Patrick Ernst

KnowLife: a versatile approach for constructing a large knowledge graph for biomedical sciences

BackgroundBiomedical knowledge bases (KB’s) have become important assets in life sciences. Prior work on KB construction has three major limitations. First, most biomedical KBs are manually built and curated, and cannot keep up with the rate at which new findings are published. Second, for automatic information extraction (IE), the text genre of choice has been scientific publications, neglecting sources like health portals and online communities. Third, most prior work on IE has focused on the molecular level or chemogenomics only, like protein-protein interactions or gene-drug relationships, or solely address highly specific topics such as drug effects.ResultsWe address these three limitations by a versatile and scalable approach to automatic KB construction. Using a small number of seed facts for distant supervision of pattern-based extraction, we harvest a huge number of facts in an automated manner without requiring any explicit training.We extend previous techniques for pattern-based IE with confidence statistics, and we combine this recall-oriented stage with logical reasoning for consistency constraint checking to achieve high precision. To our knowledge, this is the first method that uses consistency checking for biomedical relations. Our approach can be easily extended to incorporate additional relations and constraints.We ran extensive experiments not only for scientific publications, but also for encyclopedic health portals and online communities, creating different KB’s based on different configurations. We assess the size and quality of each KB, in terms of number of facts and precision. The best configured KB, KnowLife, contains more than 500,000 facts at a precision of 93% for 13 relations covering genes, organs, diseases, symptoms, treatments, as well as environmental and lifestyle risk factors.ConclusionKnowLife is a large knowledge base for health and life sciences, automatically constructed from different Web sources. As a unique feature, KnowLife is harvested from different text genres such as scientific publications, health portals, and online communities. Thus, it has the potential to serve as one-stop portal for a wide range of relations and use cases. To showcase the breadth and usefulness, we make the KnowLife KB accessible through the health portal (http://knowlife.mpi-inf.mpg.de).

KnowLife: A knowledge graph for health and life sciences

Knowledge bases (KBs) contribute to advances in semantic search, Web analytics, and smart recommendations. Their coverage of domain-specific knowledge is limited, though. This demo presents the KnowLife portal, a large KB for health and life sciences, automatically constructed from Web sources. Prior work on biomedical ontologies has focused on molecular biology: genes, proteins, and pathways. In contrast, KnowLife is a one-stop portal for a much wider range of relations about diseases, symptoms, causes, risk factors, drugs, side effects, and more. Moreover, while most prior work relies on manually curated sources as input, the KnowLife system taps into scientific literature as well as online communities. KnowLife uses advanced information extraction methods to populate the relations in the KB. This way, it learns patterns for relations, which are in turn used to semantically annotate newly seen documents, thus aiding users in “speed-reading”. We demonstrate the value of the KnowLife KB by various use-cases, supporting both layman and professional users.

DeepLife: An Entity-aware Search, Analytics and Exploration Platform for Health and Life Sciences

Despite the abundance of biomedical literature and health discussions in online communities, it is often tedious to retrieve informative contents for health-centric information needs. Users can query scholarly work in PubMed by keywords and MeSH terms, and resort to Google for everything else. This demo paper presents the DeepLife system, to overcome the limitations of existing search engines for life science and health topics. DeepLife integrates large knowledge bases and harnesses entity linking methods, to support search and exploration of scientific literature, newspaper feeds, and social media, in terms of keywords and phrases, biomedical entities, and taxonomic categories. It also provides functionality for entityaware text analytics over health-centric contents.

ESPRESSO: Explaining Relationships between Entity Sets

Analyzing and explaining relationships between entities in a knowledge graph is a fundamental problem with many applications. Prior work has been limited to extracting the most informative subgraph connecting two entities of interest. This paper extends and generalizes the state of the art by considering the relationships between two sets of entities given at query time. Our method, coined ESPRESSO, explains the connection between these sets in terms of a small number of relatedness cores: dense sub-graphs that have strong relations with both query sets. The intuition for this model is that the cores correspond to key events in which entities from both sets play a major role. For example, to explain the relationships between US politicians and European politicians, our method identifies events like the PRISM scandal and the Syrian Civil War as relatedness cores. Computing cores of bounded size is NP-hard. This paper presents efficient approximation algorithms. Our experiments with real-life knowledge graphs demonstrate the practical viability of our approach and, through user studies, the superior output quality compared to state-of-the-art baselines.

Mobile Radiology Interaction and Decision Support Systems of the Future

Clinical care and research increasingly rely on digitized patient information. There is a growing need to store and organize all patient data, including health records, laboratory reports, and medical images. Medical images have become indispensable for detecting and differentiating pathologies, planning interventions, and monitoring treatments. The effective retrieval of images builds on the semantic annotation of image contents and intelligent interaction with the image material. The semantic annotation of image contents has an automatic and a manual component. In our work, we heavily rely on automatic organ, tissue, and disease detection, which represents one of the main technical research questions in Medico. In this article, however, we will focus on intelligent interaction with the image material, i.e., what mobile radiology interaction and decision support systems of the future, based on automatic detectors, may look like.

