were recently published by Molecular Cell (Online Now) were selected with journal editors. Second, email invitations addressed to the corresponding author of each article were sent from the institutional email account of a Biofactoid team member. As in Phase I, emails contained a link to the curation tool with their article information already filled in. To our knowledge, these authors had no prior contact with the journal with regard to Biofactoid. Follow-up reminders were sent after one week to authors who had not used the system. We repeated this process with two additional groups of articles (ten total) from Molecular Cell (Online Now) that were selected by the Biofactoid team.

Phase III examined the author participation rate following an invitation. First, we selected a total of sixteen journals, with a preference towards those with a mandate for molecular biology and mechanistic studies backed by biochemical-level evidence. Journals with broad and diverse readership (e.g. Nature, Science; Table 1) were also included. Next, articles from entire journal issues were screened by a Biofactoid team member (JVW) for studies reporting evidence of one or more biological interactions compatible with Biofactoid. As before, email invitations were addressed to an author of the article. In contrast to previous emails, the recipient was only provided with a link to the biofactoid.org homepage in order to begin the data sharing workflow (no pre-fetching of manuscript metadata). A maximum of two weekly reminder emails were sent to those that had not submitted their pathway information. We manually analysed user feedback to identify bugs and aspects of the system that were not intuitive or prevented users from entering data and prioritized those to be addressed with Biofactoid software improvements.

Software availability

Request a detailed protocol

Biofactoid is available to biomedical researchers for data sharing and exploration at biofactoid.org. To support bioinformaticians and software developers, all user-contributed pathway data is openly accessible in multiple standard formats: JavaScript Object Notation (JSON) for raw data; Systems Biology Graphical Notation Markup Language (SBGN-ML) pathway visualization format using the Process Description notation (Le Novère et al., 2009; van Iersel et al., 2012); and BioPAX (Demir et al., 2010). All code, documentation and data are open source and freely available through GitHub (github.com/PathwayCommons/factoid); containerized software components are freely available on DockerHub (hub.docker.com/r/pathwaycommons/factoid) enabling others to build on and improve the Biofactoid software.

Data availability

All Biofactoid data are available under the Creative Commons CC0 public domain license. To download the data and code, please refer to the documentation on the Biofactoid GitHub repository (github.com/PathwayCommons/factoid). More information on software availability is available in Materials and methods.

References

  1. Book
    1. Norman DA
    (2001)
    The Design of Everyday Things
    MIT Press.

Decision letter

  1. Helena Pérez Valle
    Reviewing Editor; eLife, United Kingdom
  2. Peter Rodgers
    Senior Editor; eLife, United Kingdom

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Decision letter after peer review:

Thank you for submitting your article "Capturing scientific knowledge in computable form" to eLife for consideration as a Feature Article. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by two members of the eLife Features Team (Helena Pérez Valle and Peter Rodgers).

The reviewers and editors have discussed the reviews and we have drafted this decision letter to help you prepare a revised submission.

Summary:

The manuscript of Wong et al., introduces Biofactoid, a novel and intuitive tool to collect interaction data from publications. The manuscript text is clear and provides both a general background for non-specialists as well as it contains the key technical details for experts. The figures nicely illustrate Biofactoid and its application. The authors not only developed the tool but also carried out multiple tests and pilots with users, editors and invited authors. All these phases are properly documented and presented in the manuscript. However, a number of points need to be addressed to make the article suitable for publication.

Essential revisions:

1. Please revise the title so that it more precisely describes the current capabilities of Biofactoid and the use cases demonstrated.

2. The amount of information collected for an interaction (that is, after the drawing just the type of molecular event) is not enough to really utilize the power of this approach in other projects. In particular not having an option to add some kind of biological context and detection methods is a missed opportunity. Please consider introducing an optional box for the biological context, in particular to name the tissue/cell line or the condition (healthy, cancer, etc.). This could come very useful when there are conflicting entries (A activates B; A inhibits B). With the biological context it could be clear that both are correct, otherwise it would look like the same interactors but different signs for the interaction and no further information. As for the type of detection, integrating the Molecular Interactions Controlled Vocabulary (https://www.ebi.ac.uk/ols/ontologies/mi) and facilitating data entry with autocomplete functions would provide useful extra information without overly increasing the time it takes to make a submission. Please consider updating Biofactoid to include this functionality. Please also clarify what ontology (if any) is used for relationships in Biofactoid.

3. Please highlight the fact that submissions to Biofactoid can be edited by users later on using the link sent in the original email. This may be useful if users make a mistake during the submission process or if they need to update the diagram during the publication process.

