Submit a Revision

*your email :**
page name* : such as "Introduction", "Conclusion"..etc
A fundamental goal of systems biology is constructing mathematical<sup></sup>models that elucidate key features of biological processes as<sup></sup>they exist in real cells. A critical step in realizing this<sup></sup>goal is effectively calibrating models against experimental<sup></sup>data. The challenges of model calibration are well recognized<sup></sup>(Jaqaman and Danuser, <a href="http://bioinformatics.oxfordjournals.org/cgi/content/full/24/6/840#B5">2006</a>) but we have found systematizing<sup></sup>and processing data prior to calibration to be tricky as well.<sup></sup>This is particularly true as the volume of data or the complexity<sup></sup>of models grows. Few information systems exist to organize,<sup></sup>store and normalize the wide range of experimental data encountered<sup></sup>in contemporary molecular biology in a sufficiently systematic<sup></sup>manner to maintain provenance and meanwhile retaining the adaptability<sup></sup>necessary to accommodate changing methods. Partly as a consequence,<sup></sup>relatively few complex physiological processes have been modeled<sup></sup>using a combination of theory and high throughput experimental<sup></sup>data.<sup></sup><p> An information management system for experimental data must<sup></sup>record data provenance and experimental conditions, maintain<sup></sup>data integrity as various numerical transformations are performed,<sup></sup>describe data in terms of a standardized terminology, promote<sup></sup>data reuse and facilitate data sharing. The most common way<sup></sup>to achieve these requirements is via a relational database management<sup></sup>system (RDBMS, see SBEAMS—<a href="http://www.sbeams.org/">http://www.sbeams.org</a>—or<sup></sup>Bioinformatics Resource Manager for relevant examples; Shah<sup></sup><em>et al.</em>, <a href="http://bioinformatics.oxfordjournals.org/cgi/content/full/24/6/840#B8">2006</a>). Databases in biology resemble those previously<sup></sup>developed for business and have proven spectacularly successful<sup></sup>in managing data on DNA and protein sequences. In a relational<sup></sup>database, the subdivision of information and its subsequent<sup></sup>storage into cross-indexed tables follows a precise, predefined<sup></sup>schema. The granularity and stability of the schema allows an<sup></sup>RDBMS to identify and maintain links between disparate pieces<sup></sup>of information, even in the face of frequent read–write<sup></sup>operations. However, this power comes at a considerable cost<sup></sup>in terms of inflexibility. It is difficult for a relational<sup></sup>database to accommodate frequent changes in the formats of data<sup></sup>or metadata, and to incorporate unstructured information.<sup></sup></p> <p> Whereas the sequence of a human gene represents valuable information<sup></sup>independent of how sequencing was performed or of the individual<sup></sup>from whom the DNA was obtained (a statement that remains true<sup></sup>despite the value of characterizing sequence variations); such<sup></sup>is not the case for measures of protein activity or cellular<sup></sup>state. Such biochemical and physiological data are highly context<sup></sup>dependent. Data on ERK kinase activity, for example, is uninformative<sup></sup>in the absence of information on cell type, growth conditions,<sup></sup>etc. Moreover, a wide range of techniques are used to make biochemical<sup></sup>and physiological measurements, and both the assays and the<sup></sup>data they generate change over time, as new methods are developed<sup></sup>(e.g. in imaging see Swedlow <em>et al.</em>, <a href="http://bioinformatics.oxfordjournals.org/cgi/content/full/24/6/840#B10">2003</a>). Context dependence<sup></sup>and rapidly changing data formats pose fundamental problems<sup></sup>for databases because RDBMS schemes are not easily modified.<sup></sup></p> <p> Moreover, even if effective metadata standards are developed<sup></sup>to describe the context-dependence of experimental findings,<sup></sup>data from different experiments cannot be reconciled simply<sup></sup>by storing them in a single database. Subtle distinctions must<sup></sup>be made about different types of data and biological insight<sup></sup>brought to bear. Currently this is performed implicitly in the<sup></sup>minds of individual investigators, but we envision a future<sup></sup>in which the unique ability of mathematical models to formalize<sup></sup>hypotheses and manage contingent information makes them the<sup></sup>primary repositories of biological knowledge. As we work towards<sup></sup>a model-centric future, it is our contention that information<sup></sup>systems based solely on relational databases are unnecessarily<sup></sup>limiting; rarely do we modify a difficult experiment simply<sup></sup>to conform to a pre-existing database schema (whereas conformity<sup></sup>to uniform—even arbitrary—standards is a strength<sup></sup>for a business database). New approaches to data management<sup></sup>that reconcile competing requirements for flexibility and structure<sup></sup>are required.<sup></sup></p> One response to the challenges of systematizing biological data<sup></sup>has been the creation of lightweight data standards focused<sup></sup>on the most important metadata. Pioneered by the Microarray<sup></sup>and Gene Expression Data Society's <em>Minimum Information about<sup></sup>a Microarray Experiment</em> (MIAME), these ‘minimum information’<sup></sup>approaches typically define a simple data model that can be<sup></sup>instantiated as an XML file, a database schema, etc. A strength<sup></sup>of ‘minimum information’ standards is that they<sup></sup>specify that subset of the metadata that is relatively constant<sup></sup>among ever-shifting and context-sensitive experiments. The philosophy<sup></sup>is that of the <em>Pareto principle</em> or <em>80-20 rule</em>, namely that 80%<sup></sup>of the information can be captured with 20% of the effort whereas<sup></sup>the final 20% requires exponentially greater effort. An underlying<sup></sup>assumption is that a minimum information standard successfully<sup></sup>records the information needed to make experimental data intelligible.<sup></sup>In this article we implement an information processing system,<sup></sup><em>DataRail</em>, intended to bridge the gap between data acquisition<sup></sup>and modeling. A new minimum information standard (MIDAS) is<sup></sup>part of the <em>DataRail</em> system, but a series of additional tools<sup></sup>are also applied to maintain the provenance of data and ensure<sup></sup>its integrity through multiple steps of numerical manipulation.<sup></sup><em>DataRail</em> is model- rather than data-centric in that the task<sup></sup>of creating and transmitting knowledge is invested in mathematical<sup></sup>models constructed using the software, rather than the data<sup></sup>storage system itself, but it is designed to support existing<sup></sup>modeling tools rather than serve itself as an integrated modeling<sup></sup>environment. We illustrate this capacity in <em>DataRail</em> using a<sup></sup>large set of protein measurements derived from primary and transformed<sup></sup>hepatocytes; through the use of <em>DataRail</em> we derive insight both<sup></sup>into the biology of these cell types and the optimal means by<sup></sup>which to perform partial least squares regression (PLSR) modeling<sup></sup>of cue-signal-response data.
Confirmation code: Enter the code exactly as it appears. All letters are case insensitive, there is no zero.