10.1371/journal.pone.0211659.g002 Ismail M. Khater Ismail M. Khater Stephane T. Aroca-Ouellette Stephane T. Aroca-Ouellette Fanrui Meng Fanrui Meng Ivan Robert Nabi Ivan Robert Nabi Ghassan Hamarneh Ghassan Hamarneh The process of obtaining the class labels for the Cav1 blobs using wide-field CAVIN1/PTRF mask. The class labels are necessary to train the machine learning models to identify the Cav1 blobs types automatically. Public Library of Science 2019 classification Caveolae transcript release factor molecule localization microscopy 3 D cluster PC 3-PTRF cells adaptor protein polymerase PC 3 prostate cancer cells scaffold approach method caveolae point clouds 10 PC 3 accuracy Cav 1 proteins plasma membrane invaginations PointNet model SMLM data CAVIN CNN super-resolution microscopy images molecule localization microscopy data 2019-08-26 17:25:07 Figure https://plos.figshare.com/articles/figure/The_process_of_obtaining_the_class_labels_for_the_Cav1_blobs_using_wide-field_CAVIN1_PTRF_mask_The_class_labels_are_necessary_to_train_the_machine_learning_models_to_identify_the_Cav1_blobs_types_automatically_/9731924 <p>(A) The first row shows the imaged wide-field TIRF CAVIN1/PTRF mask before and after morphological closing. The morphological closing operation is used to close the small holes in the consecutive regions of CAVIN1/PTRF mask. The CAVIN1/PTRF regions are delineated in yellow to highlight the locations of the CAVIN1/PTRF regions in the cell. (B) The second row shows the Cav1 blobs and the overlay of the Cav1 blobs with the wide-field CAVIN1/PTRF mask to label the blobs into PTRF+ and PTRF-. The caveolae structures have a minimum of 60 Cav1 molecule per blob [<a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0211659#pone.0211659.ref009" target="_blank">9</a>] that can be used to stratify the PTRF+ blobs into PTRF+≥ 60 and PTRF+< 60. Our goal is to use machine learning approaches to automatically identify the PTRF+≥ 60 blobs (caveolar domains) from the rest of the non-caveolar domains (i.e. PTRF+< 60 and PTRF-) using different features and data representations of the blobs.</p>