Molecular Interactions Using New Technology: A Virtual Reality Gaming Platform to Visualize and Manipulate Molecules

The representation of complex biomolecular structures and interactions is a difficult challenge across life sciences. Researchers and students use unintuitive 2D representations to gain an intuitive understanding of 3D space and molecular interactions. Since this is cumbersome for complex structures, such as protein-ligand interactions, several solutions have been proposed to help elucidate the 3D space. However, these representations are often static or do not fully leverage the interactivity that modern computing systems can provide. Our solution, Molecular Interaction using New Technology (MINT), is the first “gaming” platform to effectively represent and manipulate structures in 3D space using virtual reality while simultaneously scoring biomolecular interactions in real-time. Utilizing this combination of manipulation and real-time feedback, MINT provides scientists with an intuitive and effective method for drug discovery. We hope the combination of an intuitive interface with a powerful chemistry backend will expand molecular understanding and drug discovery for scientists and non-scientists. Abstract The representation of complex biomolecular structures and interactions is a difficult challenge across life sciences. Researchers and students use unintuitive 2D representations to gain an intuitive understanding of 3D space and molecular interactions. Since this is cumbersome for complex structures, such as protein-ligand interactions, several solutions have been proposed to help elucidate the 3D space. However, these representations are often static or do not fully leverage the interactivity that modern computing systems can provide. Our solution, Molecular Interaction using New Technology (MINT), is the first “gaming” platform to effectively represent and manipulate structures in 3D space using virtual reality while simultaneously scoring biomolecular interactions in real-time. Utilizing this combination of manipulation and real-time feedback, MINT provides scientists with an intuitive and effective method for drug discovery. We hope the combination of an intuitive interface with a powerful chemistry backend will expand molecular understanding and drug discovery for scientists and non-scientists.


Introduction
In 1966, Cyrus Levinthal published "Molecular Model-building by Computer" along with the first interactive display for visualization and manipulation of molecular structures 1 that revolutionized the field of molecular visualization. Before the introduction of more advanced virtual systems, physical ball and stick models were developed and used by several scientists, such as Watson and Crick to investigate the structure of DNA and John Kendrew to solve the first crystal structure of protein 2 . As the capabilities for creating virtual 3D environments advanced, the use of physical models decreased, and the physical ball and stick models have been replaced by a mouse, keyboard, and computer monitor. The current way of chemical structure modification along with 3D position and orientation involves the scripting of computer programs or the use of complex graphical user interfaces where inputs are given by a mouse and keyboard. [3][4][5][6][7][8] Manipulation of molecules through such methods is complicated as it often requires extensive knowledge of programming or a deep understanding of the user interface. Such systems lag behind the state-of-the-art tools developed for human-computer interaction. Relatively simple actions such as repositioning molecules are only mastered after a steep learning curve as the many nuances required are difficult to understand for both professional researchers and students 9 .
Although the current software packages afford a great deal of flexibility in representation and visualization styles, they lack intuitive manipulations because of their reliance on a mouse and keyboard. The proper representation of biomolecules in a 3D space is crucial to the understanding of various intermolecular interactions. 10 However, 2D displays can misrepresent the understanding of such interactions in 3D and the steep learning curve to manipulate structures is a bottleneck for widespread use of tools beyond scientists 11 . Conversely, physical models avoid these pitfalls by offering an environment where users can manipulate objects intuitively 12 whereby anyone can manipulate physical model by bending angles, breaking bonds, adding new atoms and functional groups, and changing positions of multiple atoms by rotation around one or more bonds. However, the physical models for larger complexes are expensive to make, hard to maintain, and lack real-time feedback to understand molecular interactions. We believe that the intuitive nature of physical models needs to be incorporated into in silico 3D modeling software; a feature that can be accomplished with the application of Virtual Reality (VR) hardware.
Currently, there are numerous platforms offering visualization and manipulation of molecular structures [6][7][8] , and several more that have the capability to visualize molecules within a VR environment [13][14][15][16] . Noteworthy examples include Molecule Viewer 17 which allows for protein visualization and UnityMol 18 which provides an immersive environment for exploring molecules. Neither platform allow the user to edit and manipulate the chemical environment, a shortcoming addressed by Nanome, a collaborative VR environment implemented with a wide variety of molecular manipulation functionalities, and ChimeraX VR 19 , an application utilizing UCSF Chimera and the SteamVR toolkit to enable molecular data analysis and manipulation through simple controller-based input commonly seen in VR applications. While these new tools offer visualization and manipulation capabilities, they are unable to provide insight into the underlying chemical significance of these interactions. Simply porting this functionality from conventional 2D molecular visualization systems, such as PyMol or Jmol, into VR does not exploit the full potential of the new technology for learning and adoption by scientists and layman alike. Additionally, these tools lack the collaborative nature and scalability to be effectively applied in the classroom. The motivation for utilizing virtual reality for molecule visualization over conventional 2D displays lies in the inherent intuitiveness and 3D nature of virtual reality, which in turn promotes interaction with the molecule. We believe that interactions between the user and the molecular structure via a feedback-driven system is a key aspect of molecular visualization because the synergy of these features empowers a viewer's own natural curiosity to further explore, study, and research chemistry and biochemistry in a unique and rewarding manner.
To address the lack of chemical insight provided by current VR implementations, we have developed the Molecular Interactions using New Technology (MINT), a virtual reality biomolecular visualization platform. Our implementation serves to lift biomolecular visualization to the forefront of the technological frontier and foster a mainstream understanding of the biomolecular research that accompanies drug discovery. We seek to provide an easy-to-use, intuitive, and powerful platform to simultaneously visualize and manipulate molecular structures, allowing any user, regardless of scientific training, to optimize molecular structures and receive real-time visual feedback through MINT's comprehensive virtual toolkit. MINT introduces additional new features not present in the visualization platforms mentioned previously: (1) it is integrated with a backend computational chemistry platform our lab to efficiently compute scoring functions to monitor how manipulation behaviors change the chemical environment, (2) it 'gamifies' the process of molecular optimization, fostering a playful relationship between user and molecule as well as competition between users for the creation of optimal structures, and (3) it is scalable across multiple devices from smartphones to workstations. Herein, we present MINT's features that allow for intuitive manipulation and visualization of molecules, followed by a discussion on how these features lead to gamification and the creation of a platform intended for the instruction of chemistry.