4. Please address whether there is a process for removing a submission altogether (e.g. if a submission is found to be incorrect or if a paper is being retracted).

5. The two sub-figures of Figure 2B are not easily reproducible. With https://biofactoid.org/ as an entry point, it was not possible to get any networks spanning more than the small ones extracted from a single article as shown in Figure 2A. With pathwaycommons.org as an entry point, it was not possible to find the interactions given in the figure. Concretely, on https://www.pathwaycommons.org/, trying "senp1 sirt3" and clicking on "Interactions", the result was "No interactions to display". Please provide more detailed instructions so the figures shown can be reproduced, or modify Figure 2B to reflect a network that can be obtained using Biofactoid.

6. Please include a discussion of how the facts stated in the paper, regarding the use of Biofactoid, could be qualitatively or quantitatively assessed. Currently, an author working with Biofactoid at all is taken as a successful end point, but surely it will be important to verify that the tool works as described and is, therefore, useful to the community. An assessment could be done, for example, in comparison to professional curators' work, as is often done for text mining approaches.

7. Please include the overall number of recorded facts generated over the multi-step process, including the number of papers used and the facts per paper.

8. Please temper your statements regarding the search and exploration functions of the system. Currently biofactoid.org seems to have no query interface. Since the data is not yet in PathwayCommons, interactive access is effectively only by browsing the papers shown in the landing page, but the figures and text imply much more functionality than is actually available to the user at this point.

9. Please include a comparison to related approaches, considering the citations referenced in line 235 of the manuscript.

10. Please state under what license contributors to Biofactoid are contributing their data, and make it clear both in the article and also in the website, so that users know how they can (re)use the data.

11. Please provide each of the Figures in your manuscript as a separate figure file.

https://doi.org/10.7554/eLife.68292.sa1

Author response

Essential revisions:

1. Please revise the title so that it more precisely describes the current capabilities of Biofactoid and the use cases demonstrated.

Our title has been amended to emphasize the role of authors in pathway curation:

“Author-sourced capture of pathway knowledge in computable form using Biofactoid”

2. The amount of information collected for an interaction (that is, after the drawing just the type of molecular event) is not enough to really utilize the power of this approach in other projects. In particular not having an option to add some kind of biological context and detection methods is a missed opportunity. Please consider introducing an optional box for the biological context, in particular to name the tissue/cell line or the condition (healthy, cancer, etc.). This could come very useful when there are conflicting entries (A activates B; A inhibits B). With the biological context it could be clear that both are correct, otherwise it would look like the same interactors but different signs for the interaction and no further information.

We have updated the Biofactoid curation tool to include a context input box that encourages authors to “List terms that describe the context (e.g. T cell, cancer, genome stability)”. Allowing authors to provide a list in free text form has the benefit of accommodating any terms or concepts they consider important for specifying the research context. This information is presented verbatim alongside the network in the Biofactoid Explorer and included in the underlying BioPAX ontology representation, which is also available in Biofactoid’s export files. As a future direction, we will explore ways, such as using natural language processing, to better structure this contextual information without requiring the user to learn the details of complex controlled vocabularies. Users can try the context input box through the Biofactoid curation tool demonstration website at https://biofactoid.org/demo.

As for the type of detection, integrating the Molecular Interactions Controlled Vocabulary (https://www.ebi.ac.uk/ols/ontologies/mi) and facilitating data entry with autocomplete functions would provide useful extra information without overly increasing the time it takes to make a submission. Please consider updating Biofactoid to include this functionality.

Biofactoid aims to capture biological pathway knowledge supported by evidence described in a research article. In this case, authors are free to describe an interaction that represents the findings of one experiment or a set of different experiments. Because entering this information using controlled vocabularies like MI could be complex, even for a trained database curator, we prefer to let authors work with easy to enter plain text data entry in this case (e.g. in the context field). In the future, we will consider ways to help authors annotate the experimental evidence supporting their interactions, for instance by listing the relevant MI terms or by using natural language processing, while still maintaining an easy to use curation interface.

Please also clarify what ontology (if any) is used for relationships in Biofactoid.

Relationships in Biofactoid are mapped to the standard BioPAX ontology that our team developed and that is in use by most major pathway database groups. BioPAX has controlled vocabulary terms for major interaction types (e.g. molecular interaction, catalysis), and extends this to use more specific types by reusing the standard molecular interaction (MI) controlled vocabulary “interaction type” subtree (e.g. “phosphorylation reaction”). We use these in Biofactoid.

