PheneBank: automatic extraction and validation of a database of human phenotype-disease associations in the scientific literature

Free text scientific literature has the potential to be an incredibly valuable source of data for uncovering the often hidden relationships between genes, diseases and phenotypes. Phenotypic descriptions cover abnormalities in anatomical structures, processes and behaviours. For example 'growth delay' and 'body weight loss'. Such descriptions form the basis for determining the existence and treatment of a disease but, because of their inherent complexity, have previously received less attention by the text mining community.
In recent years, significant effort has been spent by a small number of expert curators to create coding systems for phenotypes (called "ontologies"), such as the Human Phenotype Ontology (HP) [1] and the Mammalian Phenotype Ontology (MP). The PheneBank project proposes to support and speed up curation using terms discovered directly from the literature and to automatically integrate them with such standard ontologies
There are three major challenges we seek to address: (1) knowledge brokering: to develop state of the art text mining approaches to identify phenotypic descriptions in scientific texts; (2) knowledge management: to create a structured resource of phenotype terms used in scientific texts and link them to existing coding systems; and (3) adding insight to evidence: to work with domain experts to utilize statistical association algorithms to identify meaningful phenotype-disease / phenotype-gene profiles. The disease profiles will be evaluated against hand curated standards in human disease databases (e.g. Online Mendelian Inheritance of Man and OrphaNet) with a focus on rare diseases. Mined data will be provided in a machine understandable database - a definitive output of the project - to support clinicians and scientists.
At the technological level the project will pioneer new methods for text mining that exploit machine learning (ML). Scientific texts remain a challenging area for a variety of reasons: descriptive naming, high levels of ambiguity/out of vocabulary words, use of complex sentence structures and an evolving vocabulary. Current techniques in term recognition employ ML in pipelines to search for continuous sequences of words that represent genes, proteins and cells etc. State of the art models include conditional random fields using feature sets based on dictionaries as well as the local and topical context where the term is located. However, phenotype descriptions are often represented by discontinuous sequences, such as 'growth in the patient was delayed'. One key aspect not previously addressed is in the capture of such non-canonical terms. This requires a different paradigm based on grammatical parsing algorithms that capture structural relations as well as joint learning techniques that can leverage large numbers of features simultaneously and optimise these across the diverse contexts in which phenotypes are mentioned.
The project also seeks to harness texts for extracting statistically significant associations between phenotypes, diseases and genes. Earlier approaches have suffered from not providing deep semantic descriptions of the relations they tried to target. This means that association scores merge notions of genetic, pharmacological, and epidemiological relations etc. without distinction. Our parsing-based approach is an attempt to overcome this issue by discovering more precise relationships. The approach follows ground breaking work at the Wellcome Trust Sanger Institute (WTSI), including terminology alignment of phenotypes using pairwise scoring of the conceptual elements that make up the phenotype.
An exciting aspect of this project is inter-disciplinary collaboration across stakeholders to build a resource of phenotype-disease profiles: (a) computer scientists from the Universities of Cambridge, Colorado and Manchester; (b) bioinformaticians and life scientists from the WTSI, McGill University and EMBL-EBI, and (c) clinicians from the NIHR Bioresource.

