ASpecD: A Modular Framework for the Analysis of Spectroscopic Data Focussing on Reproducibility and Good Scientific Practice

Abstract

Reproducibility is at the heart of science. However, most published results usually lack the information necessary to be independently reproduced. Even more, most authors will not be able to reproduce the results from a few years ago due to lacking a gap-less record of every processing and analysis step including all parameters involved. There is only one way to overcome this problem: developing robust tools for data analysis that, while maintaining a maximum of flexibility in their application, allow the user to perform advanced processing steps in a scientifically sound way. At the same time, the only viable approach for reproducible and traceable analysis is to relieve the user of the responsibility for logging all processing steps and their parameters. This can only be achieved by using a system that takes care of these crucial though often neglected tasks. Here, we present a solution to this problem: a framework for the analysis of spectroscopic data (ASpecD) written in the Python programming language that can be used without any actual programming needed. This framework is made available open-source and free of charge and focusses on usability, small footprint and modularity while ensuring reproducibility and good scientific practice. Furthermore, we present a set of best practices and design rules for scientific software development and data analysis. Together, this empowers scientists to focus on their research minimising the need to implement complex software tools while ensuring full reproducibility. We anticipate this to have a major impact on reproducibility and good scientific practice, as we raise the awareness of their importance, summarise proven best practices and present a working user-friendly software solution.

Towards more reproducible and FAIRer research data: documenting provenance during data acquisition using the Infofile format

“…Other aspects of reproducible research [40] such as a complete audit trail of data processing and analysis are out of scope, but dealt with, e.g. scientific workflow systems, such as the ASpecD framework [41].…”

“…In any case, these key-value pairs allow for a 'semantic understanding' by the routines for data processing and analysis. [41] Furthermore, the metadata should be long-lasting and as permanent as possible. This imposes severe contraints on the available file formats.…”

“…This requires a thorough understanding of the processes resulting in a mental model of a dataset that can only be achieved and created by means of sufficient experience. One possible implementation of this model, consisting of the recorded (mostly numeric) data and the accompanying, machine-actionable metadata, is the dataset in context of the ASpecD framework [41]. The data model needs to be modular and expandable, another application for the open-closed principle [44, p. 57] mentioned above.…”

“…Originally, the format has been developed for use with a series of MATLAB ® toolboxes used for processing and analysing different kinds of spectroscopic data, i.e. the predecessors of what later became the ASpecD framework for reproducible spectroscopic data processing and analysis [41,49] and the packages based on it [50][51][52][53]. The information contained in these files is transferred to the associated dataset, one of the key concepts not only of the ASpecD framework, the inseparable unit of data and its accompanying metadata.…”

“…If your setup is controlled by lab-written software, make sure all the best practices of (scientific) software development apply, as a bare minimum using a version control system and unique version numbers. For an overview see [58] and [41] and references therein. Measured values should have both, numeric value and unit.…”