Datasets common types

Cross comparisons across toxicogenomic datasets provides new statistical questions and subsequent challenges in data analysis. Meta-analyses may integrate datasets from multiple experimental studies consisting of different models (species, source), platforms (array type), statistical techniques (normalization) and design. The first challenge that needs to be addressed is how to properly make comparisons across datasets. To normalize datasets, better results may be achieved when data is first normalized internally and then externally (88). Secondly, equivalent and current annotation is needed to identify common genes across platforms, models, etc. 

Several European intensive short-term ( campaign-type ) projects have provided important information on the atmospheric aerosol properties in Europe, usually by concentrating on specific aerosol properties or interactions. However, these kinds of campaign-type measurements do not necessarily represent the seasonal or annual variations of the aerosol concentrations and can overestimate some properties of the aerosol populations. Long-term measurements, especially with intercalibrated instruments and common data handling and calibration protocols make the data comparison between stations much more reliable and provide the end users (e.g., atmospheric modelers) good datasets to compare with. 

Validation The use of TMAs enables analysis of large data sets, however this does not by any means suggest that the data set is not skewed. This skewing may be the result of the institution s location (population distributions with regards to race, ethnicity, access to health care), type of practice (community hospital versus referral center). These collectively might influence the tumor size, grade and subtype composition of the cases in the dataset. Such abnormalities of the dataset need to be compensated the involvement of a biostatistician from the start (i.e from case selection) helps to prevent the creation of biased TMAs. It is useful to perform common biomarker analysis on sections from the created TMA to confirm the normal distribution of known parameters. Comparison of this data with prior clinical data (e.g. ER analysis) obtained from whole section analysis is particularly useful to validate utility of the TMA. Alternatively the incidence of expression of a number of biomarkers in the TMA should be compared to that in published literature (using whole sections). 

In rare and interesting cases it is possible to rank the size of the variables along each column. The suitability depends on the type of preprocessing performed first on the rows. However, a common method is to give the most intense reading in any column a value of I and the least intense 1. If the absolute values of each variable are not very meaningful, this procedure is an alternative that takes into account relative intensities. This procedure is exemplified by reference to the dataset C, and illustrated in Table 6.4. 

In contrast to the previous applications, which rely on predefined classes or states, it is also possible to analyze the expression datasets in order to discover classes or subgroups. This type of analysis, also known as clustering, searches for subgroups of samples with common features within them and distinct from others. One application of these methods is to identify molecular classes and improve the understanding of the differences arising within apparently homogeneous conditions. In medicine, this method has been successfully applied to identify new subtypes of cancer (43). 

In contrast, federated architectures tend to be more flexible and are more generally applicable. Typically they either leave data in its native format or require that data be put in a format common to all the datasets. They do not rely on any domain-specific abstractions but instead model the generic features of data and employ some kind of query-based logic for their API abstractions. These solutions tend to be much more extensible and require configuration rather than a programmatic effort when bringing new data types into the system. These federated architectures can be either local or distributed. 

Managing reference data can be particularly troublesome when two or more systems are merged for example when two local systems are being replaced by a single alternative and there is a need to preserve historical cliifical data. Each system is likely to have its own reference data and, particularly over time, it is common for these datasets to have similar purposes but with different content. For example, a Patient Administration System may have five different options for Admission type whilst the Electronic Health Record may have seven different options for that same field. Whilst it might be possible to live with this discrepancy on a day-to-day basis, the situation suddenly becomes very complex when it is necessary to merge the two datasets. Any proposed solutions need a careful safety assessment. 

---