![]() Interestingly, removing this information allows the regressor to perform far better when it comes to the shape parameters of the point cloud ('shape' 1–14). As expected, CFOR normalization removes all information on cloud size ('scaling') and orientation ('rotation' 1–2). ( A) Performance in predicting different generative parameters of point clouds in a synthetic dataset from either a raw or a size- and rotation-corrected (cell frame of reference, CFOR) embedding. This feature space can be transformed with PCA to emphasize relevant variation across the sample population. Finally, features are computed to describe each cell's point cloud relative to these common reference points, resulting in an embedded feature space. A representative subset of the resulting clouds is overlaid and k-means clustering is performed on the overlay, yielding a set of common reference points. Input point clouds of cells are either rotated according to a registration across tissues (Tissue Frame Of Reference, TFOR) or are volume-normalized and re-represented as a subset of the pairwise distances between points, removing size and rotational information (Cell Frame Of Reference, CFOR). The resulting image is normalized and used to stochastically sample points for the point cloud. To sample from cell shapes, the 1vxl-wide outer shell of the segmentation is set to 1, all other voxels to zero. To sample from intensity distributions, images are masked by setting voxels outside of the segmentation to zero and a simple background subtraction is performed. Note that the most distinguishing morphological feature of the two example cells, namely the outcropping of cell a at the bottom, is reflected in a large difference in the corresponding cluster's distance values (cluster 4, blue). CBE proceeds by performing clustering on both clouds combined (middle) and then extracting the distances along each axis from each cluster center to the centroid of its ten nearest neighbors (right). In this 2D mock example, two cells are being embedded based on point clouds of their outlines (left). ( D) Illustration of CBE, our algorithm for embedding point clouds into a feature space. Shown are a slice of an input image (left), here a membrane-labeled cell in the pLLP (scale bar: 2 μm), the landmarks sampled from this image (middle), here oversampled compared to the standard pipeline for illustration purposes, and the resulting 3D point cloud (right). ( C) Illustration of ISLA, our algorithm for conversion of voxel-based 3D images to representative point clouds. See Figure 3-figure supplement 1 for a more detailed version. ( B) Adapted workflow for morphometrics of arbitrary fluorescence intensity distributions. ( A) A classical workflow in landmark-based geometric morphometrics. ![]() ( F) Finally, all of the resulting data are explored and interpreted through various visualizations and statistics. ( E) A similar strategy can be used to map manually annotated contextual knowledge (top) into the dataset (bottom), in this case specific cell archetypes chosen based on prior knowledge of the tissue's biology. ( D) Such well-structured data simplify the application of machine learning techniques for data integration, which here is performed based on cell shape as a common reference measurement. ( C) Next, data extraction takes place to arrive at numerical features representing the cell shapes (yellow) and the various fluorescent protein distributions of additional markers (other colors). ( B) Using an automated image analysis pipeline, single cells are automatically segmented based on the membrane marker to prepare them for analysis, illustrated here by shifting them apart. Each sample is labeled with a membrane marker to delineate cell boundaries (top) and samples can additionally be labeled with various other markers of interest (bottom, colored). ![]() ( A) Image data of the tissue of interest are acquired using 3D confocal fluorescence microscopy. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |