Efficient GPU computation of large protein Solvent-Excluded Surface

Cyprien Plateau--Holleville1, Maxime Maria1, Stéphane Mérillou1, Matthieu Montes2, 3
1XLIM, UMR CNRS 7252, Université de Limoges 2Laboratoire GBCM, EA 7528, Conservatoire National des Arts et Métiers, Hésam Université 3Institut Universitaire de France (IUF)
Publisher Paper (36.6Mo) Paper compressed (1Mo) Code
The solvent excluded surface of the complete HIV capsid.

The complete geometry of the Solvent-Excluded Surface of the complete HIV capsid (PDB ID: 3J3Q) computed in 700ms on an NVIDIA RTX 2080 8GB with our method.

Abstract

The Solvent-Excluded Surface (SES) is an essential representation of molecules which is massively used in molecular modeling and drug discovery since it represents the interacting surface between molecules. Based on its properties, it supports the visualization of both large scale shapes and details of molecules. While several methods targeted its computation, the ability to process large molecular structures to address the introduction of big complex analysis while leveraging the massively parallel architecture of GPUs has remained a challenge. This is mostly caused by the need for consequent memory allocation or by the complexity of the parallelization of its processing. In this paper, we leverage the last theoretical advances made for the depiction of the SES to provide fast analytical computation with low impact on memory. We show that our method is able to compute the complete surface while handling large molecular complexes with competitive computation time costs compared to previous works.

Performance

We compare our method to Megamol's implementation of GPU Contour-Buildup, the fastest method in our experiments. On the same GPU, our method is able to compute the surface of 30x larger protein without being restricted to the exterior surface.

Computation benchmark

Regarding computation time, our methods perform similarly to GPU Contour-Buildup, which allows the fast computation of the complete surface of large proteins.

Computation benchmark

Illustration

Thanks to reduced computation time and complete geometry computation, we can compute illustrations of molecular bodies with various atom counts and shading effects emphasizing the properties of the molecular surface.

PDB ID: 6BT3

PDB ID: 2D1S

PDB ID: 1CD3

Interactive visualisation

Finally, our method can be used for interactive visualisation of proteins with per-frame updates.

Per-frame update with randomized geometry update. PDB ID: 1AON

BibTeX

@article{PlateauHolleville2024,
    author={Plateau—Holleville, Cyprien and Maria, Maxime and Mérillou, Stéphane and Montes, Matthieu},
    journal={IEEE Transactions on Visualization and Computer Graphics}, 
    title={Efficient GPU computation of large protein Solvent-Excluded Surface}, 
    year={2024},
    volume={},
    number={},
    pages={1-12},
    doi={10.1109/TVCG.2024.3380100}
}