3. Please highlight the fact that submissions to Biofactoid can be edited by users later on using the link sent in the original email. This may be useful if users make a mistake during the submission process or if they need to update the diagram during the publication process.

The concluding paragraph of the “Data sharing workflow” section has been updated accordingly:

“Submitted pathways are shared publicly with the research community and users retain the ability to edit their pathway via the link they received at the start of the workflow.”

4. Please address whether there is a process for removing a submission altogether (e.g. if a submission is found to be incorrect or if a paper is being retracted).

The final sentence of the “Data sharing workflow” section has been updated accordingly:

“Authors who wish to have their submissions removed from Biofactoid can do so by contacting the support team (support@biofactoid.org).”

Also, the Biofactoid Explorer web app has been updated so that if PubMed identifies an article as retracted, then users will see a “Retracted Publication” message for that article’s record in Biofactoid.

5. The two sub-figures of Figure 2B are not easily reproducible. With https://biofactoid.org/ as an entry point, it was not possible to get any networks spanning more than the small ones extracted from a single article as shown in Figure 2A. With pathwaycommons.org as an entry point, it was not possible to find the interactions given in the figure. Concretely, on https://www.pathwaycommons.org/, trying "senp1 sirt3" and clicking on "Interactions", the result was "No interactions to display". Please provide more detailed instructions so the figures shown can be reproduced, or modify Figure 2B to reflect a network that can be obtained using Biofactoid.

The results in Figure 2B are intended to demonstrate the utility of information added to Biofactoid. In particular, we show that interactions contributed by two different authors could be used to bridge otherwise disjoint mitochondrial pathways in Pathway Commons.

To create Figure 2B, we manually searched public data in Pathway Commons and Biofactoid with genes of interest and loaded them into the free Cytoscape software for network visualization (cytoscape.org). As a specific example, we executed the following steps to generate the first network in Figure 2B:

1. Visit pathwaycommons.org and search for genes that are part of interactions in Figure 2A: SENP1, SIRT3, UCP1, SIRT5

2. In the search results, listed under “Pathways”, select the Reactome pathways entitled “SUMO is proteolytically processed”, “Transcriptional activation of mitochondrial biogenesis” and “The fatty acid cycling model”. Download a Simple Interaction Format (SIF) file for each. This downloads all of the Reactome pathway information for these genes.

3. Also in the search results, select “Interactions” and download it in SIF. This downloads interactions from many databases involving these genes.

4. Using Cytoscape, import each SIF file as a new network then merge the individual networks using Cytoscape’s merge tool. Manually add the two Biofactoid interactions in Figure 2A to the merged network within Cytoscape.

The description of how these figures were created has been added to the section "Visualization of network data across sources" in Materials and methods.

In the future, we would like to develop additional features for Biofactoid or Cytoscape such that the steps used to create the networks in Figure 2 could be more easily reproduced using automated tools. Specifically, we are working to integrate Biofactoid data continually in Pathway Commons to make querying across all publicly available pathway data easier.

6. Please include a discussion of how the facts stated in the paper, regarding the use of Biofactoid, could be qualitatively or quantitatively assessed. Currently, an author working with Biofactoid at all is taken as a successful end point, but surely it will be important to verify that the tool works as described and is, therefore, useful to the community. An assessment could be done, for example, in comparison to professional curators' work, as is often done for text mining approaches.

In the Discussion (paragraph 6) we have described how Biofactoid currently supports authors in high-quality self-curation, followed by a proposal for efficiently monitoring and improving data quality. Briefly, we will aim to:

  • Monitor data accuracy using trained curators:

– Recruit at least two curators experienced in collecting pathway knowledge (e.g. Reactome curators)

– Curators independently rate a sample of records with regards to accuracy of entity grounding, interactions, as well as record completeness

– The curator’s ratings are compared to each other to estimate variability

  • Support community-based data accuracy monitoring:

– Encourage researchers with questions or concerns about a pathway in Biofactoid to communicate directly with the author(s) of the article to resolve issues and, if necessary, the authors can edit the pathway to fix the issues

– Enable researchers to directly report an issue to the Biofactoid support team

– Reserve the right for the Biofactoid support team to lock pathway editing, flag it as disputed or remove it from the collection

7. Please include the overall number of recorded facts generated over the multi-step process, including the number of papers used and the facts per paper.

