Abstract
Cryo-electron tomography enables the visualization of macromolecular complexes within native cellular environments but is limited by incomplete angular sampling and the maximal electron dose that biological specimens can be exposed to. Here, we developed cryoTIGER (Tilt Interpolation Generator for Enhanced Reconstruction), a computational workflow leveraging deep learning-based frame interpolation to generate intermediate tilt images. By interpolating between tilt series projections, cryoTIGER improves angular sampling, leading to enhanced 3D reconstructions, more refined particle localization, and improved segmentation of cellular structures. We evaluated our interpolation workflow on diverse datasets and compared its performance against non-interpolated data. Our results demonstrate that deep learning-based interpolation improves image quality and structural recovery. The presented cryoTIGER framework offers a computational alternative to denser sampling during tilt series acquisition, paving the way for enhanced cryo-ET workflows and advancing structural biology research.