CryoET Data Annotation Machine Learning Competition

Cryogenic Electron Tomography (CryoET) is an advanced imaging technique for visualizing cellular structures at near-atomic resolution. CryoET has led to significant scientific advances in our understanding of protein structure, cellular function, and disease. These insights come from analyzing annotated CryoET data where different parts of the cell and even individual proteins are segmented.

However, annotating CryoET data by hand is very time-consuming, taking weeks or months for a single dataset. In addition, particle picking, the annotation process of identifying particles of interest within tomograms, is challenging, especially for particles smaller than ribosomes since CryoET data has a low signal-to-noise ratio compared to other imaging modalities. Particle picking is generally done by template matching, which compares a predefined template to potential regions containing the particles of interest in the tomograms. Once particles are picked, they are aligned and averaged to reconstruct structures at near-atomic resolutions, providing insights into cellular mechanisms.

Our machine learning competition focuses on enhancing particle picking for a range of particle sizes through the development of machine learning models to generate semantic segmentation masks within 3D CryoET tomograms. The primary aim is to refine the template matching process by identifying areas of interest where template matching should be concentrated. This method narrows down the search space, thereby boosting the accuracy and efficiency of the CryoET computational workflow. Participants are tasked with crafting models that produce masks highlighting the precise locations for template matching, facilitating the identification of specific particles across various tomograms.