At the time of writing, Biofactoid contains data for 73 articles in which a total of 330 entities (263 unique) from 7 organisms (H. sapiens, M. musculus, D. rerio, S. cervisiae, D. melanogaster, C. elegans, E. coli) are involved in 216 interactions (~3 interactions/article). This information has been included in the last paragraph of the Introduction.

8. Please temper your statements regarding the search and exploration functions of the system. Currently biofactoid.org seems to have no query interface. Since the data is not yet in PathwayCommons, interactive access is effectively only by browsing the papers shown in the landing page, but the figures and text imply much more functionality than is actually available to the user at this point.

We have altered statements from our initial submission that more precisely reflect the existing capabilities of Biofactoid.

– Abstract: “The resulting data is interoperable with public information resources, allowing biological facts from different research reports to be unambiguously connected and making it possible for author-curated knowledge to be appreciated in the context of all existing computable knowledge.”

– Data sharing and exploration: “Beyond the “Explorer”, Biofactoid data represented in the formal BioPAX ontology is interoperable with external pathway and interaction knowledge using technologies for BioPAX processing and analysis, including Paxtools (Demir et al., 2013) and cPath2 (Cerami et al., 2006). This permits author-contributed information to be connected to existing information, and leverage tools for searching, visualizing and analyzing data across different sources.”

– Figure 2

– Legend title: “Biofactoid data is connected to information sources and establishes a bridge between related pathways”.

– Legend: “Biofactoid data establishes a bridge between related pathways described by structured biological knowledge from distinct data sources.”

– Figure 2B, subtitle: “Bridging related pathways.”

9. Please include a comparison to related approaches, considering the citations referenced in line 235 of the manuscript.

A summary and comparison of related approaches is presented in Table 2 which is referenced in the Discussion. This table includes projects that support community curation of biological pathway and interaction knowledge. Given this refined criteria, we have removed references to NDEx (Pratt et al., 2015) and GraphSpace (Bharadwaj et al., 2017), since they are primarily concerned with network information storage and dissemination; removed the reference to Wikidata (Waagmeester et al., 2020) as it only stores metadata for pathways imported from elsewhere (e.g. Reactome, WikiPathways); and removed the reference to INDRA-IPM (Todorov et al., 2019) as it does not involve data storage.

10. Please state under what license contributors to Biofactoid are contributing their data, and make it clear both in the article and also in the website, so that users know how they can (re)use the data.

We have added a "Data availability section to Materials and methods that indicates all contributed data is made freely available without restriction under a Creative Commons CC0 public domain licence. Likewise, the Biofactoid homepage now states that “All Biofactoid data is made freely available to download to the research community under a public domain-equivalent license (Creative Commons CC0 license)” (under section “By researchers, for researchers'').

11. Please provide each of the Figures in your manuscript as a separate figure file.

Each figure is provided as a JPEG file.

https://doi.org/10.7554/eLife.68292.sa2

Article and author information

Author details

  1. Jeffrey V Wong

    Jeffrey V Wong is in The Donnelly Centre, University of Toronto, Toronto, Canada

    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Software, Validation, Visualization, Writing – original draft, Writing – review and editing
    For correspondence
    jeffvin.wong@utoronto.ca
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-8912-5699
  2. Max Franz

    Max Franz is in The Donnelly Centre, University of Toronto, Toronto, Canada

    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Software, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-0169-0480
  3. Metin Can Siper

    Metin Can Siper is in the Computational Biology Program, Oregon Health and Science University, Portland, United States

    Contribution
    Data curation, Software
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-7556-093X
  4. Dylan Fong

    Dylan Fong is in The Donnelly Centre, University of Toronto, Toronto, Canada

    Contribution
    Software
    Competing interests
    No competing interests declared
  5. Funda Durupinar

    Funda Durupinar is in the Computer Science Department, University of Massachusetts Boston, Boston, United States

    Contribution
    Software
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-4915-6642
  6. Christian Dallago

    Christian Dallago is in the Department of Cell Biology and the Department of Systems Biology, Harvard Medical School, Boston, United States and in the Department of Informatics, Technische Universität München, Garching, Germany

    Contribution
    Conceptualization, Methodology, Project administration, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-4650-6181
  7. Augustin Luna

    Augustin Luna is in the Department of Cell Biology, Harvard Medical School; in the Department of Data Sciences, Dana-Farber Cancer Institute; and in the Broad Institute, Massachusetts Institute of Technology and Harvard University, Boston, United States

    Contribution
    Conceptualization, Investigation, Methodology, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-5709-371X
  8. John Giorgi

