Molecular visualization is fundamental in the current scientific literature, textbooks and dissemination materials. It provides an essential support for presenting results, reasoning on and formulating hypotheses related to molecular structure. Tools for visual exploration of structural data have become easily accessible on a broad variety of platforms thanks to advanced software tools that render a great service to the scientific community. These tools are often developed across disciplines bridging computer science, biology and chemistry. This mini-review was written as a short and compact overview for scientists who need to visualize protein structures and want to make an informed decision which tool they should use. Here, we first describe a few ‘Swiss Army knives’ geared towards protein visualization for everyday use with an existing large user base, then focus on more specialized tools for peculiar needs that are not yet as broadly known. Our selection is by no means exhaustive, but reflects a diverse snapshot of scenarios that we consider informative for the reader. We end with an account of future trends and perspectives.
Many parts of science rely on a visualization-driven cycle of experimentation, reasoning, conjecture and validation, even more so in relation with structural biology and biophysics. Molecular visualization  in particular is now broadly used in many contexts, with the purpose of illustration in the scientific literature or the aim to gain insight about primary research data. A broad interest in and need for these methods exists for many decades as described in . A key challenge is to pass information from intrinsically three-dimensional objects onto a two-dimensional support such as a sheet of paper or more commonly computer displays. Significant contributions originate in the computer science field of computer graphics. These contributions may only slowly transfer to the field of structural biology  as this process requires available end-user oriented software tools for efficient dissemination. Such tools are at the core of this survey. The survey was written as a short and compact overview for scientists who need to visualize protein structures and want to make an informed decision which tool they should use.
Tools for visualization of macromolecular structures have emerged from the longstanding need for molecular graphics and are accessible to scientists with broadly varying backgrounds as previously reviewed in . In this mini-review, we focus on a few essential and generic tools along with some novelties and more singular solutions for specific needs. We will not focus on commercial closed source software, although there are definitely tools available, amongst which we may mention Yasara  and Maestro , both freely accessible. Commercial tools often integrate or interface with modeling features and may preferably be used in industry settings or for occasional lay usage because of their user guidance, support and commercial ecosystem.
The scope of applications we target with this mini-review ranges from very generic needs to specific usages. For instance, a first generic application concerns the illustration of scientific publications and the scrutiny of hypotheses , as for example relating mutational data to structural representations . A further level of application may involve a more in-depth visual analysis of macromolecular structures and their properties , possibly related to the (spatial) distribution of charges, electrostatic properties, pockets and surface complementarities [4,9,10]. An even more specialized usage may refer to depicting and analyzing data from theoretical chemistry , computational biology  and bioinformatics approaches . The need to analyze increasingly complex molecular dynamics simulations is one prominent example that has driven the field forward. This usage naturally leads into the field of data analytics and in particular visual analytics [12–14], which will be only briefly examined in this review.
‘Swiss army knives’ for general needs
A few molecular visualization packages have been used by the community for many decades and provide robust visualization capacities suitable for a large audience. Among these tools, we will briefly discuss Chimera (latest version called ChimeraX), Jmol (and more recently JSmol), PyMol and VMD, which all serve a large user base. Such packages appear as a safe choice for visualization-based projects, with a stable codebase and API to interact with. Their usability has been refined over time, and although not necessarily easy to use for newcomers, they are well adapted to the tasks related to molecular visualization. Each tool may feature many functionalities and provide tutorials, scripts and extensions that enrich the user's experience. Table 1 provides the first overview of this core set of software tools, whereas Figure 1 provides a pictorial comparison of a typical work session and each software's user interface.
|Features\Tools .||VMD .||Pymol .||Chimera .||Jmol/JSmol .|
|Raytracing/Illustrations||✓||✓||External POVRay||External POVRay|
|Features\Tools .||VMD .||Pymol .||Chimera .||Jmol/JSmol .|
|Raytracing/Illustrations||✓||✓||External POVRay||External POVRay|
The Chimera software  dates back to 1976, then developed under the name Midas, and provides functionalities for the visual exploration and analysis of molecular structures. Related descriptive data can be examined, in particular, cryo-EM datasets with density maps, molecular dynamics trajectories and more. Creating animations is made easy by using keyframes with automatic transitions between them and the repertoire of visual depictions is rich.
Jmol was started ca. 1999 [16,17]. It is a generic platform usable either as a standalone viewer or within a web context. Many educational applications feature Jmol for structure depiction as it is versatile and easy to embed in courseware or to use as graphical frontend for the exploration of structural databases.
PyMol  was launched at the end of 1999 and, as its name suggests, is built around a python scripting environment. This tool is very popular with experimentalists as it provides useful features for crystallographic and NMR-derived structures. It's greatest user base may relate to pharmacology and drug design. PyMol provides convenient molecular editing and atom selection functionalities.
