high performance computing on graphics processing units: hgpu.org

The robust alignment of tilt-series collected for cryo-electron tomography in the absence of fiducial markers, is a problem that, especially for tilt-series of vitreous sections, still represents a significant challenge. Here we present a complete software package that implements a cross-correlation based procedure that tracks similar image features that are present in several micrographs and explores them implicitly as substitutes for fiducials like gold beads and quantum dots. The added value compared to previous approaches, is that the algorithm explores a huge number of random positions, which are tracked on several micrographs, while being able to identify trace failures, using a cross-validation procedure based on the 3D marker model of the tilt-series. Furthermore, this method allows the reliable identification of areas which behave as a rigid body during the tilt-series and hence addresses specific difficulties for the alignment of vitreous sections, by correcting practical caveats. The resulting alignments can attain sub-pixel precision at the local level and is able to yield a substantial number of usable tilt-series (around 60%). In principle, the algorithm has the potential to run in a fully automated fashion, and could be used to align any tilt-series directly from the microscope. Finally, we have significantly improved the user interface and implemented the source code on the graphics processing unit (GPU) to accelerate the computations.