Periodic Reporting for period 1 - VTX-HPC (Molecular Visualization of High Performance Computing Simulations with VTX)
Período documentado: 2022-04-01 hasta 2023-12-31
Molecular visualization is a critical task usually performed by structural biologists and bioinformaticians to aid three processes that are essential in science and fundamental to understand structural molecular biology: synthesis, analysis and communication
• Synthesis involves the process of creating a model to integrate experimental observation or to develop a structural hypothesis.
• Analysis involves the examination and exploration of data and/or models to explain observed phenomena, derive new hypotheses, or suggest new experiments.
• Communication involves the imparting or sharing of information and knowledge about the systems or methods under study.
Molecular dynamics simulation (MD) is widely used in structural molecular biology and drug discovery to study the structure, properties and dynamics of small and macromolecules. The size of the simulated systems (up to a billion atoms) and the length of the simulations (from 50 ns in 2006 to more than 50 ns/day in 2020) have dramatically increased following the recent advances in computing hardware (GPU clusters) and storage. Additionally, with the recent advances in Cryo Electron Microscopy and the release of AlphaFold2, many new models of macromolecular structures and complexes are now available for molecular simulation. Due to their more and more extreme size, the storage, analysis and visualization of the resulting data is becoming problematic. As an example, a microsecond simulation contains up to 1 billion frames, each containing up to 80 gb of data for the largest systems.
Within the ViDOCK project (ERC #640283), we created a free for academics and open-source software called VTX (open beta available at http://vtx.drugdesign.fr) that is highly usable and capable of high-performance molecular visualization. VTX is optimized to handle efficiently the big data from High Performance Computing Molecular Dynamics Simulation (HPC MD). It is based on a high-performance 3D engine including cutting-edge computer graphics methods, adapted to molecular scenes, that can handle several million atoms on a standard laptop computer. It also offers a video game based minimalistic task-oriented GUI to maximize its usability and comfort of use.
The goal of the VTX-HPC project is to improve the performance, usability and sustainability of VTX for the visualization of HPC MD big data by developing:
1. a solution for streaming MD for increased performances and reduced environmental load.
2. a meshless version of the widely used Solvent Excluded Surface (SES) representation, generated on-the-fly during rendering for scalability purposes and real-time visualization.
3. Adapted manipulation and controls to facilitate the generation of illustrations and movies of static and dynamic molecular scenes.
The addition of these features will make VTX unprecedented software for molecular visualization.
• Synthesis involves the process of creating a model to integrate experimental observation or to develop a structural hypothesis.
• Analysis involves the examination and exploration of data and/or models to explain observed phenomena, derive new hypotheses, or suggest new experiments.
• Communication involves the imparting or sharing of information and knowledge about the systems or methods under study.
Molecular dynamics simulation (MD) is widely used in structural molecular biology and drug discovery to study the structure, properties and dynamics of small and macromolecules. The size of the simulated systems (up to a billion atoms) and the length of the simulations (from 50 ns in 2006 to more than 50 ns/day in 2020) have dramatically increased following the recent advances in computing hardware (GPU clusters) and storage. Additionally, with the recent advances in Cryo Electron Microscopy and the release of AlphaFold2, many new models of macromolecular structures and complexes are now available for molecular simulation. Due to their more and more extreme size, the storage, analysis and visualization of the resulting data is becoming problematic. As an example, a microsecond simulation contains up to 1 billion frames, each containing up to 80 gb of data for the largest systems.
Within the ViDOCK project (ERC #640283), we created a free for academics and open-source software called VTX (open beta available at http://vtx.drugdesign.fr) that is highly usable and capable of high-performance molecular visualization. VTX is optimized to handle efficiently the big data from High Performance Computing Molecular Dynamics Simulation (HPC MD). It is based on a high-performance 3D engine including cutting-edge computer graphics methods, adapted to molecular scenes, that can handle several million atoms on a standard laptop computer. It also offers a video game based minimalistic task-oriented GUI to maximize its usability and comfort of use.
The goal of the VTX-HPC project is to improve the performance, usability and sustainability of VTX for the visualization of HPC MD big data by developing:
1. a solution for streaming MD for increased performances and reduced environmental load.
2. a meshless version of the widely used Solvent Excluded Surface (SES) representation, generated on-the-fly during rendering for scalability purposes and real-time visualization.
3. Adapted manipulation and controls to facilitate the generation of illustrations and movies of static and dynamic molecular scenes.
The addition of these features will make VTX unprecedented software for molecular visualization.
Some of the tasks required a better modularity of the software. We have then completely refactorized VTX into separate and independent modules (Core, renderers, etc..). We also included automated unit tests and automated test builds of the code which are essential for the production and release of the future stable versions of an industrial level software.