Results and Discussion
MINT provides an intuitive interface to chemistry MINT utilizes intuitive controls and an immersive environment to allow for a unique visualization and manipulation environment. MINT's workflow is a 4-step procedure consisting of input, visualization, manipulation and output (Figure 1). MINT starts by interpreting binary files that contain molecule structure information (Figure 1A). With the molecular structures obtained from the input files, MINT generates and displays a 3D model and presents this model through a VR headset such as the HTC Vive. The 3D models are fully interactable, allowing users to reposition and manipulate the entire molecule by working with a menu interface consisting of three different panels: Manipulation, Visualization, and Utilities. This interface groups the essential elements found on many conventional molecule editors into one simplistic and VR centric format (Figure 2A).
The Chemical Algorithms for Network-based Decisions on Interactions for modeling reactivitY (CANDIY) software suite is integrated into MINT to bridge the manipulations performed by the user and the underlying chemistry of the VR representation. CANDIY aids the scientific community in their efforts to model how molecules interact with their environment by providing a platform for the development of algorithms and procedures tailored for specific purposes. Role of MINT in this software suite is to allow for user manipulation in an intuitive manner, opening the use of this software to the layman. After each manipulation performed by the user, CANDIY validates the user input and provides feedback through a combination of haptic and visual interfaces. This creates a relationship which ensures the chemical legitimacy of each operation without the need to instruct the user in advanced scientific concepts. Currently, the CANDOCK 20 package is the most integrated into MINT, but we plan on integrating other packages and machine learning methods that we are developing, such as our biomolecular structure searching software, Lemon, in the near future 21 .
CANDIY provides the ability to interpret 3D coordinates and molecular topology obtained from molecular file formats 22 (Figure 1A). The molecular information is passed onto MINT for visualization, where it is processed and rendered. CANDIY calculates the interaction between biomolecules, such as a ligand and a protein, in real-time by using a generalized statistical potential function 23 . When a user manipulates the ligand or protein in the VR environment, the changes in the 3D conformation of the molecule and protein are communicated to CANDIY which in turn returns a numeric score to the user. This process is key to the gamification concepts presented later in this work.

