Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Nature Methods (2022)
46
Metrics details
Here we discuss barriers to reproducibility in regard to microscopes and related hardware, along with best practices for sharing novel designs created using computer-aided design (CAD). We hope to start a fruitful community discussion on how instrument development, especially in microscopy, can become more open and reproducible, ultimately leading to better, more trustworthy science.
You have full access to this article via your institution.
Microscopy has often been at the heart of new biological discoveries, and as cutting-edge experiments become more complex, so do the new microscopes required to image them. This has led to rapid growth in interdisciplinary collaborations to develop novel instruments. Increasingly, this has been done using reusable building blocks.
Many scientists successfully use CAD software to design custom parts, assemble setups from their components and render graphics. However, these detailed and reusable designs are rarely openly archived. In recent years, funders and publishers have formulated extensive guidelines and policies on the sharing of data and software in the context of open science1,2,3. In contrast, hardware development and its publications often lack suitable standards to ensure the completeness and quality of designs that are shared, even though such standards are available4,5. Indeed, there is often no requirement to share hardware designs at all. The usual practice of including a parts list and rough schematic in Supplementary Material is frequently incomplete and insufficient to allow readers reproduce the instrument without extensive involvement of the original authors. We believe that the community of researchers, journals and granting bodies should start treating design files as research data.
In this paper, we discuss how CAD and improved publication of design files can make the scientific process more open, reproducible and valuable. Using and properly sharing CAD files accelerates the dissemination of scientific knowledge, allows reproducibility at lower cost, permits a design’s reuse and improvement, and so promotes innovation in the rapidly evolving field of biological imaging. With this article, we hope to start a conversation about this vision, the associated difficulties and ways to overcome them.
Reproducing a novel scientific instrument can be a lengthy process. First, specific off-the-shelf components must be obtained, and any custom-made parts must be fabricated. Then, the component parts must be assembled into a complete instrument. Finally, the instrument must be aligned, commissioned and put into use. With the advent of rapid prototyping technologies such as three-dimensional (3D) printing and on-demand machine-shop services, scientists are increasingly adopting these tools in research6,7,8,9. A CAD file can describe both custom-made and commercially available parts and how to assemble these parts into complete instruments (Fig. 1). This is a convenient way to create and share the manufacturing instructions for custom parts, an accurate bill of materials describing off-the-shelf parts and information about how the components fit together into a complete instrument.
a, A mechanical assembly including custom parts from the UC2 toolbox modeled in Autodesk Inventor. Adapted with permission from ref. 22, copyright its authors. b, Assembly of the OpenFlexure optics module, rendered using OpenSCAD. Adapted with permission from ref. 23, IEEE. c, An assembled OpenSPIM showing beam path, modeled with SolidWorks.
A complete set of CAD files can provide most of the information required to reproduce a design of a hardware part or system, though usually a comprehensive set of assembly instructions is also required. A CAD file that is not shared, or is shared only in an inaccessible format, makes reproducibility difficult and costly. Improving the design may then require the part to be recreated from scratch in suitable format (an inefficient and time-consuming process). It is therefore crucial that design files linked to published works be archived in long-term repositories with stable links such as digital object identifiers (DOIs).
The use of 3D schematics to present protocols, workflow and instrument designs is commonplace and supported by the emergence of specialized journals and social platforms such as Nature Methods, JoVE (the Journal of Visualized Experiments), HardwareX and the Journal of Open Hardware. These visually appealing schematics spark interest and provide useful conceptual understanding. A short survey of Nature Methods papers (2009–2020) reveals that the majority of publications do employ some form of CAD, but crucial files are often missing, particularly photographs and CAD files (Fig. 2). More than half of the papers present CAD renderings, but do not attach original files to the publication. This is not a technical accessibility issue, as 5 out of 50 papers do provide extensive files in the Supplementary Data section.
Survey of microscopy-related publications in Nature Methods published in 2009–2020 that include any amount of information related to design of the experimental setup. Many publications use CAD byproducts (renderings), but few publications present any CAD files. When files are made available, the formats are inconsistent.
