Xmipp is a software package for cryo-EM data reconstruction developed by CNB CSIC. GPU-accelerated versions of the hotspots in Xmipp are developed as a joint effort of ICS MUNI and CNB CSIC.

The novel protocol allows Xmipp to analyze the continuous heterogeneity of a specimen (typically a macromolecule), i.e., identify different conformers and transition states of the molecule. In its current implementation, the multi-core CPUs are utilized. The processing speed of the current implementation limits both practical usage of the protocol and its further improvements.

The master thesis will focus on GPU acceleration of the protocol for continuous heterogeneity analysis. The bottleneck of the protocol is a function computing displacement and rotation of a voxel field to match other voxel fields as close as possible (the voxel fields represent a 3D structure of the molecule). The student will implement and optimize the CUDA kernel, which replaces the function's CPU implementation and integrates this kernel into Xmipp. The master thesis will evaluate the original CPU-based implementation, the new GPU-based implementation, and proposed optimizations of the GPU-based implementation.