MINT provides multiple visualization and manipulation modes
The visualization of molecular structures in 3D is a necessary component of the MINT workflow. MINT's VR interface embraces a range of visualization techniques to improve the understanding of a 3D environment, e.g., the binding of a drug to a protein, in a versatile and robust way (Figure 3). Protein structures can be rendered via a surface model with dynamic lighting and shadow effects ( Figure 3A) where different atom types are represented by different colors on the surface. In MINT, this is the default rendering mode for large molecules due to its intuitive demonstration of a molecule's size and spatial information. Figures 3B, 3C and 3D show a ligand structure rendered in various forms and forge the basis of most molecular manipulations performed by the user. Figures 3E and 3F depict protein structure in ribbon form and a specialized rendering of the backbone, respectively. Both visualization options allow the user to develop a more holistic comprehension of the biomolecular structure. The user can dynamically tailor the virtualization using visualization panel, giving them the freedom to mix and match different options to create unique and complex visualizations. Figure 4 shows a comparison between a visualization provided by PyMol, a standard molecule visualization program, and a visualization provided by MINT. While both programs offer specular textured surfaces and ball and stick representations, MINT does not require complex scripting like PyMol to represent the binding site tunnel. Instead, MINT helps the user achieve these actions via the VR interface. MINT allows its users to perform several different manipulations using the toolkit depicted in Figure 5. By linking these simple manipulations together, users can quickly perform complex maneuvers in a short time as compared to traditional methods of interaction, such as scripting or 2D graphical user interfaces.
• Hand tool ( Figure 5A), maneuver the ligand as if it was a rigid object.
• Bond rotate tool (Figure 5B), rotate a portion of the molecule via an axis of rotation. To define this axis, the user selects two atoms to create a vector pointing from the first atom to the second). This movement is directly inspired by physical models which allowed for different configurations to be created by quick twists and turns. • Bond tool (Figure 5C), make and break bonds by clicking on two different atoms simultaneously. This action is monitored by the backend program CANDIY that prevents the creation of invalid molecules. • Selection tool (Figure 5D), select specific atoms to manipulate instead of working with the whole entity. • Surface trekking ( Figure 5E), a quick way to navigate in the environment by walking on a surface model. • By linking these simple manipulations together, users can quickly perform complex maneuvers in an exponentially shorter time than traditional methods of interaction like scripting or 2D graphical user interfaces.
To illustrate the MINT workflow, we have detailed out each step of the workflow for PDBID 4XUF in Figure 6. MINT begins by interpreting the PDB input file ( Figure 6A) and converts the textual atomic records into atom data arrays that form a complete representation of the molecule in working memory ( Figure 6B). MINT produces an intuitive VR visualization using these coordinates which the user can interact with to optimize the docking score of the ligand ( Figure 6C). Once a user has performed a manipulation on the ligand such as rotation or translation, the change is reflected in working memory ( Figure 6B). Finally, MINT outputs the modified data as a PDB file that can be used in other applications or reopened in MINT for further analysis ( Figure 6E).
The driving force of such workflow is MINT's ability to consolidate every manipulation made by the user into a numeric 'score' which represents the chemical validity of these actions. A detailed description of this score is given in the section entitled Scoring of Player's ligand conformations. In Figure 7, PDB 4XUF, a protein-ligand complex, has a score of 331 in its initial state ( Figure 7A). The user then performs a bond rotation on the ligand structure through the VR controllers and interface ( Figure 7B). This action results in the score increasing to 333 in real-time, and indication that the user has improved the potential effectiveness of the ligand towards the target protein.