For task 1, due to the industrial constraints of the industrial partner to additionally get a web based version of the application for a better integration in their software platform, we finally opted for a full remote technology. . This solution allowed the use of VTX as an API in the industrial partner web-based software suite.
For Task 2, the main objective was to deliver real-time and offline graphics rendering representations optimized for large molecular systems structure and trajectories. We focused on developing a meshless Solvent Excluded Surface (SES) analytical representation adapted to real time rendering and computed efficiently on GPU. As planned we successfully developed and implemented a solution allowing the fast computation of the SES on parallel architectures such as GPUs (Plateau-Holleville et al, under minor revision in Trans Vis Comput Graph).
For task 3, the objective was to design, develop and integrate in VTX an intuitive and usable top quality illustration and movie generation module. We integrated fully customizable representations and displays that include multiple shaders for visual effects such as fog, outline, materials (glossy, diffuse, flat color). We designed a snapshot/export image function that allows to produce high resolution images of the displayed molecular scene (up to 16K resolution in multiple formats) with customizable transparent background. Concerning the molecular movies module, we developed and integrated an interface that allows to place and save landmark viewpoints within the 3D scene.
We also worked on developing automatic camera pathfinding procedures in order to automatically produce the most relevant camera viewpoints for a complete overview of the molecular scene. The pathfinding procedure is based on the evaluation of the scene Shannon entropy over an icosahedron that englobe the molecular scene. The prototype was published in 2023 (Larroque et al, Eurographics 2023) and will be implemented in VTX in a future release.
For task 1, due to the industrial constraints of the industrial partner to additionally get a web based version of the application for a better integration in their software platform, we finally opted for a full remote technology. . This solution allowed the use of VTX as an API in the industrial partner web-based software suite.
For Task 2, the main objective was to deliver real-time and offline graphics rendering representations optimized for large molecular systems structure and trajectories. We focused on developing a meshless Solvent Excluded Surface (SES) analytical representation adapted to real time rendering and computed efficiently on GPU. As planned we successfully developed and implemented a solution allowing the fast computation of the SES on parallel architectures such as GPUs (Plateau-Holleville et al, under minor revision in Trans Vis Comput Graph).
For task 3, the objective was to design, develop and integrate in VTX an intuitive and usable top quality illustration and movie generation module. We integrated fully customizable representations and displays that include multiple shaders for visual effects such as fog, outline, materials (glossy, diffuse, flat color). We designed a snapshot/export image function that allows to produce high resolution images of the displayed molecular scene (up to 16K resolution in multiple formats) with customizable transparent background. Concerning the molecular movies module, we developed and integrated an interface that allows to place and save landmark viewpoints within the 3D scene.
We also worked on developing automatic camera pathfinding procedures in order to automatically produce the most relevant camera viewpoints for a complete overview of the molecular scene. The pathfinding procedure is based on the evaluation of the scene Shannon entropy over an icosahedron that englobe the molecular scene. The prototype was published in 2023 (Larroque et al, Eurographics 2023) and will be implemented in VTX in a future release.
We have demonstrated the capabilities of VTX to handle massive data using a witness system provided by collaborators from Groningen University. They built a coarse grained whole cell model of a Mycoplasma that contains 561 million coarse grained pseudoatoms (Stevens et al, Frontiers in Chemistry, 2023). Using VTX, we are able to visualize this whole cell model on a consumer laptop computer. The loading of the model takes reasonable time (8 to 11 minutes depending on the available RAM) and the fluidity and ease of manipulation is beyond expectations (from 60 to 120 frames per second on consumer laptop GPUs). The real-time renderer includes SSAO lighting effects and implicit only representations that allow pixel perfect quality molecular representations. This demonstrates the high performance of our real-time meshless molecular graphics engine which is unmatched yet in the literature to our knowledge.
We also evaluated the handling of massive molecular dynamics data with a trajectory of 500ns of a 60 million atom system provided by collaborators from Groningen University. We are able to load this massive system in an amount of time similar to the one obtained with the most optimized software on this task in the literature, VMD. Since VTX allows much increased features including molecular graphics performance, selection, navigation and manipulation, as compared to VMD, this is a great step towards next generation molecular visualization software.
We also evaluated the handling of massive molecular dynamics data with a trajectory of 500ns of a 60 million atom system provided by collaborators from Groningen University. We are able to load this massive system in an amount of time similar to the one obtained with the most optimized software on this task in the literature, VMD. Since VTX allows much increased features including molecular graphics performance, selection, navigation and manipulation, as compared to VMD, this is a great step towards next generation molecular visualization software.