Source data
Requiring existing CAD files in their original, editable format, as well as exported formats that are easier to view or print, will not create undue burden but instead will increase the value to readers and allow greater reproducibility, improvement and adaptability. In our opinion, a design that is presented but not properly shared suggests a conflict of interest that ought to be declared; it suggests that the manuscript is promoting a product that is (or will be) commercialized without allowing full scrutiny and reproduction of the science. This is inconsistent with policies on data and code, which already require a statement linking to original files, or explaining why this is not done. Not every hardware design uses CAD, and so editable CAD files may not exist—but where they do exist, sharing them ought to be the default course of action.
There are many different CAD systems, each with its own native file format, and very few are intercompatible (see the online repository https://github.com/HohlbeinLab/OpenMicroscopy10). Most of these are commercial products with expensive licenses, meaning that a given institution rarely has access to all of them. This creates a huge barrier to collaboration between institutions or companies and can result in researchers losing access to their own work when they take a new position. Some cloud-based CAD platforms are now available, often at no cost but with no guarantee of future accessibility. Open-source CAD solutions are also available, which are free from restrictive license terms, but currently offer a less polished user experience. Manual technical drawings are an alternative to CAD, and if used correctly can convey information that is just as complete. However, a skilled draftsperson is needed to produce them, and subsequent modifications can be more difficult. It is also more difficult than with CAD to check whether a design is properly constrained and self-consistent.
Most CAD packages allow parametric modeling, in which the final 3D object is described in terms of meaningful dimensions and geometric constraints. Typically, 2D ‘sketches’ or technical drawings are used to define geometry that is then converted to 3D, and both the sketches and the operations to make a 3D model are retained (Fig. 3). Alternatively, primitive 3D shapes can be combined in different ways to build up a more complex model. In either approach, the editable CAD file has much more information than the final shape, as the relationships between different dimensions, and the sequence of operations required, must be retained to change the design effectively.
Native file formats are specific to particular CAD packages but preserve full parametric control and editability. STEP and STL files preserve shape, but not design constraints and parameters.
Custom parts are exported from the CAD package into a transfer format that can be interpreted by a computerized numerical control (CNC) machine or 3D printer. These files allow interoperation of different systems, but lack the parametric information needed for meaningful editing. The de facto standard for 3D-printed parts is an STL (STereoLithography) file, in which the model is represented by a triangular mesh that can be viewed or printed but is hard to modify. STEP (Standard for the Exchange of Product Data; ref. 11) files can preserve assembly arrangement and material properties such as the colors of parts, can accurately represent complex 3D objects including curved faces and can be edited more easily than STL meshes. Many CAD packages have reverse-engineering tools that allow STEP and STL files to be edited. This editing, however, is limited in scope because the original design constraints are lost.
Most CAD packages can export technical drawings for manual machining, and these often preserve more of the original design constraints. Good technical drawings can also make it easier for others to design accessories or incorporate a piece of apparatus into another experiment. However, reconstructing the editable CAD model based on drawings is a laborious process.
A universal parametric CAD format would solve a great many technical issues, not only for microscopy but for many fields in industry and in academia. For now, the best solution is to include both the proprietary, native CAD file and as many exported formats as is possible and appropriate for a particular design, including renderings and technical drawings. As with data, it is always important to document the files, indicating which are the originals and which are generated, and to ensure any information not contained within the CAD files is properly documented.
Optical supply companies such as ThorLabs, Newport, Edmund Optics, McMaster-Carr and recently also the microscopy manufacturer Olympus provide CAD models of their components. This allows designers to check compatibility, preservation of optical axes and physical constraints before purchasing and building a system. It also allows the design of specific parts ahead of assembly (for example, adapters with specific threads). This may be done with only main details of the part, such as rough outline and main openings, but can also go further, adding specific optical elements such as mirrors and lenses to see how the housing will accommodate them, or designing and planning for every screw. A CAD model can even be used to simulate or optimize a system’s optical performance12,13,14.
Creating a virtual assembly can save time and reduce effort wasted in procuring incompatible parts, as well as making it easier to share precise designs. Automatically generated bills of materials can make it much easier to obtain all the parts for a system, and exported images make publication-ready, informative figures (Fig. 4). Assemblies can be represented using many of the CAD formats described previously, and the same recommendation applies: both the native files and appropriate transfer formats (including bills of materials) should be archived as part of a complete design.
a, Virtual assembly can quickly create an instrument from off-the-shelf parts. Distances and angles can be precisely set, and conflicts checked and avoided early in design. b, Technical drawings and bills of materials are generated from the model.
