ABSTRACT
In the last decade, a large number of knowledge graph (KG) completion approaches were proposed. Albeit effective, these efforts are disjoint, and their collective strengths and weaknesses in effective KG completion have not been studied in the literature. We extend Plumber, a framework that brings together the research community's disjoint efforts on KG completion. We include more components into the architecture of Plumber to comprise 40 reusable components for various KG completion subtasks, such as coreference resolution, entity linking, and relation extraction. Using these components, Plumber dynamically generates suitable knowledge extraction pipelines and offers overall 432 distinct pipelines. We study the optimization problem of choosing optimal pipelines based on input sentences. To do so, we train a transformer-based classification model that extracts contextual embeddings from the input and finds an appropriate pipeline. We study the efficacy of Plumber for extracting the KG triples using standard datasets over three KGs: DBpedia, Wikidata, and Open Research Knowledge Graph. Our results demonstrate the effectiveness of Plumber in dynamically generating KG completion pipelines, outperforming all baselines agnostic of the underlying KG. Furthermore, we provide an analysis of collective failure cases, study the similarities and synergies among integrated components and discuss their limitations.
ABSTRACT
BACKGROUND: New technologies for analyzing biological samples, like next generation sequencing, are producing a growing amount of data together with quality scores. Moreover, software tools (e.g., for mapping sequence reads), calculating transcription factor binding probabilities, estimating epigenetic modification enriched regions or determining single nucleotide polymorphism increase this amount of position-specific DNA-related data even further. Hence, requesting data becomes challenging and expensive and is often implemented using specialised hardware. In addition, picking specific data as fast as possible becomes increasingly important in many fields of science. The general problem of handling big data sets was addressed by developing specialized databases like HBase, HyperTable or Cassandra. However, these database solutions require also specialized or distributed hardware leading to expensive investments. To the best of our knowledge, there is no database capable of (i) storing billions of position-specific DNA-related records, (ii) performing fast and resource saving requests, and (iii) running on a single standard computer hardware. RESULTS: Here, we present DRUMS (Disk Repository with Update Management and Select option), satisfying demands (i)-(iii). It tackles the weaknesses of traditional databases while handling position-specific DNA-related data in an efficient manner. DRUMS is capable of storing up to billions of records. Moreover, it focuses on optimizing relating single lookups as range request, which are needed permanently for computations in bioinformatics. To validate the power of DRUMS, we compare it to the widely used MySQL database. The test setting considers two biological data sets. We use standard desktop hardware as test environment. CONCLUSIONS: DRUMS outperforms MySQL in writing and reading records by a factor of two up to a factor of 10000. Furthermore, it can work with significantly larger data sets. Our work focuses on mid-sized data sets up to several billion records without requiring cluster technology. Storing position-specific data is a general problem and the concept we present here is a generalized approach. Hence, it can be easily applied to other fields of bioinformatics.
