Visualising Inconsistency and Incompleteness in RDF Gene Expression Data using FCA

1. Introduction

In data analyses, there can be a need to identify inconsistent or incomplete data (IID). IID can exist in an integrated dataset, such as in integrated biologists’ experimental result sets. For example, a gene can be detected in a biological investigation of a tissue in a specific organism, at a particular developmental stage and also not detected by another biologist in a different experiment with the same tissue, at the same developmental stage. There will be instances of IID when these result sets are integrated into a single dataset. These anomalies affect the validity of conclusions drawn from the dataset. Consequently, deductions such as the cause of an abnormality in the investigated organism can be incorrect or misleading when IID is not considered in the analyses. This is because such deductions might have contradicting data in the dataset. The deduction can also be incomplete in the dataset. It is therefore necessary that any conclusion drawn from an integrated dataset should include the analyses of its associated IID.

Any datum with some (but not all) missing attribute values is referred to as an incomplete datum (Hathaway & Bezdek, 2001) while any datum that does not conform to the rules governing its design is referred to as an inconsistent datum (Nwagwu, 2013). In Bleihoder & Naumann (2008), there is recognition of incompleteness and inconsistency in and between sources, as a problem in a data integration setting. Also, the existence of IID in online distributed life science resources is discussed in McLeod & Burger, (2007).

World Wide Web Consortium (W3C) recommended Resource Description Framework (RDF) as a means of increasing the interoperability of web data (Klyne et al., 2004). Even so, IID can still exist in an RDF dataset or in the retrieved information from the dataset. In an RDF dataset, an object can be associated with various attribute values which can result in inconsistency or incompleteness in the dataset. This is because contradictory attribute values can be associated to an object in RDF dataset. Also, incompleteness will exist in an object which does not have a required value. As stated in Klyne et al., (2004) “RDF does not prevent anyone from making assertions that are nonsensical or inconsistent with other statements or the world as people see it. Designers of applications that use RDF should be aware of this and may design their applications to tolerate incomplete or inconsistent sources of information.”

IID can be identified through statistical or mathematical formula (Drumond et al., 2012) which can also be presented as numeric figures or in tabular representations. But these methods of visualising and identifying IID can be boring and difficult to perceive. It can also be challenging for non-statisticians to comprehend statistical information. In addition, IID displayed in a tabular form will not be easy to analyse especially where the data is in many rows and columns. For example, there are difficulties with visualising at a glance relationships existing in IID across many rows of a table.

This work limits its study to identifying and visualising IID existing in an RDF dataset. It demonstrates how information containing IID can be retrieved from RDF dataset. It describes an FCA approach for identifying and visualising IID in the retrieved information (recordset). We note that more can be done with IID, such as predicting the missing data but, those are outside the scope of this work. The investigations are limited to objects in a recordset whose attribute values are mutually exclusive. We use an EMAGE¹ dataset as our use case in illustrating our approach. Owlim-SE triple store² and FCA tools³ which include FcaBedrock⁴ and Concept Explorer⁵ are used as our processing engine. In section 2, Formal Concept Analysis (FCA). IID of RDF dataset is described in section 3. We explained how to identify and visualise IID of a recordset through FCA in section 4. We provided the background of our examined dataset, our approach to identifying and visualising IID, and our results and analyses in section 5. Finally, the evaluation of this work, its conclusion and future work are presented in section 6.