Software and data are now routinely archived openly in support of published research, and funders usually require a statement detailing access arrangements and justifying any restrictions on sharing as part of each publication. Moreover, many universities and institutions have established policies regarding open science, often related to government guidelines and funders. Hardware designs, however, are often handled quite differently15.
Universities typically require researchers to allow their intellectual property office to ‘protect’ promising technologies with invention disclosures or patents, so that the institution can attempt to license future use of the work. Patents are one way to publicly communicate an idea, protecting that idea from being patented by another entity, but filing a patent is expensive and takes months, and thus is usually delayed until a strong business case can be made. If a novel instrument is to be patented, designs cannot be openly shared until patents are filed. However, sharing a design openly creates ‘prior art’, thereby acting like a patent to stop other entities from patenting the design. This saves both the cost and time of registering a patent, but the researcher must usually obtain the university’s agreement to share designs without patenting them. Often this means the publication of designs is delayed and the use of these data at conferences, job interviews, etc. is limited unless a non-disclosure agreement has been signed. Consequently, a great many hardware designs are neither patented nor shared openly. Patents (even if they are never exploited) are often counted when evaluating researchers, and this provides an incentive for individual scientists to opt for patenting a design, even when open sharing of a design would provide many of the same benefits more quickly and cheaply.
There is a need for clarity on the cost–benefit rationale of this patent-by-default approach as well as the relationship between public funding of research and the intellectual property system in financed institutions: the software community has demonstrated that commercialization and openness are not in opposition16. Companies such as the open-source pipetting robot provider Opentrons17 and the 3D printer manufacturer Prusa18 that release their entire hardware and software sources suggest that a re-think is taking place in industry as to the importance of patents for hardware products. Hardware requires resources and know-how to produce, especially for scientific instruments, and thus customers are usually willing to pay for a quality product from the original manufacturer even if it could be legally supplied by another company. Suppliers of optomechanical parts and even companies that sell microscopy hardware are starting to provide detailed design files for easier adaptation19, recognizing that this improves their product and is good for business. The possibility of building an active community of users and developers holds great potential for long-term customer relations.
Most journals provide a “conflict of interest” section for authors to disclose financial relationships relevant to their work. It is commonly accepted that hardware designs may not be shared due to patent and licensing concerns, but this is often not stated explicitly and so is not considered in the review process. Failing to disclose full hardware designs is detrimental to the reproducibility and credibility of an experiment, and a paper describing a novel, proprietary instrument blurs the distinction between advertising and research—a factor that we argue should be noted by editors and reviewers when deciding to accept such a work and declared as a conflict of interest. It is also important that the authors familiarize themselves with the rules of their funders and institutions. Requiring a statement justifying why designs are not shared fully, as is already done for code and data, would provide an incentive for researchers and universities to decide to either patent or release instrument designs, rather than keeping them a secret.
Our primary recommendation for researchers sharing a new hardware design is to include both the original, native CAD file and all appropriate transfer formats, drawings and bill of materials information in a permanent archive associated with the publication (via DOIs, university data archives or other stable repositories). Detailed guidance on structuring and sharing a hardware project are available from Open Hardware Makers20 or the Open Source Hardware Association (OSHWA)4. As an example, established open projects such as OpenSPIM21, UC222, OpenFlexure23 and MesoSPIM24 aim for documentation that links STL files, native CAD files and assembly tutorials to maintain easy replication. One can publish a design on available platforms, including accessible wiki solutions25, hardware-specific platforms that include viewers for common formats26,27,28 and software-focused platforms29,30 allowing extensive, custom automation31. A well-formulated list of metadata to include is provided by the Open KnowHow specification32. Most of these platforms do not offer permanent archival or DOIs, so we recommend archiving a snapshot of the design with Zenodo or another permanent repository.
