Electron tomography is the most widely applicable method for obtaining 3D information by electron microscopy. It has become a powerful tool for elucidating 3D architectures of biological samples at resolution of about 4-6 nm and is the only method suitable for investigating polymorphic structures such as organelles, cells and tissues. However, in addition to the relatively low resolution, electron tomograms inevitably suffer from a low signal-to-noise ratio and some data-collection artifacts. These factors significantly hamper development of algorithms for reliable detection of structural features, which poses a severe barrier to progress in the field. As of today, the tasks of extracting information from these highly complex cellular tomograms are, for the most part, painstakingly carried out manually. Apart from the subjectivity of the process, the time consuming (and tiring) nature of this manual task all but precludes the prospects of the high throughput necessary to take full advantage of the method’s potential.
Here, we present a novel tool for the detection of filaments in cellular tomograms that is based on reduced representation templates. Reduced representations consist of small sets of 3D points that capture the characteristics of the underlying structure. The use of these representations results in a reduction of computational complexity that allows scanning large volumes in real space in a relatively short time. This approach is specifically useful for detecting structures with higher order such as filaments and bundles. The use of reduced representations allows efficient adjustment of the scoring function for variations in signal-to-noise level, background, and surrounding environment (crowding), all factors that significantly hamper reliable detection using traditional correlation-based template matching. As a result, the approach is capable of matching or even exceeding the detection performance of a human operator.
Funding was provided by NIH grant P01 GM098412