Shape classification is one of the most important tasks in computer vision as demonstrated by the large body of work dealing with 3D shape analysis. Recent advances in 3D data acquisition as well as the availability of large 3D repositories have been instrumental in the design of new and more efficient algorithms for shape classification. 3D shapes may be represented by graphs and consequently, graph techniques may be strongly useful for their classification. In the GRADIENT project, we address the problem of 3D protein deformable shapes classification. Proteins are macromolecules characterized by deformable and complex shapes which are linked to their function making their classification an important task namely for drug discovery and disease characterization. Protein shapes can be standardly and robustly generated from their high-resolution 3D structures available in the Protein Data Bank. Their conformational space can be sampled using molecular dynamics simulations. In the GRADIENT project, proteins are assimilated to 3D dynamic deformable objects and their surfaces are represented by graphs, such as triangular tessellations or meshes. Since molecular dynamics can be used to efficiently sample the trajectories of molecular 3D objects, they constitute a perfect case of study for dynamic graph matching.

The goal of the GRADIENT project is to develop a platform containing robust techniques and algorithms based on graphs, algorithms and machine learning able to classify and analyze 3D deformable (dynamic) shapes of proteins. To do that, we will address the following research hypothesis and challenges:

1. Data representation, descriptors and features for 3D protein deformable structures.
2. Graph distances to compare 3D protein deformable shapes.
3. Robust models algorithms combining graph and machine learning techniques to classify 3D protein deformable shapes.
4. Trajectories tracking analysis of protein dynamics.

This platform will be open for biologists and chemists to study their proteins.