When building new experimental setups, we follow design process. Computer-aided design (CAD) allows better and faster engineering, and can make the design of experimental setups open, reproducible and adaptable. Making these files available will lead to faster, more consistent and more reproducible biological experiments. However, nowadays, although CAD is an essential tool for designing and presenting new hardware, not every researcher fully utilizes these benefits due to lack of training, complexity of tools or absence of guidelines. In the case of published works primarily describing a novel instrument design, the time is ripe to establish a culture of best practice to improve the reproducibility of such work. We see how reproducibility and openness create new scientific communities that work together, so sharing and documentation of designs ensure that projects stay alive in the long term, even if the original creators are no longer involved, and offer the advantage of decentralized data collection and evaluation.
There are technical barriers to sharing designs fully. CAD files are usually either proprietary, restricting their usefulness to researchers with access to specific commercial packages, or incomplete, describing a final shape but not the design constraints required to edit them. The goal of fully interoperable CAD files is a long way off, and while promising open software exists, it is unlikely to replace proprietary systems soon. However, sharing both the native files and the appropriate transfer formats and documentation is a reasonable solution that can be implemented immediately.
Organizational barriers and deterrents are more difficult to mitigate. University policies on intellectual property are at odds with the principles of open science, often resulting in valuable designs being neither shared nor commercialized. We argue that treating hardware designs in the same manner as software or data, and specifically requiring an explanation in the paper if designs are not fully shared, is an important action that journals and funders can take to help drive a shift in culture and policy.
Even small changes made by the scientific community can realize the benefits of sharing CAD files in a manner compatible with open science. Editors and reviewers should scrutinize works that claim to share designs but omit crucial files or, even better, should enforce proper file-sharing and policies, as is already done for data. Researchers can use guidelines like OSHWA’s4 and workshops like the Open Hardware Makers20 to document hardware effectively. Ultimately, it is essential that global standards for CAD be developed, and become part of good scientific practice, so that curating design files comes to be considered just as important for the reproducibility of experiments as documenting biological protocols. International frameworks should better document how to use existing repositories, where essential design files can be stored for longer than a grant lifetime. Ideally, global funding schemes should create a basis for the development of open-source yet professional CAD packages that allow scientists to share reproducible results regardless of restrictive licenses or financial situations—because we believe that the community of researchers, journals and grant bodies should start treating design files as research data.
Data presented in Fig. 2 are available as Source Data and at the GitHub repository https://github.com/HohlbeinLab/OpenMicroscopy (https://doi.org/10.5281/zenodo.6406820). Source data are provided with this paper.
Center for Open Science. TOP guidelines (accessed 9 September 2021); https://www.cos.io/initiatives/top-guidelines
European Union. Kamraro (2014) open science (open access) (accessed 9 September 2021); in Horizon 2020 https://ec.europa.eu/programmes/horizon2020/en/h2020-section/open-science-open-access
UNESCO. Open science (accessed September 2021); https://en.unesco.org/science-sustainable-future/open-science
Open Source Hardware Association. Best practices for open source hardware 1.0 (accessed September 2021); https://www.oshwa.org/sharing-best-practices/
DIN SPEC Consortium. DIN SPEC 3105-1:2020—open source hardware—Teil 1: Requirements for technical documentation; text in English (accessed 9 September 2021); https://webstore.ansi.org/standards/din/dinspec31052020
Baden, T. et al. PLoS Biol. 13, e1002086 (2015).
Article Google Scholar
Stirling, J. & Bowman, R. Design J. 24, 299–314 (2021).
Article Google Scholar
Parker, A. & Novak, A. Building blocks for better science: case studies in low-cost and open tools for science (Wilson Center, 2020); https://www.wilsoncenter.org/publication/building-blocks-better-science-case-studies-low-cost-and-open-tools-science
Ravindran, S. Nature 587, 509–511 (2020).
CAS Article Google Scholar
Hohlbein, J. et al. HohlbeinLab/OpenMicroscopy: V01_release (2022); https://github.com/HohlbeinLab/OpenMicroscopy
US Library of Congress. STEP-file, ISO 10303-21 (2017); https://www.loc.gov/preservation/digital/formats/fdd/fdd000448.shtml
FX64. FX64 LambdaSpect (accessed 9 September 2021); https://www.fx64online.com/light-simulation
3DOptix. 3DOptix free optical simulation & design software (accessed 9 September 2021); https://3doptix.com/
Zemax (accessed 14 December 2021); https://www.zemax.com/
Fabrizio, K. & Di Minin, A. Res. Policy 30, 30 (2008).
