How collaborative projects that involve complicated electrophysiological data sets profit from workflow design

Michael Denker (Institute of Neuroscience and Medicine (INM-6), Research Center Jülich, 52428 Jülich, Germany), Andrew Davison (Unité de Neurosciences, Information et Complexité (UNIC), CNRS UPR-3293, 91198 Gif sur Yvette, France), Markus Diesmann (Institute of Neuroscience and Medicine (INM-6), Research Center Jülich, 52428 Jülich, Germany), Sonja Gruen (Institute of Neuroscience and Medicine (INM-6), Research Center Jülich, 52428 Jülich, Germany)

The recent years have seen a rapid increase in the complexity of electrophysiology experiments. This complexity arises firstly from the interest in simultaneously analyzing the activity recorded from large numbers of channels in order to investigate the role of concerted neural activity in brain function. These efforts have led to advances in data analysis methods [1] that exploit the parallel properties of such data sets [2]. A second source of complexity is in the sophistication of stimulus protocols. To take the visual system as an example, typical visual stimulation has progressed from simple moving bars or drifting gratings to natural movies, Gabor noise, apparent motion stimuli. In the somatosensory system, new technology now allows the entire rodent whisker array to be stimulated in essentially arbitrary patterns. However, an often neglected aspect of these technological advances is that both massively parallel data streams and highly complex stimuli place new demands on handling their complexity during all stages of the project [3]: from the initial recording, throughout the analysis process, to the final publication.

Three factors contribute these new demands: First, the sheer quantity of data complicates the organization of data sources, and the resulting automatization of analysis steps renders the validation of interim and final results difficult. Second, modern analysis methods often require intricate, multi-layered implementations, leading to sophisticated analysis toolchains [4]. Third, a growing number of projects needs to be carried out in teams, within a laboratory or in collaborative efforts, requiring transparent workflows that guarantee smooth interaction. Taken together, the increase in complexity calls for a reevaluation of the ad-hoc traditional approaches to such projects. Can we derive general guiding principles that may be adopted for designs of efficient workflows? How could these improve our confidence in handling the data by providing better cross-validation of findings, reliably managing provenance data, and enabling tighter collaborative research, while at the same time leaving the scientist with the flexibility required for creative research?

Although several projects are devoted to finding solutions for specific aspects of a workflow design (e.g., [5-7]), on a more general level there is lack of a thorough discussion on what goals are expected from a workflow, and which of these can be realistically addressed. Here, we summarize feedback received from experimenters and theoreticians that pinpoints the fundamental problems typically encountered in the analysis of high-dimensional electrophysiological data. Illustrated by examples from our own experience, we further show obstacles that prevent us from harmonizing workflows to common guidelines. For selected issues we draw parallels to other communities that are faced with similar problems (e.g., neuronal network modeling [8-9]; neuroimaging [10]). Lastly, we propose how existing concepts and software [9,11] could assist in practically implementing workflows that are tailored to the needs of a specific project, yet guarantee high standards by adhering to general guidelines of accepted best-practice.

Acknowledgements: This project was supported by the European Union (FP7-ICT-2009-6, BrainScales).

1. Brown EN, Kass RE, Mitra PP: Multiple neural spike train data analysis: state-of-the-art and future challenges. Nat Neurosci 2004, 7:456-461.
2. Stevenson IH, Kording, KP: How advances in neural recording affect data analysis. Nat Neurosci 2011, 14:139-142.
3. Buzsáki G: Large-scale recording of neuronal ensembles. Nat Neurosci 2004, 7:446-451.
4. Denker M, Wiebelt B, Fliegner D, Diesmann M, Morrison A: Practically trivial parallel data processing in a neuroscience laboratory. In: Analysis of parallel spike trains. New York: Springer-Verlag; 2010.
5. CARMEN: Code analysis, repository & modeling for e-neuroscience []6. Herz AVM, Meier R, Nawrot MP, Schiegel W, Zito T: G-Node: An integrated tool-sharing platform to support cellular and systems neurophysiology in the age of global neuroinformatics. Neural Networks 2008, 21:1070-1075.
7. CRCNS - Collaborative research in computational neuroscience []
8. Nordlie E, Gewaltig MO, Plesser HE: Towards reproducible descriptions of neuronal network models. PLoS Comput Biol 2009, 5:e1000456.
9. Sumatra: automated electronic lab book []
10. LONI Pipeline []
11. VisTrails []; Taverna []; Kepler []


Preferred presentation format: Poster
Topic: Electrophysiology

Document Actions