Gamification of molecular interactions is an integral component in MINT
The influence of video games on contemporary culture is immeasurable and the practice of utilizing factors that involve game mechanics like challenges, tasks, and levels into the design of non-game consumer software has surged in recent years 24,25 . The notion of a 'serious game' 26 , for example, is a practice parallel to gamification and is often categorized by its emphasis on training the player for a specific real-world task or completion of non-entertainment objectives through specially oriented gameplay. The incentive of incorporating gamification into non-game software amplifies the user's engagement with the experience and stimulates motivation and curiosity to further facilitate accomplishing an objective regardless of whether it is learning, training or simulation. The benefits of gamification have been explored in many studies [27][28][29][30] .
One excellent example that combines biochemistry, protein folding, and gamification is Foldit 31 , a platform which presents a multiplayer puzzle game to help solve protein folding questions. This 'game' takes each protein structure as a challenge or a level for the player to conquer by using the intuitive folding tools provided by the application and leverages the crowdsourcing nature of gameplay to unite all players and further facilitate biochemistry experimentation and research. The Foldit player base has achieved remarkable accomplishments including helping decipher the crystal structure of a monomeric retroviral protease linked to HIV/AIDS 32 . In a similar manner, we aim to incorporate gamification elements like player collaboration/competition, challenges, scores, and a playful user interface into the design of MINT and to excel at being intuitive and engaging with the help of VR.
The scoring feature in MINT (Figure 7) functions as an indicator of the interaction energy between two structures such as a protein receptor and a small molecule ligand, or as a method for players to selfvalidate their in-game actions. Given the importance of score for the 'gamification' of a given objective, we present this interaction score as the core mechanism for gamification in MINT. A receptor and ligand complex obtained from the protein databank (PDB) 33 is imported into MINT to produce a level or a quest, where players can compete against each other to find the optimal score (provided by CANDIY). To do so, the players must manipulate the conformation and topology of the ligand, yielding a drug discovery platform which is naturally crowdsourced. One can further extend this pipeline to rank scores obtained from different players on the same complex on a leaderboard in order to encourage competition between players. Such practice can be achieved through a backend server that collects players' gameplay data, providing an implementation for crowdsourced drug design.
A series of playful aesthetics are utilized by the MINT user interface to instill a game-like theme throughout the gameplay experience. For example, the coloration of each element in the program, including the menu interface (Figure 2A), the controllers ( Figure 2B) and the 3D models rendered in MINT (Figure 3), tend to fall on the brighter sides of the color spectrum, and are selected to have a high contrast with one another. The menu interface takes the form of a virtual clipboard that user can hold using the controller and the textual elements such as tooltips on the interface are pixelated and 2D image icons are used for representing each functional item on the menu. Furthermore, the toon shading technique 34 is used in the program for rendering a surface model of macromolecules to give them outline on the edges and produce simple lighting visual effects, yielding an environment which imitates a comic book drawing.
Haptic feedback 35 is the sensorial mechanism used to simulate a sense of touch and is used to convey the application of motion or forces, the difference between the weight of virtual objects, or the textural feeling of geometry or surface. HTC Vive's hand controller ( Figure 2B) has a built-in haptic feedback mechanism which vibrates to simulate a sense of weight and friction. MINT exploits this feature to make the overall user interface responsive and lively. The variation of vibration depends on both its duration and its strength and adjusting these two factors opens different dialogs with user: For example, clicking a button on the menu interface generates short and mild vibration that imitates the sense of pressing a mechanical button, while clashing a protein and a ligand by dragging them together produces long and strong vibration to indicates the physical collision of such a clash. Similarly, rotating angular bonds between atoms returns a consecutive and blunt vibration in short intervals on to resembles the sense of turning a crank.
Another important feature for immersive gameplay is the use of gesture-based and motion-based interaction, therefore a major component of MINT's manipulation system is performed using intuitive gestures and motions. For example, breaking a bond is performed by pulling two atoms apart from each other instead of having users simply click on two connected atoms with a computer mouse. This motion can be augmented with the gradation of vibration on Vive's hand controllers to express the energy cost associated with the operation. We have implemented these features in MINT to add complexity and a sense of skill to gameplay, ultimately contributing to the gamification of molecular manipulation.

