REL 2021-09-20 2021-10-01 2022-02-28 0000-0002-0462-6540 Sjors H.W. Scheres MRC Laboratory of Molecular Biology Francis Crick Avenue Cambridge Cambridgeshire United Kingdom CB2 0QH 0000-0002-0462-6540 Sjors H.W. Scheres MRC Laboratory of Molecular Biology Francis Crick Avenue Cambridge Cambridgeshire United Kingdom CB2 0QH Kimanius D Dong L Sharov G Nakane T Scheres SHW 20.7 10.6019/EMPIAR-10812 Exception Kimanius D Dong L Sharov G Nakane T Scheres SHW The Biochemical journal Biochem J 24 478 4169 4185 2021 English 10.1042/bcj20210708 34783343 Kimanius D Dong L Sharov G Nakane T Scheres SHW bioRxiv 10.1101/2021.09.30.462538 2D class averages for training neural network in region_class_ranker /data class averages MRCS MRCS 18051 1 32 BIT FLOAT variable variable variable variable

Each subdirectory with 12 random characters contains a single 2D classification run, with an image file run_class.mrcs that contains the actual 2D class averages, the files run_model.star, run_data.star, run_sampling.star and run_optimiser.star with the corresponding metadata from RELION's 2D classification run (see RELION documentation for details), a file job_score.txt that contains the manually assigned job score for that 2D classification run, a backup_selection.star file that contains the different categories of assigned classes, which are converted to individual class scores in the class_ranker program (see function ClassRanker::getClassScoreFromJobScore inside src/class_ranker.cpp), and a file features_normalized.star that contains the features calculated by the region_class_ranker program. One can visualise the images for each class, e.g. in directory cahg4Zo4Goos, with the following command: relion_display --sort rlnClassDistribution --reverse --class --i cahg4Zo4Goos/run_optimiser.star --fn_imgs cahg4Zo4Goos/backup_selection.star Classes shown in red (1 in backup_selection.star) are the best according to the manually assigned class labels in backup_selection.star; magenta (5) are second-best; green (2) third-best; and blue (3) or cyan (4) fourth-best. Yellow classes (6) or non-coloured classes (0) are the worst (score=0). The normalised_features.star file was calculated in RELION-4.0, running the following command in csh from the main directory: foreach opt (*/run_optimiser.star) set dir=`echo ${opt} | awk -F"/" '{print $1}'` echo $dir relion_class_ranker --train --do_granularity_features --extract_subimages --subimage_boxsize 64 --nr_subimages 25 --opt ${dir}/run_optimiser.star --select ${dir}/backup_selection.star --fn_score ${dir}/job_score.txt --o ${dir} --write_normalized_features end