Data marts

Figure 9.17. Reagent Selector—an example of a chemical data mart. Various components of the system are shown, including the data sources, the daemon program that automatically updates the mart, the concordance database, and the client/server architecture, which is implemented in a three-tier system.

Dimension Tables. In a data mart or warehouse, the dimension tables store non-redun-dant information about the entries in the fact table of the database. For the chemical example of an inventory data mart, the fact table stores the various source database identifiers cf each unique structure in the data mart. A dimension table of molecular formulas would store the formula for the unique structure in the mart, rather than storing the same formula for each occurrence of that structure in the various source databases. 

Multidimensional Database. A relational database in which multiple general types of data are stored, indexed, and cross-referenced, for use by several different groups. In chemistry, an example would be a database containing reactions, 2D structures, perhaps generic structures or libraries, and 3D models. Such a database would be used by synthetic, chemical informatic, and molecular modeling scientists. A data warehouse is often a multidimensional database, whereas a data mart is usually single-dimensional. 

Figure 2 Database models data warehouse (A) and data mart (B).

A plethora of data marts are created with subsets of data extracted out of a central data warehouse. 

Separate, smaller warehouses typically defined along organization s departmental needs. This selectivity of information results in greater query performance and manageability of data. A collection of data marts (functional warehouses) for each of the organization s business functions can be considered an enterprise warehousing solution. 

The data staging area is the data warehouse workbench. It is the place where raw data are brought in, cleaned, combined, archived, and eventually exported to one or more data marts. 

Data warehouse A repository for data organized in a format that is suitable for ad hoc query processing. Data warehouses are buHt from operational databases used for day-to-day business processes. The operational data is cleaned and transformed in such a way that it is amenable to fast retrieval and efficient analysis. A single-purpose data warehouse is sometimes referred to as a data mart. ... 

Initially, it was planned to implement a separate interactive front-end for the CIDB to allow for a native access to the database contents. However, our experience shows that such a front-end is rarely used in practice and that it is sufficient to provide access via workflow tools (e.g., Pipeline Pilot [6] and KNIME [7]). Extracts of the data are made available to end users through the project data marts described in Section 13.2.5. 

FIGURE 13.2 An example for a BICEPS planning database that is used to collect and characterize synthesis ideas (virtual compounds) for a research project. A number of molecular properties are predicted and used to prioritize synthesis ideas. BICEPS planning databases are a special type of project data mart. 

All newly synthesized structures and all experimental results are stored in the CDB. The data model of this data warehouse is necessarily very complex and flexible, which makes it often difficult and slow to directly retrieve data in a useful format. Therefore, a specialized project data mart is set up that contains only the relevant structures and results for a particular project. This includes sample-related data such as availability or analytics results, but also project-related assay data as well as experimental physicochemical and ADMET results. All results are pivoted to a suitable format, normalized to a certain unit, and presented in a conveniently readable form. This includes the calculation of mean values (and other statistical parameters) in the case of multiple measurements and conditional formatting according to critical thresholds. An automated daily update procedure retrieves new structures and results from the CDB and adds them to the data mart. In addition to experimental results, the BICEPS mechanism is used to provide access to calculated properties (cf. Section 13.2.4). Since project data marts serve as central data exchange and communication platform within research projects, they are a natural reporting front-end for many other Cl tasks such as the results from HTS data analyses (cf. Section 13.3.1) or the BioProfile of a compound (cf. Section 13.3.2). The communication within the project is facihtated by a number of manual comment fields that can be used by the team to add comments or suggestions and to track the status of requests for assay testing. 

As these project data marts are stored in an Oracle [18] database instance, there are several possibilities to access the data. Analysis of the data can be done with work-flow tools like Pipehne Pilot [6] or KNIME [7]. Visualization in mass data viewers like Spotfire [19] is also possible. Currently, medicinal chemists often use ISIS/Base [20], using a predefined set of standard forms, which can be adjusted if necessary. An example for such a form is shown in Figure 13.2. 

The outcome of a BIMESH analysis is thus a convenient information package that is distributed to the project team and used to prioritize and select the most promising structural classes for further profiling and analysis. The static reports provide an overview of the various structural classes while the project data mart can be used to retrieve detailed information about the different molecirles in a given cluster. This interplay between the different outputs allows project teams to get a quick and substantiated overview of the resrrlts of a screening campaign. 

We rely mainly on workflows built in KNIME workflows to analyze this data. The per-compound data are also available through our project data marts as described and shown in Section 13.2.5. In the following, we describe two application examples. The first one is a compound-based analysis using primary assay data to identify frequent hitters. The second example is an analysis based on the dose- esponse data of a complete hit set. 

---