Since its inception in 1993, the VMD software focused on the graphical analysis of molecular dynamics data, hence its acronym Visual Molecular Dynamics. It manages even quite large systems smoothly and provides many extensions, in particular for advanced visual analysis, through plugins.
An important feature common to all these packages is the possibility to automate and script tasks for easy re-use and batch deployment.
Despite their long history, all packages still provide regularly novel functionalities and adapt to the progressing hardware technologies, for instance in the graphics card market. Sometimes such extensions may require an extensive re-design of the underlying code as has been documented for instance for ChimeraX . Such extensions may address more specific needs and usages adapting to an ever-evolving field as described in the subsequent section.
Emerging needs for molecular visualization
Using molecular visualization as a ‘service’ in a broad sense is one of the more recent evolutions. This concept may relate to web integration, interactive visual notebooks such as with Jupyter [20,21] and web-based exploration with novel technologies such as accelerated graphics rendering through WebGL. The NGLviewer  and more recently Mol* viewer (molstar.org)  address many of these particular aspects. Mol*, for instance, is now implemented as the default viewer on the PDBe web pages. JSmol and EzMol  provide easy web integration as well.
To generate illustrative and compelling images for publications is another widespread and very common need reviewed by Goodsell and Jenkinson . David Goodsell has provided for many years the captivating images of the molecule of the month , which can now be reproduced by everyone with the Illustrate software . Previously, the introduction of ambient occlusion lighting with the Qutemol software  had already led to a significant improvement in the depth perception from static illustrative images. The possibility to continuously vary the abstraction level added further possibilities . For particularly eye-catching results, tools from the cinema and animation industry can be used as well [32–36]. Such publication-quality images can be produced with the majority of packages described here, either directly within the package or through a subsequent ray-tracing step.
Interactive raytracing [37,38] is needed for the highest quality and most realistic images produced almost on the fly and to be explored interactively providing a convincing illusion of the reality of the raytraced objects. Real-time, often GPU-powered, path-tracing approaches become more widely accessible, for instance through the Unity game engine. Alternative approaches attempt to produce visually similar results .
The visual exploration of specific data sets represents an important aspect of protein visualization. Historically this aspect is illustrated by the need to visualize experimental electron density maps in crystallography, spawning many tools, among which O [40,41], Coot , CCP4mg  and uglymol  just to mention a few. More generally the advanced exploration of specific experimental data may represent a specific feature of some packages, such as Chimera, with a focus on CryoEM data. Similarly, through its tight integration with the protein data bank, Mol* provides many useful features for crystallographic data.
The integrative analysis of data spawning different data spaces is a particular challenge in the current era of integrative and computational biology. This task is nicely illustrated by the visual analytics approach implemented in the Aquaria software [45,46] enabling a user to link the evolutionary dimension of sequence-space with structure space. The MolArt structure annotation and visualization tool pursues similar avenues . Molecular dynamics simulations — and more generally computational simulation approaches including normal modes, Poisson–Boltzmann electrostatics, finite element and elastic network models — represent another source of specific data to be visualized. Historically, the VMD package was the first to focus on molecular simulations, but many others now provide such possibilities, most recently, for instance, MDsrv  or HTMoL  on the web  and Unitymol in virtual reality (VR).
Such VR exploration  of protein structures addresses the need to render the complex spatial features of macromolecules very efficiently in immersive 3D thereby helping shape perception and benefiting from advanced interaction devices. Web delivery of VR contents is possible through WebVR. Several packages have reported VR capacities such as Molecular Rift , ChimeraX , UnityMol [25,51], Narupa , Nanome  and CootVR. WebVR-capable molecular viewers are less widespread, possibly due to the currently limited browser-support for this technology. VRMol  is a complete web-based VR tool available at vrmol.net which has recently been released. UnityMol experiments with WebVR have been reported as well .
Virtual reality was mentioned in the previous section, featuring currently available software tool implementations for protein visualization. In the future, much is to be expected from augmented reality (AR) headsets or glasses, which promise a great potential for multi-user interactions without isolating the scientists in a virtual world. These collaborative and social aspects are important for efficient scientific exchange as well as to lower the barrier to use such novel technology. AR is currently limited by three main barriers related to the available hardware. First of all, the characteristics of the devices need to improve for instance in terms of field of view, second, better graphics performance and processing power are required when they come embedded in the device such as is, for instance, the case with the 1st generation of the HoloLens , and finally a good human–computer interaction interface with the device needs to be available. Current interaction paradigms through finger gestures and/or voice recognition are a useful step to get a glimpse of the possibilities. Together with several laboratories we carried out extensive tests and came to the conclusion that the current paradigms are in the long run not satisfactory for a continued regular usage of these devices as their current error rates build up user frustration rather quickly.