Scalability for collaboration and education
While gamification is a defining difference between MINT and conventional molecular visualization and manipulation software, it is only one of the several factors that improve the scalability of MINT as a collaborative project over competitors. Education, research, and entertainment are the three pillars guiding the developmental roadmap of MINT and applying the molecular visualization and manipulation capability of MINT in education and research spans the gambit from classroom teaching to drug design prototyping. Since the number of active VR users worldwide is increasingly rapidly 36 , a large potential user base is anticipated to become participants in this project, benefiting drug design and discovery. Therefore, we want to catalyze the popularity of MINT by introducing mobile versions and multiplayer gameplay features. Due to our use of the Steam VR toolkit and the Unity3D engine, our visualization platform can run on multiple hardware platforms. Although we targeted the HTC Vive due to its superior support for human-computer interaction, the Oculus Rift is also a potential target for our platform as others have attempted to use this platform to target drug design 37 . However, we believe that our program is both more intuitive and scalable than these approaches due to the better human-computer interface offered by the Vive.
Molecular data representation in memory space is a crucial component of the processing workflow for scalability to create vivid and intuitive graphics in VR. To manage different sections of the workflow, MINT has a hierarchy of data classes and helper classes that are dedicated to representing and managing molecular data receiving from CANDIY. All the entries in the PDB file that describe atoms are used to generate atom data arrays, which are stored within the molecule data class. From a single molecule data object, various in-game representations can be generated (Figure 8). First, MINT's algorithm generates a molecule representation base from the atom data arrays, which connects the molecular data with its in-game representation, because a single molecule often possesses many different types of molecule representation forms. Next, MINT uses the marching cube algorithm 38 to generate a mesh that simulates the protein's surface and uses native features in the Unity game engine such as game object instantiation and line renderers to simulate atoms and bonds for ball-and-stick representation. The molecule representation base forms the basis for atom manipulation and interaction with the user by enabling collision detection with user's VR controller. Physical collision provides the player with useful feedback about the position and orientation of the molecule. The changes that the user makes on the molecular structure, such as transformation, making/breaking bonds and angular bond rotation, update the molecule and atom data which then go through CANDOCK for error checking, automatic optimization, and most importantly, validation of these VR operations to ensure the scientific accuracy. Finally, CANDIY updates the molecule data, which is then returned to the user through visualization via the molecule representation in Unity.
We have released a version of MINT on the Google Play Store that targets the Android Platform (MINT Mobile). This version is compatible with Google Cardboard, a low-cost head-mounted VR platform developed by Google for smartphones. Currently, the mobile version only supports molecular visualization and is equipped with a user interface that is tailored towards smartphones, taking into consideration of smartphone's limited computational power and the lack of physical controllers when compared to PC. In this version, the user can load molecule structures as different visualization on the fly and study the visualization using tools like surface trekking and camera orbiting. Additionally, MINT mobile includes the environment grid guide (Figure 9A), which is intended to help offer the reference of camera orientation and position in the virtual environment.
To enable collaboration of molecular exploration at real-time, we have developed a multiplayer version of MINT which allows for multiple users to cooperate in the same virtual space (Figure 9B). In the multiplayer mode, one user hosts a virtual environment using the HTC Vive headset and controllers, allowing them to manipulate and modify molecular structures. Other users can enter the hosted environment as guests through the use of the mobile version of MINT and spectate the host user's actions in real-time. Guest users can walk around in the virtual environment, observe the structures from different perspectives, and suggest manipulations to the host.

Conclusion
MINT is a VR platform that challenges the conventional molecular visualization and manipulation tools used in a 3D environment. Equipped with an intuitive interface and a variety of features in visualization, MINT brings ease of use and better comprehension to biochemistry research and study. Due to its use of CANDIY, our program pipeline is user-friendly because it allows for input and output compatibility with conventional chemical file formats and is responsive towards user actions while keeping a chemically accurate simulation. Users do not need to possess specialized programming or scripting knowledge to perform complex manipulations in MINT. Instead, a few quick movements with a VR controller can surpass what many lines of codes can do in other molecular visualization software and in an exponentially shorter time.
The concept of gamification is ingrained into every component of MINT, from the design of interaction between molecule and user, to the aesthetics of interface and the feedback of each action happened in VR. A fun and enjoyable user experience is born through the utilization of such elements and ultimately yields a more scalable software that can reach a broader audience through the implementation of an intuitive interface for molecular manipulations. These features along with tight integration of our platform with the CANDIY suite for evaluating the molecular interactions of small molecules, provide a unique functionality equivalent to traditional molecular visualization packages like PyMol and Jmol while offering a unique experience that is immersive and interactive due to the power of VR. MINT allows users to develop visual comprehension of molecular structures while making it easier to manipulate the structures in a short period of time.
Finally, MINT will be released as an open-source project which welcomes collaborative efforts from all members of the VR and chemistry community. Decades have passed since Levinthal's system was first introduced, revolutionizing the way we perceive the microscopic world and we hope MINT can be part of the next revolution by incorporating modern technology and other advancements previously unavailable.

