Tag: Big Data

Pensoft Annotator – a tool for text annotation with ontologies

By Mariya Dimitrova, Georgi Zhelezov, Teodor Georgiev and Lyubomir Penev

The use of written language to record new knowledge is one of the advancements of civilisation that has helped us achieve progress. However, in the era of Big Data, the amount of published writing greatly exceeds the physical ability of humans to read and understand all written information. 

More than ever, we need computers to help us process and manage written knowledge. Unlike humans, computers are “naturally fluent” in many languages, such as the formats of the Semantic Web. These standards were developed by the World Wide Web Consortium (W3C) to enable computers to understand data published on the Internet. As a result, computers can index web content and gather data and metadata about web resources.

To help manage knowledge in different domains, humans have started to develop ontologies: shared conceptualisations of real-world objects, phenomena and abstract concepts, expressed in machine-readable formats. Such ontologies can provide computers with the necessary basic knowledge, or axioms, to help them understand the definitions and relations between resources on the Web. Ontologies outline data concepts, each with its own unique identifier, definition and human-legible label.

Matching data to its underlying ontological model is called ontology population and involves data handling and parsing that gives it additional context and semantics (meaning). Over the past couple of years, Pensoft has been working on an ontology population tool, the Pensoft Annotator, which matches free text to ontological terms.

The Pensoft Annotator is a web application, which allows annotation of text input by the user, with any of the available ontologies. Currently, they are the Environment Ontology (ENVO) and the Relation Ontology (RO), but we plan to upload many more. The Annotator can be run with multiple ontologies, and will return a table of matched ontological term identifiers, their labels, as well as the ontology from which they originate (Fig. 1). The results can also be downloaded as a Tab-Separated Value (TSV) file and certain records can be removed from the table of results, if desired. In addition, the Pensoft Annotator allows to exclude certain words (“stopwords”) from the free text matching algorithm. There is a list of default stopwords, common for the English language, such as prepositions and pronouns, but anyone can add new stopwords.

Figure 1. Interface of the Pensoft Annotator application

In Figure 1, we have annotated a sentence with the Pensoft Annotator, which yields a single matched term, labeled ‘host of’, from the Relation Ontology (RO). The ontology term identifier is linked to a webpage in Ontobee, which points to additional metadata about the ontology term (Fig. 2).

Figure 2. Web page about ontology term

Such annotation requests can be run to perform text analyses for topic modelling to discover texts which contain host-pathogen interactions. Topic modelling is used to build algorithms for content recommendation (recommender systems) which can be implemented in online news platforms, streaming services, shopping websites and others.

At Pensoft, we use the Pensoft Annotator to enrich biodiversity publications with semantics. We are currently annotating taxonomic treatments with a custom-made ontology based on the Relation Ontology (RO) to discover treatments potentially describing species interactions. You can read more about using the Annotator to detect biotic interactions in this abstract.

The Pensoft Annotator can also be used programmatically through an API, allowing you to integrate the Annotator into your own script. For more information about using the Pensoft Annotator, please check out the documentation.

How the names of organisms help to turn ‘small data’ into ‘Big Data’

Innovation in ‘Big Data’ helps address problems that were previously overwhelming. What we know about organisms is in hundreds of millions of pages published over 250 years. New software tools of the Global Names project find scientific names, index digital documents quickly, correcting names and updating them. These advances help “Making small data big” by linking together to content of many research efforts. The study was published in the open access journal Biodiversity Data Journal.

The ‘Big Data’ vision of science is transformed by computing resources to capture, manage, and interrogate the deluge of information coming from new technologies, infrastructural projects to digitise physical resources (such as our literature from the Biodiversity Heritage Library), or digital versions of specimens and records about specimens by museums.

Increased bandwidth has made dialogue among distributed data centres feasible and this is how new insights into biology are arising. In the case of biodiversity sciences, data centres range in size from the large GenBank for molecular records and the Global Biodiversity Information Facility for records of occurrences of species, to a long tail of tens of thousands of smaller datasets and web-sites which carry information compiled by individuals, research projects, funding agencies, local, state, national and international governmental agencies.

The large biological repositories do not yet approach the scale of astronomy and nuclear physics, but the very large number of sources in the long tail of useful resources do present biodiversity informaticians with a major challenge – how to discover, index, organize and interconnect the information contained in a very large number of locations.

In this regard, biology is fortunate that, from the middle of the 18th Century, the community has accepted the use of latin binomials such as Homo sapiens or Ba humbugi for species. All names are listed by taxonomists. Name recognition tools can call on large expert compilations of names (Catalogue of Life, Zoobank, Index Fungorum, Global Names Index) to find matches in sources of digital information. This allows for the rapid indexing of content.

Even when we do not know a name, we can ‘discover’ it because scientific names have certain distinctive characteristics (written in italics, most often two successive words in a latinised form, with the first one – capitalised). These properties allow names not yet present in compilations of names to be discovered in digital data sources.

The idea of a names-based cyberinfrastructure is to use the names to interconnect large and small distributed sites of expert knowledge distributed across the Internet. This is the concept of the described Global Names project which carried out the work described in this paper.

The effectiveness of such an infrastructure is compromised by the changes to names over time because of taxonomic and phylogenetic research. Names are often misspelled, or there might be errors in the way names are presented. Meanwhile, increasing numbers of species have no names, but are distinguished by their molecular characteristics.

In order to assess the challenge that these problems may present to the realization of a names-based cyberinfrastructure, we compared names from GenBank and DRYAD (a digital data repository) with names from Catalogue of Life to assess how well matched they are.

As a result, we found out that fewer than 15% of the names in pair-wise comparisons of these data sources could be matched. However, with a names parser to break the scientific names into all of their component parts, those parts that present the greatest number of problems could be removed to produce a simplified or canonical version of the name. Thanks to such tools, name-matching was improved to almost 85%, and in some cases to 100%.

The study confirms the potential for the use of names to link distributed data and to make small data big. Nonetheless, it is clear that we need to continue to invest more and better names-management software specially designed to address the problems in the biodiversity sciences.

###

Original source:

Patterson D, Mozzherin D, Shorthouse D, Thessen A (2016) Challenges with using names to link digital biodiversity information. Biodiversity Data Journal, doi: 10.3897/BDJ.4.e8080.

Additional information:

The study was supported by the National Science Foundation.