#	Pagina
attuale pagina	/open-h2020/projects/197237/results.html
-1	/open-h2020/projects/218787/index.html
-2	/open-h2020/projects/214559/index.html

Report

Teaser, summary, work performed and final results

Periodic Reporting for period 3 - VIDOCK (2D Conformal mapping of protein surfaces: applications to VIsualization and DOCKing software)

Teaser

Summary

Protein-protein interactions play a crucial role in biological processes such as cellular communication, immune defense, cell-growth or programmed cell death. These interactions occur between the respective surfaces of the proteins involved. The characterization of their shape is thus of major importance.

Different methods have been developed to compare protein subparts but very few methods taking into account the whole protein surfaces are available. In the ViDOCK project, we are developing tools for 2-dimensional representations of protein surfaces that will be applied for:
1. Fast shape similarity search
2. Real-time visualization and interactive simulation tools

Work performed

Regarding the 1st objective: We have developed tools for generating 2D protein representations (maps). The surface maps are generated using a two-step procedure: A. The protein surface is projected onto a sphere. B. The sphere is then projected onto a plane which resuls into a map of the surface of the protein. These maps are then used to compare the shape of the proteins.

We evaluated and published the performance of our prototype against a reference benchmark of the computer vision scientific community (TOSCA) and against another community benchmark of protein shapes in a reference conference of the computer vision scientific community (SHREC2017).

The prototype of the shape similarity search method compares two maps within a microsecond but lacks sufficient precision to be usable as is. The ViDOCK team hence improved the sensitivity of method by adding an optimization procedure.

We have constituted a new benchmarking dataset to evaluate the performance of similarity search methods of protein surfaces that will be used as a community benchmark for the SHREC2018 event.

For the 2nd objective, we started the development of a usable open-source simulation system based on the Udock engine (Levieux et al, Faraday Discuss 2014; http://udock.fr). We used the surface maps developed in objective 1 as a complementary representation to the usual 3D representation of structural biology. We are developing new ways to represent and manipulate multiple 3D objects in real time using methods from the video game industry. This will allow biologists and chemists not familiar with molecular visualization software to use them in their own professional environment.

Different features will be available in our system:
The prototype allows interactive real-time assembly of multiple proteins
The prototype includes a virtual camera which allows a local exploration and visualization of the properties of the interfaces of protein assemblies.

Final results

We have successfully applied computational geometry tools used in robotics and computer vision to generate 2 Dimensional representation of protein surfaces. The maps are then compared in order to compare the shapes. With this approach, we have unconventionally applied a method initially developed and used for generic shape retrieval (Furniture, faces) in computational structural biology. This fast 2D-based shape comparison methods will allow the ViDOCK team to compare the shapes of all proteins in the Protein Databank. This understanding of the shape similarity between protein structures could provide a better insight on protein partners involved in cellular mechanisms and drugs off-targets responsible for adverse effects.

We have developed a new way to interact and visalize objects in a molecular visualization software by using a local user-controlled virtual camera. This representation/manipulation is classical in 3rd person manipulation of a character in video games but has not been used so far in molecular visualization software. Using a local virtual camera as a complement to the classical global camera gives an insight notably of the parts of the system that are difficult to visualize such as the interface between opaque objects (like protein surfaces) or very crowded environments.
With this virtual camera, we unconventionally applied interaction methods from the video game industry which is one of the industrial domains where the quality and the usability of the computer-human interaction is the most critical. All video games with approximate controls or uninformative camera angles ended as industrial failures.

These new representation and interaction methods will be included in a visualization and simulation tool that we will provide to the community and that will be more easily accessible and usable for non experts in computational structural biology.

In this project, we have successfully and unconventionally applied methods and algorithms routinely used in robotics, computer vision and video game industries.