Surface Generation
Surface generation utilizes a modified version of the marching cubes algorithm 38 specifically tailored towards Unity, which results in a continuous, dough-like surface. While this algorithm is especially useful for generating protein surfaces, it can be exceedingly costly as well. Limits upon the number of faces a procedural mesh can contain in Unity requires certain models to be generated and pieced together in smaller parts. Generation speed and stability of the surface depend heavily upon thresholds set by the player and the size of the datasets. To optimize performance, the existing serial density field code was modified to run in parallel on multiple threads. To accomplish this, a collection of work threads is generated and assigned a group of cells. This code is structured analogous to SIMD systems, as each cell's final density is independent of the surrounding cells.

Molecule Input/output
To generate and further manipulate a 3D molecular model, MINT requires preset data input that defines the molecule's structural formation. MINT's visualization and manipulation pipeline is focused on the atomic records and bonding records that exist in the PDB file. Each atomic record is a line of text that starts with the label "ATOM" or "HETATM", followed by the atom's index number, element type and other information along with its 3D coordinates. Some PDB files contain both receptor atomic records and ligand atomic records. MINT's interpretation of the input PDB file starts by delegating file reading to CANDIY. The various column-aligned parameters are read and interpreted by CANDIY according to standard PDB file protocol, which then transforms textural data into memory space and sends them back to MINT. After receiving the data returned by CANDIY, MINT puts them into data arrays in Molecule Data class which the Unity Engine can understand and further passes them down to the graphics rendering pipeline. The structural changes associated with this operation update the arrays in the memory space. Finally, the atom data arrays existing in the memory space are written out as a new PDB file with the modified atomic records reflecting the various manipulations that the user enacted. The new PDB file generated from MINT can be passed down to other visualization platforms to create display renderings of the new 3D molecule model that reflect the structural modifications from MINT or enter another round of MINT's input/output cycle to be further studied on in VR. In addition, MINT has a PDB-fetching VR panel that searches and fetches PDBs when users input the name of the PDB to the interface. MINT communicates with RCSB cloud backend to retrieve PDBs in-app so there is no need for the user to take off the VR headset and download resources manually.

VR Interaction
The hand tool is used to grab the ligand. In this mode, the ligand will follow the orientation and position of the player's hand in 3D space when the user activates the trigger button. Haptic feedback on the controllers is provided to give the player a feeling of weight and resistance, as well as a signal when ligand molecules are brushing against the protein's surface model. Players are also given the ability to freely rotate atoms along with their bonds by using the rotate tool, a method of interaction is directly inspired by molecular modeling kits. To perform this action, players must first grab an atom that they wish to act as an "anchor" with their offhand. Next, players use their primary hand to select an atom that determines their axis of rotation. This atom is referred to as the directional atom. An axis of rotation is defined from the vector pointing from the center of the anchor atom to the center of the directional atom. Several steps are required to calculate the angle of rotation. First, a perpendicular plane is formed using the vector between the two key atoms. Next, the position of the player's primary hand is then projected onto the plane. This point in space always lies upon the plane that is perpendicular to the axis of rotation. This new local space can be conceptualized as a 2D plane, where the (x,y) position of the two atoms are located precisely at (0,0). Every frame, an angle in degrees is calculated between the center of this space and the projected coordinate of the primary controller. The current angle is compared with that of the previous frame and a delta is calculated. The anchor atom is then rotated by this delta angle to produce a rotation that mimics the rotation of the hand around the axis of rotation. In addition, bonds between two atoms may be created or destroyed with the 'Make 'n Break' tool. This works very similarly to the rotate tool, by creating or destroying a bond once the user has selected two atoms.
Players are also allowed to scale themselves around the atom they are interacting with, enabling the player to resize themselves in the atomic world. Currently, the method of scaling requires the player to perform a "pinch-zoom" gesture with both controllers to allow for fine control of scale when the player's hands are spaced farther apart. In addition, the world is also simultaneously translated based on the vector between the two controllers. This is done to make the scale tool feel much more natural to use.
MINT also supports a trekking feature which allows players to navigate the surface of the protein molecule as if they were walking inside a cavern. Trekking is a special form of teleportation that allows the player to move to points of the surface while keeping the relative local space unaltered. When the trekking tool is enabled, the player points their offhand at any point along the surface of the receptor. A transparent disk is shown at the point of contact, along with a perpendicular pole that shows where the player's up vector will point after the move. Upon pressing the designated button on their controller, the player is quickly warped to the new location. In addition, the scale of the world is increased dramatically, inspiring the impression that they are standing on a surface of titanic proportions. To reduce simulator sickness, the screen is blurred slightly, and their position is quickly linearly interpolated between their origin and destination. Any additional change in rotational orientation is performed instantly to avoid nausea.