    John Giorgi is in The Donnelly Centre, University of Toronto, Toronto, Canada

    Contribution
    Software
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-9621-5046
  9. Igor Rodchenkov

    Igor Rodchenkov is in The Donnelly Centre, University of Toronto, Toronto, Canada

    Contribution
    Software
    Competing interests
    No competing interests declared
  10. Özgün Babur

    Özgün Babur is in the Computer Science Department, University of Massachusetts Boston, Boston, United States

    Contribution
    Conceptualization, Methodology, Supervision
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0239-5259
  11. John A Bachman

    John A Bachman is in the Laboratory of Systems Pharmacology, Harvard Medical School, Boston, United States

    Contribution
    Resources
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-6095-2466
  12. Benjamin M Gyori

    Benjamin M Gyori is in the Laboratory of Systems Pharmacology, Harvard Medical School, Boston, United States

    Contribution
    Resources
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-9439-5346
  13. Emek Demir

    Emek Demir is in the Computational Biology Program, Oregon Health and Science University, Portland, United States

    Contribution
    Conceptualization, Funding acquisition, Methodology, Project administration, Supervision, Writing – review and editing
    For correspondence
    demire@ohsu.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-3663-7113
  14. Gary D Bader

    Gary D Bader is in The Donnelly Centre, The Department of Computer Science and the Department of Molecular Genetics, University of Toronto; The Lunenfeld-Tanenbaum Research Institute, Mount Sinai Hospital; and the Princess Margaret Cancer Centre, University Health Network, Toronto, Canada

    Contribution
    Conceptualization, Funding acquisition, Methodology, Project administration, Supervision, Writing – review and editing
    For correspondence
    gary.bader@utoronto.ca
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-0185-8861
  15. Chris Sander

    Chris Sander is in the Department of Cell Biology, Harvard Medical School; in the Department of Data Sciences, Dana-Farber Cancer Institute; and in the Broad Institute, Massachusetts Institute of Technology, Harvard University, Boston, United States

    Contribution
    Conceptualization, Funding acquisition, Methodology, Project administration, Supervision, Writing – review and editing
    For correspondence
    sander.research@gmail.com
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-6059-6270

Funding

National Human Genome Research Institute (U41 HG006623)

  • Jeffrey V Wong
  • Max Franz
  • Metin Can Siper
  • Dylan Fong
  • Funda Durupinar
  • Christian Dallago
  • Augustin Luna
  • John M Giorgi
  • Igor Rodchenkov
  • Özgün Babur
  • Emek Demir
  • Gary D Bader
  • Chris Sander

National Human Genome Research Institute (U41 HG003751)

  • Jeffrey V Wong
  • Max Franz
  • Metin Can Siper
  • Dylan Fong
  • Funda Durupinar
  • Christian Dallago
  • Augustin Luna
  • John M Giorgi
  • Igor Rodchenkov
  • Özgün Babur
  • Emek Demir
  • Gary D Bader
  • Chris Sander

National Human Genome Research Institute (R01 HG009979)

  • Max Franz
  • Gary D Bader

National Institute of General Medical Sciences (P41 GM103504)

  • Jeffrey V Wong
  • Max Franz
  • Metin Can Siper
  • Dylan Fong
  • Funda Durupinar
  • Christian Dallago
  • Augustin Luna
  • John M Giorgi
  • Igor Rodchenkov
  • Özgün Babur
  • Emek Demir
  • Gary D Bader
  • Chris Sander

Defense Advanced Research Projects Agency (Big Mechanism)

  • Metin Can Siper
  • Funda Durupinar
  • Özgün Babur
  • John A Bachman
  • Benjamin Gyori
  • Emek Demir

Defense Advanced Research Projects Agency (Communicating with Computers)

  • Metin Can Siper
  • Funda Durupinar
  • Özgün Babur
  • John A Bachman
  • Benjamin Gyori
  • Emek Demir

Defense Advanced Research Projects Agency (ARO W911NF-14-C-0119)

  • Metin Can Siper
  • Funda Durupinar
  • Özgün Babur
  • John A Bachman
  • Benjamin M Gyori
  • Emek Demir

Defense Advanced Research Projects Agency (ARO W911NF-15-1-054)

  • Metin Can Siper
  • Funda Durupinar
  • Özgün Babur
  • John A Bachman
  • Benjamin M Gyori
  • Emek Demir

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank Quincey Justman, Miao-Chih Tsai, and Anita DeWaard for feedback on making Biofactoid useful for editors and authors; the Reactome database team for support and feedback on the curation workflow; Alfonso Valencia, and Miguel Vazquez for early support; and the many community members in Toronto, Boston, Portland, and beyond for feedback on the design and concept of Biofactoid.