PheneBank will exploit state of the art text mining (TM) together with existing ontological resources to collect fragmented biological results about the phenotypic profiles of human diseases and integrate this into a machine understandable semantic database and tool set for use by the clinical and scientific communities in their workflows. TM will be used to automatically extract phenotype, disease and gene terms from the full text scientific literature and harmonise these to ontological resources. Importantly, we will exploit grammatical parsing (the BLLIP parser) for recovering discontinuous terms and then a de-compositional approach in which the elements of the complex phenotype terms are individually linked to ontologies (e.g. Gene Ontology: processes, UBERON for structures). Harmonisation will explore learn-to-rank techniques to optimize concept selection from an array of tools such as cTAKES/MetaMap. Following this, typed relations between terms will be filtered using a range of machine learning classifiers on grammatical parse trees. Domain adaptation techniques will be tested for both term and relation extraction. Validation will follow standard protocols for text mining, including construction of a gold standard annotated corpus of 5000 sentences sampled from cited literature with a focus on rare human diseases. Phenotypic profiles will be built from strongly associated and related terms (e.g. phenotype-disease) by exploring a range of measures (e.g. Jaccard index, Information Content). Validation will use existing expert curated profiles in the OMIM and OrphaNet databases. Additionally we will explore the cross-species harmonisation of the human phenotype terms to closely associated mouse phenotypes. This will take place through conceptual mappings of PheneBank terms to the HPO and MP. The semantic database will include phenotypes and associated links for querying, navigation and download in a variety of formats (OWN/RDF/JSON).

The PheneBank project aims to revolutionise how health experts leverage phenotypic evidence from the literature for their own clinical and scientific studies. This will be done using automatic text mining (TM) techniques that encode the scientific literature according to a machine understandable semantic representation. PheneBank goes beyond traditional TM by discovering structured associations across the literature as well as cross-species mappings between human and mouse phenotypes. This is highly relevant to a range of experts across lifescience domains. Semantic integration of phenotypes between the scientific literature and ontological coding systems brings us onto a path for full integration with electronic patient records.
Who will benefit from this research?
1. Life scientists and clinicians from a variety of disciplines will benefit from a novel database of evidence about phenotype associations. They will be able to more accurately, thoroughly and efficiently access phenotypic evidence from the scientific literature for use in their own workflows.
2. Bioinformaticians and database curators involved in knowledge discovery and data integration will benefit from data and tools that they can incorporate into their own workflows. This is highly relevant to knowledge integration efforts in initiatives such as ELIXIR.
3. Researchers and engineers in human language technologies, e-Science and information retrieval will benefit from new techniques, tools and data.
How will they benefit?
1. PheneBank will be of particular benefit to life scientists and clinicians investigating rare diseases through our partnership with Prof. Willem Ouwehand (NIHR Bioresource for Rare Diseases) and advisory group member Prof. Paul Lasko (McGill University and Chair of the International Rare Disease Research Consortium). In particular we expect that mined phenotype profiles will contribute to the understanding, diagnosis and treatment of rare diseases by bringing together previously hidden evidence from across the literature using term harmonisation techniques. Building on the project partner's network, we will engage directly with the rare disease community, providing a pathway to deploying the tools and data.
2. There will be multiple opportunities throughout the project to engage with bioinformaticians and database curators. To take just three examples: (1) Partnership with Dr. Peter Robinson leader of the Human Phenotype Ontology will allow data dissemination to one of the most widely used coding systems and engagement with his rare disease partner network. PheneBank outputs will supplement scarce human expertise using high throughput text processing, bridging traditional disciplines and be of a measurable quality. (2) Collaboration with Dr. Skarnes at the Wellcome Trust Sanger Institute (WTSI) will facilitate collaboration with UK groups involved in the informatics of rare diseases such as the Deciphering Developmental Disorders project. (3) Working with Dr. Jo McEntyre at Europe PMCprovides an opportunity to make our annotations available to the scientific research and database curation community,engage with curators and industrial groups (e.g. through the EMBL-EBI Industry Programme).
3. The PheneBank project pioneers new methods for Natural Language Processing (NLP) and Machine Learning (ML) on scientific literature. We propose to develop a novel combination ML approaches on maximally rich NLP features in order to understand the meaning of disjoint scientific terms, harmonise them to clinical ontology standards and explore a range of association measures against human curation standards. Researchers and engineers will benefit from tools, novel data sets (e.g. CRAFT and the Europe PMC corpus) and techniques released through publication.
A stakeholder workshop will be organised in the second year of the project at a major biomedical informatics conference such as ISMB/ECCB to raise awareness of the project and invite feedback.