A Spatio-anatomical Medical Ontology and Automatic Plausibility Checks

In this paper, we explain the peculiarities of medical knowledge management and propose a way to augment medical domain ontologies by spatial relations in order to perform automatic plausibility checks. Our approach uses medical expert knowledge represented in formal ontologies to check the results of automatic medical object recognition algorithms for spatial plausibility. It is based on the comprehensive Foundation Model of Anatomy ontology which we extend with spatial relations between a number of anatomical entities. These relations are learned inductively from an annotated corpus of 3D volume data sets. The induction process is split into two parts. First, we generate a quantitative anatomical atlas using fuzzy sets to represent inherent imprecision. From this atlas we then abstract the information further onto a purely symbolic level to generate a generic qualitative model of the spatial relations in human anatomy. In our evaluation we describe how this model can be used to check the results of a state-of-the-art medical object recognition system for 3D CT volume data sets for spatial plausibility. Our results show that the combination of medical domain knowledge in formal ontologies and sub-symbolic object recognition yields improved overall recognition precision.

Modeling the International Classification of Diseases (ICD-10) in OWL

Current efforts in healthcare focus on establishing interoperability and data integration of medical resources for better collaboration between medical personal and doctors, especially in the patient treatment process. In covering human diseases, one of the major international standards in clinical practice is the International Classification for Diseases (ICD), maintained by the World Health Organization (WHO). Several country- and language-specific adaptations exist which share the general structure of the WHO version but differ in certain details. This complicates the exchange of patient records and hampers data integration across language borders. We present our approach for modeling the hierarchy of the ICD-10 using the Web Ontology Language (OWL). OWL, which we will introduce shortly, should provide a formal ontological basis for ICD-10 with enough expressivity to model interoperability and data integration of several medical resources such as ICD. Our resulting model captures the hierarchical information of the ICD-10 as well as comprehensive class labels for English and German. Specialities such as “Exclusion” statements, which make statements about the disjointness of certain ICD-10 categories, are modeled in a formal way. For properties which exceed the expressivity of OWL-DL, we provide a separate OWL-Full component which allows us to use the hierarchical knowledge and class labels with existing OWL-DL reasoners and capture the additional information in a Semantic Web format.

Combining patient metadata extraction and automatic image parsing for the generation of an anatomic atlas

We present a system that integrates ontology-based metadata extraction from medical images with a state-of-the-art object recognition algorithm for 3D volume data sets generated by Computed Tomography scanners. Extracted metadata and automatically generated medical image annotations are stored as instances of OWL classes. This system is applied to a corpus of over 750 GB of clinical image data. A spatial database is used to store and retrieve 3D representations of the generated medical image annotations. Our integrated data representation allows us to easily analyze our corpus and to estimate the quality of image metadata. A rule-based system is used to check the plausibility of the output of the automatic object recognition technique against the Foundational Model of Anatomy ontology. All combined, these methods are used to determine an appropriate set of metadata and image features for the automatic generation of a spatial atlas of human anatomy.

HighLife: Higher-arity Fact Harvesting

Text-based knowledge extraction methods for populating knowledge bases have focused on binary facts: relationships between two entities. However, in advanced domains such as health, it is often crucial to consider ternary and higher-arity relations. An example is to capture which drug is used for which disease at which dosage (e.g. 2.5 mg/day) for which kinds of patients (e.g., children vs. adults). In this work, we present an approach to harvest higher-arity facts from textual sources. Our method is distantly supervised by seed facts, and uses the fact-pattern duality principle to gather fact candidates with high recall. For high precision, we devise a constraint-based reasoning method to eliminate false candidates. A major novelty is in coping with the difficulty that higher-arity facts are often expressed only partially in texts and strewn across multiple sources. For example, one sentence may refer to a drug, a disease and a group of patients, whereas another sentence talks about the drug, its dosage and the target group without mentioning the disease. Our methods cope well with such partially observed facts, at both pattern-learning and constraint-reasoning stages. Experiments with health-related documents and with news articles demonstrate the viability of our method.

Disambiguation of Entities in MEDLINE Abstracts by Combining MeSH Terms with Knowledge

Entity disambiguation in the biomedical domain is an essential task in any text mining pipeline. Much existing work shares one limitation, in that their model training prerequisite and/or runtime computation are too expensive to be applied to all ambiguous entities in real-time. We propose an automatic, light-weight method that processes MEDLINE abstracts at largescale and with high-quality output. Our method exploits MeSH terms and knowledge in UMLS to first identify unambiguous anchor entities, and then disambiguate remaining entities via heuristics. Experiments showed that our method is 79.6% and 87.7% accurate under strict and relaxed rating schemes, respectively. When compared to MetaMap’s disambiguation, our method is one order of magnitude faster with a slight advantage in accuracy.