Robust Cross-Platform Workflows: How Technical and Scientific Communities Collaborate to Develop, Test and Share Best Practices for Data Analysis

Abstract

Information integration and workflow technologies for data analysis have always been major fields of investigation in bioinformatics. A range of popular workflow suites are available to support analyses in computational biology. Commercial providers tend to offer prepared applications remote to their clients. However, for most academic environments with local expertise, novel data collection techniques or novel data analysis, it is essential to have all the flexibility of open-source tools and open-source workflow descriptions. Workflows in datadriven science such as computational biology have considerably gained in complexity. New tools or new releases with additional features arrive at an enormous pace, and new reference data or concepts for quality control are emerging. A well-abstracted workflow and the exchange of the same across work groups have an enormous impact on the efficiency of research and the further development of the field. High-throughput sequencing adds to the avalanche of data available in the field; efficient computation and, in particular, parallel execution motivate the transition from traditional scripts and Makefiles to workflows. We here review the extant software development and distribution model with a focus on the role of integration testing and discuss the effect of common workflow language on distributions of open-source scientific software to swiftly and reliably provide the tools demanded for the execution of such formally described workflows. It is contended that, alleviated from technical differences for the execution on local machines, clusters or the cloud, communities also gain the technical means to test workflow-driven interaction across several software packages.

refnx: neutron and X-ray reflectometry analysis in Python

“…In order for a given scattering analysis to be fully reproducible by others, a general set of conditions need to be met (Helliwell et al, 2017;Mö ller et al, 2017):…”

“…An alternative way of recreating the environment is by using a virtual machine, or another container environment such as Docker. The strengths and weaknesses of various software distribution practices and their relationship with reproducible science have been discussed in detail elsewhere (Mö ller et al, 2017).…”

“…Reproducibility in research is an underlying principle of science, but unfortunately it is not always possible to reproduce the results of others (Stark, 2018), because there is frequently not enough information provided in journal articles to repeat the analyses. Even if the data sets and the software packages used to analyse them are supplied in supporting information (most often they are not), a comprehensive, ordered set of instructions or a codified workflow would need to be provided (Mö ller et al, 2017). One example addressing this reproducibility issue is the set of guidelines from the small-angle scattering community for the deposition of data and and associated models (Trewhella et al, 2017;Pauw, 2013).…”

Cloud-Based High Throughput Virtual Screening in Novel Drug Discovery

“…Several tools are available to prepare receptors and ligands. Glue language can be used to integrate different processes of such calculations [87]. In addition, there are workflow environments for CADD in particular like OPAL [88] or KNIME [89].…”

nf-core: Community curated bioinformatics pipelines

“…While being strongly represented in the analysis of genomics data, the community also provides a more general framework for building and maintaining analysis pipelines in proteomics, metabolomics, and other life science applications. In theory, the same should also be applicable to the broader field of scientific data analysis [17][18][19] including more data-intensive fields such as chemistry and physics.…”