Scoring of Player's ligand positions
The CANDIY uses a generalized statistical potential function derived from the Cambridge Structural Database (CSD) 39 . This scoring function is applicable to a variety of chemical environments including small molecules, proteins, RNA complexes, metal ions, cofactors, water molecules, etc. The score of a given molecular pose is calculated as the sum of all pair-wise interactions occurring between the small molecule and the biomolecule of interest within a cutoff of 15 angstroms. The interaction between two atoms with distance r is defined by the ratio of a functional term by a reference term, given in Eq (1).
Here, $% &' is the distance between atom i of type a and atom j of type b. The numerator (functional term) can be defined with Eq (2), which is derived from the radial distribution function: Here, 3 is the number of times an atom of type b is found within a given distance from an atom of type a and 3 is the volume of a sphere with radius r. The denominator (reference term) of Eq. (1) is defined as follows:

Automatic optimization of user-created molecules
One of MINT's important features is the relevance of chemistry presented to players during the VR experience. Since we cannot expect all players to be experts in chemistry, we have included an energy minimization functionality to help players correct potential mistakes such as bad bond lengths, stretches, angles, and contacts. This functionality is optional but highly recommended for ensuring proper chemistry is incorporated into the game. CANDIY provides this functionality via the use of the OpenMM toolkit 40 . Bonded forces are calculated using the Amber Forcefield 41 and non-bonded forces are calculated using the aforementioned scoring function.

Compatibility on mobile devices
Molecular visualization and manipulation can be costly computational wise, especially for proteins, as it oftentimes deals with a large quantity of atomic data. We have found that for mobile platform surface model generation and rendering of a protein takes approximately 5 to 6 times more duration to complete than for the desktop version. Plus, virtual reality display, compared to conventional 2D display, is innately more expensive due to its requirement from hardware to render an image twice, one for the left eye and the other for the right eye. To combat the above limitation and to improve the overall user experience, we have been looking into optimizing the program via multi-threading processing and GPU processing. Also, as MINT is built using Unity3D game engine, Unity's recent engine update in 2018, which improves its support for developers to build a high-performance application, will contribute to the optimization of MINT as well.

Code Availability
MINT is hosted publicly at https://chopralab.github.io/MINT/index.html where it can be downloaded and installed on Windows PCs and Android phones. WZ, CS, and JM were responsible for the implementation of the program on the Unity platform and made figures in the paper. JF helped with backend implementation by interfacing CANDIY C++ code with the Unity platform for structural and chemical accuracy of biomolecules and helped write the manuscript with WZ. GC conceived the idea, helped with the implementation choices and overall design of the program, managed the entire project, and wrote the paper.     , snapshot on the left of (A) shows the state of molecule structure before Hand tool manipulation is operated, and snapshot on the right of (A) shows the state after Hand tool manipulation is operated. The hand tool is used for moving molecular cluster in VR environment.