Google Scholar
European Commission. Study about the impact of open source software and hardware on technological independence, competitiveness and innovation in the EU economy (accessed December 2021); https://digital-strategy.ec.europa.eu/en/library/study-about-impact-open-source-software-and-hardware-technological-independence-competitiveness-and
Opentrons (accessed 9 September 2021); https://opentrons.com/
Prusa Printers. 3D models database: world map of 3D printer users (accessed 9 September 2021); https://www.prusaprinters.org/
Selchow, O. Zeiss lightsheet sample holder standard L-shaped (accessed 9 September 2021); https://www.thingiverse.com/thing:4171299
Open Hardware Maker (accessed September 2021); https://ohwmakers.netlify.app//
Pitrone, P. G. et al. Nat. Methods 10, 598–599 (2013).
CAS Article Google Scholar
OSA | UC2—a 3D-printed general-purpose optical toolbox for microscopic imaging (accessed 9 September 2021); https://www.osapublishing.org/abstract.cfm?uri=isa-2019-ITh3B.5
Stirling, J. et al. The OpenFlexure Project. The technical challenges of co-developing a microscope in the UK and Tanzania. in 2020 IEEE Global Humanitarian Technology Conference (GHTC) (IEEE, 2021).
Voigt, F. F. et al. Nat. Methods 16, 1105–1108 (2019).
CAS Article Google Scholar
Appropedia (accessed 9 September 2021); https://www.appropedia.org/Welcome_to_Appropedia
GrabCAD (accessed 14 December 2021); https://grabcad.com/
Thingiverse.com (accessed 14 December 2021); https://www.thingiverse.com/
US National Institutes of Health. NIH 3D print exchange (accessed 9 September 2021); https://3dprint.nih.gov/
GitHub (accessed 14 December 2021); https://github.com/
GitLab (accessed 14 December 2021); https://about.gitlab.com/
Stirling, J., Bumke, K., Collins, J., Dhokia, V. & Bowman, R. HardOps: Utilising the software development toolchain for hardware design. Preprint at TechRxiv https://doi.org/10.36227/techrxiv.15052848.v1 (2021).
Internet of Production Alliance. Open know-how (accessed 14 December 2021); https://www.internetofproduction.org/open-know-how
Download references
The authors thank for comments Pete Dahlberg, Davis Perez, Jim Haseloff, Kay Oliver Schink and Scott E. Fraser. M.C.M. acknowledges a Marie Skłodowska-Curie fellowship (MSCA-IF-EF-ST grant agreement no. 842893). N.V. is supported by URPP Adaptive Brain Circuits in Development and Learning (AdaBD). R.W.B. and J.S. acknowledge financial support from the Royal Society (URFR1180153, RGFEA181034) and EPSRC (EP/R013969/1, EP/R011443/1).
Leibniz Institute of Photonic Technology, Jena, Germany
Benedict Diederich
School of Physics and Astronomy, University of Glasgow, Glasgow, UK
Caroline Müllenbroich
Brain Research Institute, University of Zürich, Zürich, Switzerland
Nikita Vladimirov
Department of Physics, University of Bath, Bath, UK
Richard Bowman & Julian Stirling
School of Biomolecular and Biomedical Science, University College Dublin, Dublin, Ireland
Emmanuel G. Reynaud
California Institute of Technology, Pasadena, CA, USA
Andrey Andreev
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
All authors contributed equally.
Correspondence to Emmanuel G. Reynaud.
The authors declare no competing interests.
Nature Methods thanks Alfred Millett-Sikking and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.
Statistical data presented in Fig. 2
Reprints and Permissions
Diederich, B., Müllenbroich, C., Vladimirov, N. et al. CAD we share? Publishing reproducible microscope hardware. Nat Methods (2022). https://doi.org/10.1038/s41592-022-01484-5
Download citation
Published: 04 May 2022
DOI: https://doi.org/10.1038/s41592-022-01484-5
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
You have full access to this article via your institution.
Advertisement
Advanced search
© 2022 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.
