Chapter 3
Biomedical Big Data - CHIMeRA project
Every second a large quantity of data are produced and shared along the Internet and web-pages. Data are collected by social networks, chat messages, video streaming and images. Everyone, in fact, can easily create new data sources and share or put them in Internet pages. The growth of these data is not limited to multimedia data but it involves many different fields. This is one of the most important features of the contemporary time, the so-called Big Data era: this huge volume of data has created a new field in data processing which is called Big Data Analytics that nowadays has positioned among the top ten strategic technologies (Gartner Research, 2012).
It is still difficult to provide a definition of what exactly are the Big Data and we can find many slight different nomenclatures and categories which aim to formulate it. Moreover, Big Data does not define a particular data type but more than we normally think sources can be labeled as it. The International Journal of Computer Applications defined them as “[…] a collection of large and complex datasets that cannot be processed and analyzed using traditional computing techniques. It is not a single technique or a tool, rather it involves many areas of business and technology”. This definition certainly involves many aspects of Big Data processing but it does not provide any definition about their nature. Moreover, it is easy to identify them as “big” and thus difficult to analyze, but they are all around us every day and just using the Internet connection every smart-phone or laptop can extract and visualize our web queries so could be not properly correct to define them in this way. However, it is sure that standard computing techniques have to be reviewed to face this vast amount of data and a even more important attention has to be payed on the implementation of algorithms.
While a global definition of them is evidently difficult, we can provide a description of them using some of their “essential” features. One of the most common and used set of labels for this purpose is given by the so-called 5 V’s of Big Data: volume, velocity, variety, veracity and value. Despite the first twos are quite obvious (the Big Data are certainly big in volume and they are produced very fast), the remaining three need a particular attention. We have already treated problems about the volume of data (ref. Chapter 1) and the need of very fast processing and algorithmic optimizations (ref. Chapter 2) so in this chapter we want to focus on the remaining three characteristics of Big Data Analytics.
As pre-announced there are many different sources able to provide data and this feature describes the extreme heterogeneity and variety of them. We can however broadly classify this variety into three global classes: structured, semi-structured and unstructured data. A dataset is structured if we can easily manage the information in it or, in other words, if it is described using the standard formats of data and thus if it can be “quearable”. On the other side we have the completely unstructured datasets in which data are disorganized and we need one or multiple pre-processing steps before use them or we have to develop different techniques to handle them. The intermediate format is given by the semi-structured data in which only a part of them could be handled with standard techniques or if they can be easily brought to their structured version. The organization of data was a crucial task on this work of thesis and we will return on this topic in the next sections.
The fourth essential characteristic of Big Data is their veracity due to data inconsistency and incompleteness. The data are shared very fast using Internet and we can find some ambiguities and/or deceptions between different data sources. If we want to merge and aggregate different kinds of information we have also to face this kind of problems. The final task of every Big Data Analytic application is, in fact, to process these large quantities of data and obtain a unique answer to a problem which can not vary in relation to the portion of samples or datasets used.
The last and probably most important feature is certainly their value: it is good to have access to large amount of data but unless we can turn them into value they are useless. In this vast amount of data only a small part of them can be considered as informative and it is always harder to extract this core of information from them. Moreover, we have to take into account also the difficulties about the management of these data and their more or less complex structure. However, also in this case, it is hard to generalize this property to all the amount of data contained in Internet: every day we see a large quantity of useless information in the Web and it is hard to figure out that some of them can be useful for research applications. A key role is played by the questions which we ask: for every data source there is always an appropriate question which can be answer using it and that can give a value to it and vice versa. In this way also the seemingly useless datasets can acquire importance for an appropriate research project.
In this chapter we are going to discuss about one of the latest project developed during my PhD and which is still in work in progress: the CHIMera (Complex Human Interactions in MEdical Records and Atlases) project.
The project is founded on Big Data applications and it is accidentally born as separated branch of the INFN FiloBlu project (ref. next sections and Appendix E for further information about the FiloBlu project) which financed my last PhD year.
CHIMeRA aims to create a unified database of biomedical records using Natural Language processing techniques.
Its final purpose is to merge multiple data sources available on-line into a single network structure which highlights the relevant interactions between biomedical information, i.e starting from diseases to the biological agents and compounds involved into their causes and consequences.
The realization of the first version of CHIMeRA required a lot of time and the development of novel pipelines of data processing.
The project does not still achieve its conclusions but in this chapter we are going to cross through the main key points which allowed its construction.