Another trend lies in social features relating to annotating protein visualizations, sharing them, collaborating with other scientists and disseminating findings community-wide with technologies akin to current social network tools. The jolecule tool (jolecule.com) introduced easily shareable annotated views using a web-based approach. Mol* also supports shareable remote states. To some extent, the collaborative social features may be targeted by current VR approaches, such as [50,53,55], whereas foundations were laid a long time ago [56–59].
A particular challenge for future molecular visualization will be to embrace the next scale of huge biomacromolecular assemblies and models. With experimental developments such as the resolution revolution of CryoEM  and the rise of integrative methods , the foreseeable future is in the representation of not only a single small protein but the representation of full organelles. The molecular modeling field significantly contributes to this expansion of the scale of biological objects that can be investigated down to (quasi)-atomistic resolutions [62–64]. Exploration of such complex models would definitely benefit from immersive technologies to better comprehend the global molecular system, which is not possible to apprehend on a standard 2D display. This task may also necessitate new ways to interact with such complex systems. This interaction will be possible by using technologies such as virtual reality CAVEs, sophisticated stereoscopic display walls or possibly future generations of VR gear. Current limitations for VR hardware relate to limited display resolution for crowded information-dense scenes and delivering a high enough rendering bandwidth for fully fluid exploration at such high resolutions. At the same time, these objects and simulations will even further increase the big amount of data to analyze, and immersive visual analytics approaches may contribute significantly to our ability to analyze such data sets and identify non-obvious complex patterns therein. Better tools to generate the analysis data, for instance through advanced analysis of molecular dynamics trajectories, may lead the way into this data-driven future. Some tools already exist [65,66] and the constitution of intuitively accessible databases including access to the raw data and visualizations thereof may be a way to delve into the wealth of information generated by such approaches. Several of these databases already exist, such as [67–69], to mention only a few, however, interactive visual analysis (and visual analytics in particular) is currently rather limited.
Whereas the overwhelming part of protein visualization so far may have focused on displaying existing models, the future may focus more intensely on methods to visually create complex molecular landscapes from scratch. Novel technology may be instrumental to this task [70–73].
A final general and more philosophical comment about the progress of molecular graphics will conclude our collection of thoughts on future trends and perspectives. There still exists a huge potential for substantial improvements of visualization methods by better integration of different scientific communities, in particular computer scientists involved in the field of computer graphics and biologists with visualization needs. Much untapped potential can be found in exciting new methodologies developed in computer science that do not or only slowly transfer into tools accessible to structural biologists as described in more detail in . Among the upcoming new challenges in visualization that would benefit from such integration, we may mention for instance the depiction of uncertainty, see  for a recent exploration of this topic, or how to capture the dynamics governing macromolecules in static pictures (briefly discussed in  as well).
Protein visualization is nowadays a vibrant and striving field, with an extraordinary longevity since the first drawings by hand that date back before the 1960s. It has reached a broad audience in diverse subfields of modern biology and is ubiquitous on the web, in publications and on a broad variety of display devices used by researchers, from mobile phones to 3D theaters. This success has been driven by the availability of high-quality software tools, often developed and refined collectively by the scientific community. Beyond the classical axiom: ‘bigger (structure), larger (dataset), faster (visualization)’, future new developments related to molecular visualization of protein structures will serve our increasing need to integrate heterogeneous data and to efficiently use new hardware devices (from VR/AR headsets to smartphones).
Software tools enable protein visualization, an essential part of scientific reasoning.
Knowing such tools and their range of applicability is key to efficiently coping with molecular structures and gaining relevant insights. Ongoing revolutions in technology such as augmented and VR will drive even further the possibilities and level of visualizations that are possible.
Collaborative aspects may be more natively integrated in visualization tools as heralded by trends in multi-user VR applications. The complexity of visualizations is increasing and new concepts may require new visual metaphors; increasing the transfer of visualization methods from computer science to structural biology bears great potential to address these.
Conflict of Interest
All three co-authors are actively involved in the development of the UnityMol visualization software that is featured among other tools in this mini-review.
This work was supported by the ‘Initiative d'Excellence’ program from the French State (grant ‘DYNAMO’, ANR-11-LABX-0011 and grant ‘CACSICE’, ANR-11-EQPX-0008) and received funding from the ANR-16-CE05-0024 ENZYMOR project. X.M. and M.B. thank Sesame Ile-de-France for co-funding the display wall used to assess several software packages described in this review. M.C. acknowledges support from CNRS-MITI grants PEPS MPI 2018 and ‘Modélisation du vivant’ 2019. X.M. and M.B. thank UCB Biopharma for support.