- Strategies to improve the preservation, searchability, and discoverability of research data are a priority. To facilitate these efforts in cell electrophysiology and biophysics we propose that ontologies be used to design and annotate data, as they provide a substantive metadata structure, with reasoned-definitions arranged in a logical, hierarchal structure where the meaning of data are unambiguously assigned. We illustrate this by describing our cell electrophysiology data with an ontology. We then make this hierarchal structure with definitions the basis of the data architecture which is implemented upon transforming the data into the storage format: Hierarchical Data Format version 5 (HDF5).
Data were inconsistently reported in the past, although data forms the back-bone of much of scientific discovery. There was no motivation for researchers to develop intuitive humanunderstandable (and machine readable) data structures that both the public, and their scientific peers, could readily access. The limits of this approach for many scientific disciplines have revealed a number of deficiencies, including the inability to reproduce key findings, coupled with excessive (additional) costs to the tax payer [1]. The need to provide data sets that support key findings of research is particularly relevant in fields where data collection is slow and requires the sacrifice of mammals. This is the case in auditory electrophysiology. To facilitate data preservation and sharing, we describe our effort to transform electrophysiological data from private to public use.
The data were generated by whole-cell voltage clamping isolated outer hair cells obtained from the domestic guinea pig (i.e., cavia porcellus). The electrical properties (e.g., membrane capacitance) of the outer hair cells were then determined [2] from the electrical recordings. The data encompass results from the assay, as well as the associated properties of the animals and cells; the experimental conditions employed; and a description of the devices used.
This work was funded by NIH NLM and NIDCD R01DC000354-S1.
To provide for the durability of the data across both time and space so that others may reasonably expect to make use of it, we describe the data with an ontology; essentially, a structure which places defined concepts in a logical relationship with one another as a way to describe things, events, or ideas that exist in the objective world. Many ontologies exist, including a large number which focus on various aspects of the biological sciences. Of relevance is Ion Channel Electrophysiology ontology, ICEPO [3] that describes concepts that are associated with electrical and temporal characteristics of voltage-gated ion channels. Although some of the concepts (e.g., gating current, ICEPO_0000049) have been used to describe the electrical characteristics of outer hair cells, the membrane protein that forms part of the voltage-sensing component in the lateral membrane of outer hair cells is not an ion-channel. We generalize, where appropriate, the concepts introduced in ICEPO to align them with voltage-dependent behavior of membrane assemblies. Another more extensive ontology is the Ontology of Physics for Biology (OPB) [4] that was developed to annotate computational models of biological systems. It uses physical (e.g., thermodynamic) dependencies to describe biological processes, and although it also covers some relevant concepts, it does not describe the design, collection and analysis of data. Having examined other ontologies, we decided format, with available open source viewers, it has conversion and that Ontology for Biomedical Investigations, OBI [5] was the editing application programming interfaces (APIs) for a variety most robust and logical match for the data. However, it does not of languages (including MATLAB), it supports complex and address all the concepts required for this data set. Approximately large data structures, and it already enjoys significant scientific 100 are not found within OBI. We approached such cases by usage. importing the terms from other ontologies, and by creating new classes (~ 20). These imported and new terms are grafted at The six main classes become the main branches of the data suitable junctions onto a variant of OBI we have produced. design (Fig. 2, right panel). All data associated with an individual Although these terms are being considered for acceptance into the outer hair cell were arranged in a tree configuration and saved to OBI ontology, we stress that our goal is not to expand the a file. The tree associated with the organism Group (terminology ontology for its own sake, but to expand and use this application used with HDF5) is shown where the classes of the ontology now ontology to describe the data. readily become the name for subgroups, datasets (terminology also used by HDF5), and data values. Similar mapping is
The six main classes used to define the data in OBI are shown performed for the other arms (Fig. 1) that describe the data. in Fig. 1. The data are described with one electrophysiology assay: whole cell patch-clamp voltage clamp assay, with the ACKNOWLEDGMENT devices (e.g. patch-clamp device) needed to perform the assay in We thank Anita Bandrowski, Owen Ellard, Ronna Hertzano, a separate class. The assay was performed with outer hair cells Barbara Jones, Elena Pormal, Joseph Santos-Sacchi, Jeffrey isolated from cochlea of guinea pigs; hence cell, organism, and Teeters and Randy Vita for their contributions to this work. anatomical entity were the three other material entities. To extract key parameters from the electrical measurements we REFERENCES performed analysis and hence data transformation is the 6th class. 1.
These six main classes are the major arms of the data. In Fig. 2 we show the class structure for the organism arm with the easxspoecriiamteednt.daFtoar veaxlaumespleth:atthearmeantuorrimtyalalyndresceoxrdoefdthdeurainnigmaaln; 2. whether the guinea pig exhibited a normal or albino phenotype; the mass of the animal; the age since birth; and, for adult females, the estrous cycle phase. Similar class relationships are 3. formulated for the other five arms (data not shown).
The original data were stored in MATLAB (Mathworks, MA), as a struct. MATLAB is proprietary software and not suitable for expansive data sharing, as this software must be licensed at a significant cost, and is not accessible to everyone. We 5. transformed this data to Hierarchical Data Format version 5 (HDF5) [6] which was developed for storage of large and/or 6. complex sets of data. We chose it because it is an open source