Vaccine Semantics

Post-marketing management and decision-making about vaccines builds on the early detection of safety concerns and changes in public sentiment, the accurate access to established evidence, and the ability to promptly quantify effects and verify hypotheses about the vaccine benefits and risks. A variety of resources provide relevant information but they use different representations, which makes rapid evidence generation and extraction challenging. This thesis presents automatic methods for interpreting heterogeneously represented vaccine information. Part I evaluates social media messages for monitoring vaccine adverse events and public sentiment in social media messages, using automatic methods for information recognition. Parts II and III develop and evaluate automatic methods and resources for the recognition, representation, and reasoning about established vaccine-related information in scientific literature and extracting information from medical health record databases. Additionally, two user applications, CodeMapper and VaccO, are introduced to accellerate the implementation of collaborative observational studies about vaccines.

Here is an overview on the different projects in the thesis, the investigated resources, and applied approaches. Click on the grey boxes or follow the links below for further information.

• Social media: Can public messages on social medial be used in vaccine surveillance?
• CodeMapper: Harmonizing mapping of medical case definitions to clinical coding systems using the UMLS
• Religator: Identifying statements of relations between drugs and disorders in scientific literature
• VaccO: Ontology of vaccine properties for the alignment of vaccine codes and the identification of vaccine descriptions in scientific literature

• CodeMapper: Semi-automatic coding of case definitions
  https://app.vac4eu.org/codemapper
• VaccO alignment: Analyse, select, and align vaccine code descriptors
  https://app.vac4eu.org/vacco

• Identifying and normalizing vaccine descriptions in scientific literature: A comparison between ontology-based and machine learning approaches.
  Benedikt Becker, Helen Ying He, Miriam Sturkenboom, Jan Kors.
  Submitted, 2018.
• Alignment of vaccine codes using the VaccO ontology of vaccine descriptions.
  Benedikt Becker, Jan Kors, Erik Mulligen, Miriam Sturkenboom.
  Submitted, 2018.
• CodeMapper: semi-automatic coding of medical case definitions.
  Benedikt Becker, Paul Avillach, Silvana Romio, Erik Mulligen, Daniel Weibel, Miriam Sturkenboom, Jan Kors.
  Pharmacoepidemiology and drug safety, 2017 (link)
• Extraction of chemical-induced diseases using prior knowledge and textual information.
  Ewoud Pons + Benedikt Becker, Saber Akhondi, Zubair Afzal, Erik van Mulligen, Jan Kors.
  Database, 2016 (link)
• Evaluation of a multinational, multilingual vaccine debate on Twitter.
  Benedikt Becker, Heidi Larson, Jan Bonhoeffer, Erik Van Mulligen, Jan Kors, Miriam Sturkenboom.
  Vaccine, 2016. (link)
• Evaluating social media networks in medicines safety surveillance: two case studies.
  Preciosa Coloma, Benedikt Becker, Miriam Sturkenboom, Erik van Mulligen, Jan Kors.
  Drug safety, 2015. (link)

The work was carried out in Biosemantics working group and largely funded and developed in the context of the ADVANCE project.