Ethics

Human subjects: Participants of user testing provided written consent to volunteer, have their testing sessions recorded and have quotes obtained in the session published.

Publication history

  1. Preprint posted: March 11, 2021 (view preprint)
  2. Received: March 11, 2021
  3. Accepted: December 2, 2021
  4. Accepted Manuscript published: December 3, 2021 (version 1)
  5. Version of Record published: December 17, 2021 (version 2)

Copyright

© 2021, Wong et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,606
    Page views
  • 93
    Downloads
  • 8
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Jeffrey V Wong
  2. Max Franz
  3. Metin Can Siper
  4. Dylan Fong
  5. Funda Durupinar
  6. Christian Dallago
  7. Augustin Luna
  8. John Giorgi
  9. Igor Rodchenkov
  10. Özgün Babur
  11. John A Bachman
  12. Benjamin M Gyori
  13. Emek Demir
  14. Gary D Bader
  15. Chris Sander
(2021)
Science Forum: Author-sourced capture of pathway knowledge in computable form using Biofactoid
eLife 10:e68292.
https://doi.org/10.7554/eLife.68292
  1. Further reading

Further reading

    1. Cancer Biology
    2. Computational and Systems Biology
    Bingrui Li, Fernanda G Kugeratski, Raghu Kalluri
    Research Article

    Non-invasive early cancer diagnosis remains challenging due to the low sensitivity and specificity of current diagnostic approaches. Exosomes are membrane-bound nanovesicles secreted by all cells that contain DNA, RNA, and proteins that are representative of the parent cells. This property, along with the abundance of exosomes in biological fluids makes them compelling candidates as biomarkers. However, a rapid and flexible exosome-based diagnostic method to distinguish human cancers across cancer types in diverse biological fluids is yet to be defined. Here, we describe a novel machine learning-based computational method to distinguish cancers using a panel of proteins associated with exosomes. Employing datasets of exosome proteins from human cell lines, tissue, plasma, serum, and urine samples from a variety of cancers, we identify Clathrin Heavy Chain (CLTC), Ezrin, (EZR), Talin-1 (TLN1), Adenylyl cyclase-associated protein 1 (CAP1), and Moesin (MSN) as highly abundant universal biomarkers for exosomes and define three panels of pan-cancer exosome proteins that distinguish cancer exosomes from other exosomes and aid in classifying cancer subtypes employing random forest models. All the models using proteins from plasma, serum, or urine-derived exosomes yield AUROC scores higher than 0.91 and demonstrate superior performance compared to Support Vector Machine, K Nearest Neighbor Classifier and Gaussian Naive Bayes. This study provides a reliable protein biomarker signature associated with cancer exosomes with scalable machine learning capability for a sensitive and specific non-invasive method of cancer diagnosis.

    1. Computational and Systems Biology
    2. Immunology and Inflammation
    Alain Pulfer, Diego Ulisse Pizzagalli ... Santiago Fernandez Gonzalez
    Tools and Resources

    Intravital microscopy has revolutionized live-cell imaging by allowing the study of spatial–temporal cell dynamics in living animals. However, the complexity of the data generated by this technology has limited the development of effective computational tools to identify and quantify cell processes. Amongst them, apoptosis is a crucial form of regulated cell death involved in tissue homeostasis and host defense. Live-cell imaging enabled the study of apoptosis at the cellular level, enhancing our understanding of its spatial–temporal regulation. However, at present, no computational method can deliver robust detection of apoptosis in microscopy timelapses. To overcome this limitation, we developed ADeS, a deep learning-based apoptosis detection system that employs the principle of activity recognition. We trained ADeS on extensive datasets containing more than 10,000 apoptotic instances collected both in vitro and in vivo, achieving a classification accuracy above 98% and outperforming state-of-the-art solutions. ADeS is the first method capable of detecting the location and duration of multiple apoptotic events in full microscopy timelapses, surpassing human performance in the same task. We demonstrated the effectiveness and robustness of ADeS across various imaging modalities, cell types, and staining techniques. Finally, we employed ADeS to quantify cell survival in vitro and tissue damage in mice, demonstrating its potential application in toxicity assays, treatment evaluation, and inflammatory dynamics. Our findings suggest that ADeS is a valuable tool for the accurate detection and quantification of apoptosis in live-cell imaging and, in particular, intravital microscopy data, providing insights into the complex spatial–temporal regulation of this process.